diff --git a/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..8b84acb758f60a5b425c7e71a09ebeaad31189d2 --- /dev/null +++ b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt @@ -0,0 +1,108 @@ +[05:16:39][ 0s][START] target=dhanyasi-P14-mouse-cerebellum variant=macs-v11-norestart-acq-laplace budget=5.0% seed=2 | {"task": "membrane", "target": "dhanyasi-P14-mouse-cerebellum", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "laplace", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[05:16:39][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[05:16:39][ 0s][DATA] dhanyasi-P14-mouse-cerebellum: train=8649 val=1922 test=2883 patches +[05:16:41][ 3s][CALIB] source-val temperatures: c-elegans=2.58, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, oct-sfl=2.64, snemi=3.55, mouse-brain=4.08 +[05:16:42][ 4s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 1s +[05:16:48][ 9s][WEIGHTS] mixture: c-elegans=0.051, flywire=0.156, berghia=0.010, microns=0.417, oct-glia=0.040, oct-sfl=0.054, snemi=0.107, mouse-brain=0.165 | eff.sources=5.34 | MMD2 mix=0.0662 uniform=0.1039 +[05:16:48][ 9s][WEIGHTS] per-source: c-elegans=0.183, flywire=0.001, berghia=0.071, microns=0.290, oct-glia=0.195, oct-sfl=0.144, snemi=0.112, mouse-brain=0.004 | min-MMD source (NeuroADDA rule) = microns +[05:17:12][ 34s][TEACHERS] target-val Dice: c-elegans=0.413, flywire=0.107, berghia=0.158, microns=0.223, oct-glia=0.110, oct-sfl=0.291, snemi=0.297, mouse-brain=0.448 +[05:17:12][ 34s][TEACHERS] ensemble target-val Dice: mixture=0.210, mixture@raw=0.220, mixture@logit=0.189, per_source=0.193, per_source@raw=0.209, per_source@logit=0.156, uniform=0.192, uniform@raw=0.210, uniform@logit=0.154 +[05:17:12][ 34s][WEIGHTS] using mixture: c-elegans=0.051, flywire=0.156, berghia=0.010, microns=0.417, oct-glia=0.040, oct-sfl=0.054, snemi=0.107, mouse-brain=0.165 +[05:17:12][ 34s][INIT] student initialised from teacher microns (weight 0.417) +[05:17:15][ 36s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[05:17:15][ 36s][KD] distilling the mixture ensemble (prob space) into the student +[05:18:23][ 105s][KD] soft labels for 6000 patches from 8 teachers in 69s +[05:19:20][ 161s][KD] epoch 1/5 | patches=6000 | loss=0.5836 | 56s +[05:20:15][ 217s][KD] epoch 2/5 | patches=6000 | loss=0.4512 | 56s +[05:21:10][ 272s][KD] epoch 3/5 | patches=6000 | loss=0.4330 | 55s +[05:22:06][ 327s][KD] epoch 4/5 | patches=6000 | loss=0.4247 | 55s +[05:23:01][ 382s][KD] epoch 5/5 | patches=6000 | loss=0.4224 | 55s +[05:23:03][ 384s][ROUND0] before any target label: val_dice=0.1795 val_vi=0.5071 +[05:23:03][ 384s][AL] patches=8649 (pool 8647 after dropping blank; free blank negatives: True) | budget=432 (5.0%) | rounds=6 | per_round=72 | acquisition=laplace | first_round=stratified +[05:23:29][ 410s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': -0.3018138618955013, 'auroc_top20_error': 0.24228212890625} +[05:23:52][ 434s][AL] round 1: stratified cold start: 36 patches from the membrane-rich half (ensemble prevalence >= 0.184) + 36 from the membrane-poor half of 2000 candidates +[05:24:00][ 441s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.449, uniform=0.257, top1=0.413, top2=0.428, top3=0.558 -> using top3 +[05:24:02][ 443s][WEIGHTS] round 1: student kept (labeled Dice 0.228 vs best teacher mouse-brain 0.413); distilling the corrected ensemble into it +[05:24:02][ 443s][WEIGHTS] round 1: teacher Dice on 72 labeled patches + 2 blank (891480 fg px): c-elegans=0.347, flywire=0.154, berghia=0.171, microns=0.295, oct-glia=0.191, oct-sfl=0.403, snemi=0.403, mouse-brain=0.413 | weights: c-elegans=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.157, snemi=0.312, mouse-brain=0.531 +[05:24:28][ 470s][KD] soft labels for 6000 patches from 8 teachers in 26s +[05:25:23][ 525s][KD] epoch 1/2 | patches=6000 | loss=0.5636 | 55s +[05:26:18][ 579s][KD] epoch 2/2 | patches=6000 | loss=0.5549 | 55s +[05:26:27][ 589s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 9s +[05:26:30][ 592s][FT] round 1/6 epoch 1/5 | patches=74 | loss=1.4207 +[05:26:33][ 595s][FT] round 1/6 epoch 2/5 | patches=74 | loss=1.3712 +[05:26:36][ 598s][FT] round 1/6 epoch 3/5 | patches=74 | loss=1.3298 +[05:26:39][ 601s][FT] round 1/6 epoch 4/5 | patches=74 | loss=1.3434 +[05:26:43][ 604s][FT] round 1/6 epoch 5/5 | patches=74 | loss=1.3357 +[05:26:45][ 606s][AL] round 1/6 | labeled=72 | selection_seconds=49.4000 round_seconds=219.7000 restart_student_dice=0.2279 restart_teacher_dice=0.4127 val_dice=0.6071 val_vi=0.6185 val_ece=0.0518 +[05:27:11][ 632s][AL] round 2: 1218/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:27:21][ 643s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.353, uniform=0.177, top1=0.350, top2=0.374, top3=0.407 -> using top3 +[05:27:21][ 643s][WEIGHTS] round 2: teacher Dice on 144 labeled patches + 2 blank (1729464 fg px): c-elegans=0.303, flywire=0.120, berghia=0.147, microns=0.225, oct-glia=0.119, oct-sfl=0.317, snemi=0.337, mouse-brain=0.350 | weights: c-elegans=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.130, snemi=0.317, mouse-brain=0.553 +[05:27:30][ 652s][CONS] round 2: on the 72 new patches student Dice=0.510 vs ensemble 0.313 -> pseudo-labels from the student | consistency x0.03 +[05:27:35][ 656s][FT] round 2/6 epoch 1/5 | patches=146 | loss=1.1291 +[05:27:39][ 660s][FT] round 2/6 epoch 2/5 | patches=146 | loss=1.0857 +[05:27:43][ 665s][FT] round 2/6 epoch 3/5 | patches=146 | loss=1.0509 +[05:27:48][ 669s][FT] round 2/6 epoch 4/5 | patches=146 | loss=1.0033 +[05:27:52][ 673s][FT] round 2/6 epoch 5/5 | patches=146 | loss=0.9764 +[05:27:54][ 676s][AL] round 2/6 | labeled=144 | selection_seconds=26.0000 round_seconds=67.2000 pseudo_student_dice=0.5102 pseudo_ensemble_dice=0.3126 consistency_scale=0.0340 val_dice=0.6110 val_vi=0.6381 val_ece=0.0065 +[05:28:20][ 701s][AL] round 3: 1237/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:29:14][ 755s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.424, flywire=0.151, berghia=0.175, microns=0.280, oct-glia=0.177, oct-sfl=0.437, snemi=0.449, mouse-brain=0.492 | blended with labeled Dice at 0.50 +[05:30:04][ 805s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.503, uniform=0.227, top1=0.450, top2=0.464, top3=0.477 -> using mixture +[05:30:04][ 805s][WEIGHTS] round 3: teacher Dice on 216 labeled patches + 2 blank (2913575 fg px): c-elegans=0.382, flywire=0.141, berghia=0.168, microns=0.254, oct-glia=0.139, oct-sfl=0.397, snemi=0.401, mouse-brain=0.450 | weights: c-elegans=0.072, flywire=0.020, berghia=0.002, microns=0.165, oct-glia=0.005, oct-sfl=0.089, snemi=0.185, mouse-brain=0.462 +[05:30:14][ 815s][CONS] round 3: on the 72 new patches student Dice=0.592 vs ensemble 0.554 -> pseudo-labels from the student | consistency x0.31 +[05:30:19][ 821s][FT] round 3/6 epoch 1/5 | patches=218 | loss=1.0665 +[05:30:25][ 826s][FT] round 3/6 epoch 2/5 | patches=218 | loss=1.0386 +[05:30:30][ 831s][FT] round 3/6 epoch 3/5 | patches=218 | loss=1.0204 +[05:30:35][ 837s][FT] round 3/6 epoch 4/5 | patches=218 | loss=0.9218 +[05:30:41][ 843s][FT] round 3/6 epoch 5/5 | patches=218 | loss=0.9563 +[05:30:44][ 845s][AL] round 3/6 | labeled=216 | selection_seconds=25.9000 round_seconds=167.2000 pseudo_student_dice=0.5916 pseudo_ensemble_dice=0.5539 consistency_scale=0.3050 val_dice=0.5930 val_vi=0.5589 val_ece=0.0208 +[05:31:09][ 871s][AL] round 4: 1527/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:32:05][ 926s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.348, flywire=0.192, berghia=0.172, microns=0.361, oct-glia=0.240, oct-sfl=0.485, snemi=0.415, mouse-brain=0.414 | blended with labeled Dice at 0.50 +[05:32:56][ 978s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.487, uniform=0.262, top1=0.432, top2=0.445, top3=0.491 -> using mixture +[05:32:56][ 978s][WEIGHTS] round 4: teacher Dice on 288 labeled patches + 2 blank (4568760 fg px): c-elegans=0.366, flywire=0.154, berghia=0.168, microns=0.291, oct-glia=0.158, oct-sfl=0.405, snemi=0.397, mouse-brain=0.432 | weights: c-elegans=0.062, flywire=0.023, berghia=0.002, microns=0.241, oct-glia=0.006, oct-sfl=0.097, snemi=0.178, mouse-brain=0.391 +[05:33:06][ 988s][CONS] round 4: on the 72 new patches student Dice=0.600 vs ensemble 0.428 -> pseudo-labels from the student | consistency x0.33 +[05:33:13][ 994s][FT] round 4/6 epoch 1/5 | patches=290 | loss=0.9763 +[05:33:19][ 1001s][FT] round 4/6 epoch 2/5 | patches=290 | loss=0.9413 +[05:33:26][ 1007s][FT] round 4/6 epoch 3/5 | patches=290 | loss=0.9665 +[05:33:33][ 1014s][FT] round 4/6 epoch 4/5 | patches=290 | loss=0.9434 +[05:33:39][ 1021s][FT] round 4/6 epoch 5/5 | patches=290 | loss=0.8966 +[05:33:41][ 1023s][AL] round 4/6 | labeled=288 | selection_seconds=26.0000 round_seconds=175.6000 pseudo_student_dice=0.5996 pseudo_ensemble_dice=0.4281 consistency_scale=0.3320 val_dice=0.6412 val_vi=0.5584 val_ece=0.0317 +[05:34:07][ 1049s][AL] round 5: 1370/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:35:04][ 1105s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.399, flywire=0.168, berghia=0.184, microns=0.321, oct-glia=0.217, oct-sfl=0.479, snemi=0.434, mouse-brain=0.464 | blended with labeled Dice at 0.50 +[05:35:57][ 1159s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.505, uniform=0.229, top1=0.466, top2=0.482, top3=0.485 -> using mixture +[05:35:57][ 1159s][WEIGHTS] round 5: teacher Dice on 360 labeled patches + 2 blank (5969708 fg px): c-elegans=0.395, flywire=0.136, berghia=0.176, microns=0.263, oct-glia=0.142, oct-sfl=0.388, snemi=0.414, mouse-brain=0.466 | weights: c-elegans=0.073, flywire=0.017, berghia=0.002, microns=0.160, oct-glia=0.005, oct-sfl=0.072, snemi=0.187, mouse-brain=0.484 +[05:36:07][ 1168s][CONS] round 5: on the 72 new patches student Dice=0.643 vs ensemble 0.444 -> pseudo-labels from the student | consistency x0.48 +[05:36:15][ 1176s][FT] round 5/6 epoch 1/5 | patches=362 | loss=1.0087 +[05:36:23][ 1184s][FT] round 5/6 epoch 2/5 | patches=362 | loss=0.9617 +[05:36:31][ 1192s][FT] round 5/6 epoch 3/5 | patches=362 | loss=0.9284 +[05:36:39][ 1200s][FT] round 5/6 epoch 4/5 | patches=362 | loss=0.9311 +[05:36:46][ 1208s][FT] round 5/6 epoch 5/5 | patches=362 | loss=0.9370 +[05:36:49][ 1210s][AL] round 5/6 | labeled=360 | selection_seconds=26.0000 round_seconds=184.9000 pseudo_student_dice=0.6431 pseudo_ensemble_dice=0.4439 consistency_scale=0.4770 val_dice=0.6439 val_vi=0.5519 val_ece=0.0053 +[05:37:14][ 1236s][AL] round 6: 1343/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:38:13][ 1294s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.395, flywire=0.160, berghia=0.185, microns=0.301, oct-glia=0.199, oct-sfl=0.445, snemi=0.422, mouse-brain=0.477 | blended with labeled Dice at 0.50 +[05:39:07][ 1349s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.491, uniform=0.218, top1=0.478, top2=0.491, top3=0.493 -> using mixture +[05:39:07][ 1349s][WEIGHTS] round 6: teacher Dice on 432 labeled patches + 2 blank (7439104 fg px): c-elegans=0.402, flywire=0.134, berghia=0.183, microns=0.251, oct-glia=0.130, oct-sfl=0.375, snemi=0.409, mouse-brain=0.478 | weights: c-elegans=0.076, flywire=0.016, berghia=0.002, microns=0.138, oct-glia=0.004, oct-sfl=0.062, snemi=0.173, mouse-brain=0.530 +[05:39:18][ 1359s][CONS] round 6: on the 72 new patches student Dice=0.681 vs ensemble 0.499 -> pseudo-labels from the student | consistency x0.60 +[05:39:27][ 1368s][FT] round 6/6 epoch 1/5 | patches=434 | loss=0.9291 +[05:39:36][ 1377s][FT] round 6/6 epoch 2/5 | patches=434 | loss=0.9589 +[05:39:45][ 1386s][FT] round 6/6 epoch 3/5 | patches=434 | loss=0.9226 +[05:39:54][ 1395s][FT] round 6/6 epoch 4/5 | patches=434 | loss=0.8989 +[05:40:03][ 1404s][FT] round 6/6 epoch 5/5 | patches=434 | loss=0.8698 +[05:40:05][ 1406s][AL] round 6/6 | labeled=432 | selection_seconds=26.0000 round_seconds=194.0000 pseudo_student_dice=0.6814 pseudo_ensemble_dice=0.4993 consistency_scale=0.6050 val_dice=0.6561 val_vi=0.5270 val_ece=0.0070 +[05:40:08][ 1409s][CONS] final consolidation: pseudo-labels refreshed from the student +[05:40:17][ 1418s][FT] final 1/10 epoch 1/1 | patches=434 | loss=0.8707 +[05:40:26][ 1427s][FT] final 2/10 epoch 1/1 | patches=434 | loss=0.8387 +[05:40:35][ 1437s][FT] final 3/10 epoch 1/1 | patches=434 | loss=0.8128 +[05:40:44][ 1446s][FT] final 4/10 epoch 1/1 | patches=434 | loss=0.7855 +[05:40:54][ 1455s][FT] final 5/10 epoch 1/1 | patches=434 | loss=0.7703 +[05:41:03][ 1464s][FT] final 6/10 epoch 1/1 | patches=434 | loss=0.7591 +[05:41:12][ 1473s][FT] final 7/10 epoch 1/1 | patches=434 | loss=0.7416 +[05:41:21][ 1482s][FT] final 8/10 epoch 1/1 | patches=434 | loss=0.7372 +[05:41:30][ 1491s][FT] final 9/10 epoch 1/1 | patches=434 | loss=0.7155 +[05:41:39][ 1501s][FT] final 10/10 epoch 1/1 | patches=434 | loss=0.7190 +[05:41:41][ 1503s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=94.4000 val_dice=0.6658 val_vi=0.5143 val_ece=0.0212 +[05:41:41][ 1503s][WEIGHTS] final: c-elegans=0.076, flywire=0.016, berghia=0.002, microns=0.138, oct-glia=0.004, oct-sfl=0.062, snemi=0.173, mouse-brain=0.530 +[05:42:21][ 1542s][TEST] raw student (no EMA, no TTA): patch_dice=0.6969 patch_vi=0.4345 +[05:42:26][ 1548s][TEST] final model without TTA: patch_dice=0.6969 patch_vi=0.4345 +[05:42:26][ 1548s][RESOURCES] {"phases": {"setup": {"seconds": 2.22, "peak_cpu_rss_mb": 1028.3, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 33.69, "peak_cpu_rss_mb": 1800.9, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 346.12, "peak_cpu_rss_mb": 3693.7, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 1118.32, "peak_cpu_rss_mb": 3693.7, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4642.0, "rounds": 6}, "inference": {"seconds": 44.73, "peak_cpu_rss_mb": 3693.7, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4644.0, "patches": 2883, "images": 3, "patches_per_second": 64.5, "seconds_per_image": 14.91}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 3693.7, "total_seconds": 1545.1} +[05:42:26][ 1548s][TEST] patch_dice=0.7515 patch_iou=0.6503 patch_precision=0.8460 patch_recall=0.7459 patch_vi=0.4085 image_dice=0.8127 image_iou=0.6879 image_precision=0.8866 image_recall=0.7541 image_vi=0.4586 cal_ece=0.0015 cal_nll=0.1819 cal_brier=0.0545 cal_mean_confidence=0.9268 cal_accuracy=0.9255 cal_predicted_prevalence=0.2046 cal_true_prevalence=0.2301 cal_mean_probability=0.2107 +[05:42:26][ 1548s][DONE] dhanyasi-P14-mouse-cerebellum macs-v11-norestart-acq-laplace b5 s2 | test patch Dice=0.7515 VI=0.4085 | image Dice=0.8127 VI=0.4586 | 1548s diff --git a/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..1648e59a7e1666b3ff0e91213b0c58eed8352ded --- /dev/null +++ b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json @@ -0,0 +1,785 @@ +{ + "target": "dhanyasi-P14-mouse-cerebellum", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 2, + "config": { + "task": "membrane", + "target": "dhanyasi-P14-mouse-cerebellum", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "laplace", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "dhanyasi-P14-mouse-cerebellum", + "--variants", + "macs-v11-norestart-acq-laplace", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d", + "whole-mouse-brain" + ], + "alphas": [ + 0.05083387092965836, + 0.15580245740591697, + 0.010243446832791585, + 0.41728942804507835, + 0.03964191869924105, + 0.053950781313640436, + 0.10715830600341246, + 0.16507979077026083 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d", + "whole-mouse-brain" + ], + "mixture": [ + 0.05083387092965836, + 0.15580245740591697, + 0.010243446832791585, + 0.41728942804507835, + 0.03964191869924105, + 0.053950781313640436, + 0.10715830600341246, + 0.16507979077026083 + ], + "per_source": [ + 0.1831792908635141, + 0.0008117602305311557, + 0.07130039492253423, + 0.2901333759297942, + 0.19473854254781, + 0.14427929446206111, + 0.11165168345469406, + 0.0039056575890611817 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.04944674080252824, + "mmd2_mixture": 0.06619580859842833, + "mmd2_uniform": 0.10385692805437269, + "mmd2_single_source": [ + 0.35157092449944216, + 0.1914113737092883, + 0.2296477365874025, + 0.08478080568303437, + 0.272622477181822, + 0.2463089382682188, + 0.15004801624266395, + 0.17311813097471462 + ], + "median_squared_distance": 0.18230687630230707, + "gammas": [ + 1.3713141548508683, + 2.7426283097017365, + 5.485256619403473, + 10.970513238806946, + 21.941026477613892 + ], + "max_weight": 0.41728942804507835, + "entropy": 1.6749067795900086, + "effective_sources": 5.338297487596983, + "min_weight": 0.010243446832791585 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.24983749105579034, + 0.6076273510548802, + 0.31213616057205607, + 0.2194742775498013, + 0.24579726709659655, + 0.26559858325874997, + 0.28252517774848496, + 0.5039038205278614 + ], + "standardized": [ + -0.6514591536335823, + 2.0580486812336294, + -0.17967738442258652, + -0.8813968184025964, + -0.6820553769924043, + -0.5321019342588011, + -0.4039184793457027, + 1.2725604658220449 + ], + "tau": 0.5, + "max_weight": 0.2901333759297942, + "entropy": 1.7283708993213025, + "effective_sources": 5.631472196315185, + "min_weight": 0.0008117602305311557 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "microns_basil", + "seconds": 7.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.41286983026930635, + 0.10651095633970442, + 0.15770527050491268, + 0.22267858774357308, + 0.11006956393332956, + 0.2911632103623596, + 0.2970295092802324, + 0.447529566690438 + ], + "ensemble_val_dice": { + "mixture": 0.20976331913409396, + "mixture@raw": 0.2201044706093569, + "mixture@logit": 0.1889936769494014, + "per_source": 0.19317883340253061, + "per_source@raw": 0.20883607926566583, + "per_source@logit": 0.15551849586998046, + "uniform": 0.19238025487255683, + "uniform@raw": 0.21034175472151873, + "uniform@logit": 0.15367464892063687 + }, + "val_patches": 1024 + }, + "temperatures": [ + 2.575807571411133, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 2.6421806812286377, + 3.5461363792419434, + 4.075904369354248 + ], + "top_source": "microns_basil", + "weight_stats": { + "max_weight": 0.41728942804507835, + "entropy": 1.6749067795900086, + "effective_sources": 5.338297487596983, + "min_weight": 0.010243446832791585 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.2380952380952381, + "spearman_weight_taxonomy": 0.7324670207647144, + "top_source": "microns_basil", + "top_source_is_best_teacher": false + }, + "init_teacher": "microns_basil", + "student_init": "top-source", + "blank_fraction": 0.00023124060585038732, + "kd_seconds": 346.1, + "round0_val": { + "val_dice": 0.1795085898180082, + "val_vi": 0.5071205839268854, + "val_ece": 0.1431383671563685 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 72, + "selection_seconds": 49.4, + "round_seconds": 219.7, + "teacher_dice_labeled": [ + 0.34699131258491106, + 0.15400938239211243, + 0.17063725881990668, + 0.29518215496468825, + 0.1907363491430849, + 0.40286997571879657, + 0.402560670886263, + 0.4126936934594237 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 891480, + "evidence_nonempty_patches": 58, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.22785475732277613, + "restart_teacher_dice": 0.4126936934594237, + "weights": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.1573376001938336, + 0.311542520327699, + 0.5311198794784675 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.4486622376468838, + "uniform": 0.257123442242594, + "top1": 0.4126936934594237, + "top2": 0.427720538104064, + "top3": 0.5579440810273272 + }, + "val_dice": 0.6070700831590716, + "val_vi": 0.6184691251569843, + "val_ece": 0.05175704877036804 + }, + { + "round": 2, + "n_labeled": 144, + "selection_seconds": 26.0, + "round_seconds": 67.2, + "teacher_dice_labeled": [ + 0.30315355038755776, + 0.12001301230858169, + 0.1465265228793779, + 0.2254534366318072, + 0.11921934308719918, + 0.3172824812217642, + 0.33741894167974107, + 0.3498583982698843 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1729464, + "evidence_nonempty_patches": 100, + "first_correction": false, + "restart": "", + "weights": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.13042238801348177, + 0.31683400374116205, + 0.5527436082453562 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.3530103977499102, + "uniform": 0.17731911177941523, + "top1": 0.3498583982698843, + "top2": 0.37390484882696645, + "top3": 0.4072446824388192 + }, + "pseudo_student_dice": 0.5102164768161838, + "pseudo_ensemble_dice": 0.31257957043181106, + "pseudo_source": "student", + "consistency_scale": 0.034, + "val_dice": 0.6110047127116678, + "val_vi": 0.638117576109154, + "val_ece": 0.006539609706106881 + }, + { + "round": 3, + "n_labeled": 216, + "selection_seconds": 25.9, + "round_seconds": 167.2, + "teacher_dice_labeled": [ + 0.3819753440736413, + 0.14061082822423704, + 0.16821673752804672, + 0.2537591878167585, + 0.13870534762009895, + 0.3973398975358051, + 0.40137266311189335, + 0.44969605142945246 + ], + "teacher_dice_pool": [ + 0.42386395699475177, + 0.15099438303350263, + 0.17495050231510595, + 0.279604574493834, + 0.17744932514923112, + 0.43695860860524693, + 0.4493108258230955, + 0.49185366434198335 + ], + "evidence_fg_pixels": 2913575, + "evidence_nonempty_patches": 165, + "first_correction": false, + "restart": "", + "weights": [ + 0.07229111007511609, + 0.019827741191740342, + 0.0017180507822275705, + 0.1646392588259067, + 0.0049496923673013095, + 0.08946576052936685, + 0.18501163344613708, + 0.4620967527822041 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5030819805858586, + "uniform": 0.2265744778786027, + "top1": 0.44969605142945246, + "top2": 0.4635763198897255, + "top3": 0.477477686886879 + }, + "pseudo_student_dice": 0.591602265736447, + "pseudo_ensemble_dice": 0.5539129798210711, + "pseudo_source": "student", + "consistency_scale": 0.305, + "val_dice": 0.5930079496634324, + "val_vi": 0.5588676820879914, + "val_ece": 0.020844457318665577 + }, + { + "round": 4, + "n_labeled": 288, + "selection_seconds": 26.0, + "round_seconds": 175.6, + "teacher_dice_labeled": [ + 0.3659048985997267, + 0.15353753119751318, + 0.16844964751011626, + 0.2908847684179655, + 0.15789017264557856, + 0.4047283674903942, + 0.3966932197117854, + 0.4321963382581655 + ], + "teacher_dice_pool": [ + 0.3476838243870192, + 0.19217188601799204, + 0.172233344062994, + 0.3611551305335358, + 0.2396295638748975, + 0.4852648082757493, + 0.4149907624250994, + 0.4142943549311604 + ], + "evidence_fg_pixels": 4568760, + "evidence_nonempty_patches": 233, + "first_correction": false, + "restart": "", + "weights": [ + 0.062100170113087695, + 0.022762195720855558, + 0.0017371945807277208, + 0.24075133517549557, + 0.0060491979750396845, + 0.09717298181633877, + 0.17810553945189095, + 0.391321385166564 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.4866048628809274, + "uniform": 0.2619714021531271, + "top1": 0.4321963382581655, + "top2": 0.44509639703054504, + "top3": 0.49060202517976464 + }, + "pseudo_student_dice": 0.5995607250850519, + "pseudo_ensemble_dice": 0.4281460952256102, + "pseudo_source": "student", + "consistency_scale": 0.332, + "val_dice": 0.6411766760615955, + "val_vi": 0.5584266727212388, + "val_ece": 0.03166756076429024 + }, + { + "round": 5, + "n_labeled": 360, + "selection_seconds": 26.0, + "round_seconds": 184.9, + "teacher_dice_labeled": [ + 0.3953085237250244, + 0.13647534350877416, + 0.17612099976432438, + 0.2628817625648954, + 0.1419262187673558, + 0.3879575357978311, + 0.4141298405020889, + 0.46600203134969875 + ], + "teacher_dice_pool": [ + 0.3986317767680898, + 0.16823970417904313, + 0.1841090592603666, + 0.3212210790491769, + 0.21687275845407342, + 0.4787500737459872, + 0.43429163241666896, + 0.4643731595703601 + ], + "evidence_fg_pixels": 5969708, + "evidence_nonempty_patches": 301, + "first_correction": false, + "restart": "", + "weights": [ + 0.0734540416368638, + 0.016917594038813894, + 0.0016534433165064843, + 0.16039018219487342, + 0.004545606243436821, + 0.0724328074489532, + 0.18690817398135207, + 0.48369815113920034 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5052101415491518, + "uniform": 0.22916977644195366, + "top1": 0.46600203134969875, + "top2": 0.4822501640003084, + "top3": 0.48496910511121183 + }, + "pseudo_student_dice": 0.6430858944634501, + "pseudo_ensemble_dice": 0.44390733840638275, + "pseudo_source": "student", + "consistency_scale": 0.477, + "val_dice": 0.64392883350122, + "val_vi": 0.5519165497753227, + "val_ece": 0.005270940102244961 + }, + { + "round": 6, + "n_labeled": 432, + "selection_seconds": 26.0, + "round_seconds": 194.0, + "teacher_dice_labeled": [ + 0.4015779769021923, + 0.13371681261033305, + 0.18321938102607802, + 0.250881543038054, + 0.12972555025287827, + 0.37503542861674366, + 0.40930616637479933, + 0.4782651715898207 + ], + "teacher_dice_pool": [ + 0.3952798389140022, + 0.15953434509246006, + 0.1854799560811665, + 0.30070633663258745, + 0.19935759134853634, + 0.44519469823025515, + 0.4222142629817597, + 0.4765445169330087 + ], + "evidence_fg_pixels": 7439104, + "evidence_nonempty_patches": 370, + "first_correction": false, + "restart": "", + "weights": [ + 0.0758345688067689, + 0.01595813326405757, + 0.001721237600976133, + 0.13793838128455513, + 0.0039014731144281904, + 0.06172185298498267, + 0.17270423199249021, + 0.5302201209517412 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.4911926135221738, + "uniform": 0.21810257800847072, + "top1": 0.4782651715898207, + "top2": 0.4907837766601343, + "top3": 0.49343918422359895 + }, + "pseudo_student_dice": 0.6814277622500998, + "pseudo_ensemble_dice": 0.49926128889806726, + "pseudo_source": "student", + "consistency_scale": 0.605, + "val_dice": 0.6561008947304966, + "val_vi": 0.5270340585580242, + "val_ece": 0.00698206159380761 + }, + { + "round": 7, + "n_labeled": 432, + "selection_seconds": 0.0, + "round_seconds": 94.4, + "final_consolidation": true, + "val_dice": 0.6658333978014943, + "val_vi": 0.5143445223532903, + "val_ece": 0.02122605101113116 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.3018138618955013, + "auroc_top20_error": 0.24228212890625 + }, + "al_seconds": 1118.3, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.0758345688067689, + 0.01595813326405757, + 0.001721237600976133, + 0.13793838128455513, + 0.0039014731144281904, + 0.06172185298498267, + 0.17270423199249021, + 0.5302201209517412 + ], + "weight_stats_final": { + "max_weight": 0.5302201209517412, + "entropy": 1.3790735363210445, + "effective_sources": 3.971220730881249, + "min_weight": 0.001721237600976133 + }, + "top_source_final": "whole-mouse-brain", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.6968826503970063, + "patch_iou": 0.5873942958945824, + "patch_precision": 0.7688665778942382, + "patch_recall": 0.7397190870694399, + "patch_vi": 0.434540009497679, + "n_patches": 2883, + "cal_ece": 0.013984583116174568, + "cal_nll": 0.20291160477716605, + "cal_brier": 0.05947698933410016, + "cal_mean_confidence": 0.933616504075245, + "cal_accuracy": 0.9196319209590704, + "cal_predicted_prevalence": 0.20512520865851544, + "cal_true_prevalence": 0.23013442215140478, + "cal_mean_probability": 0.2089750948953597 + }, + "test_notta": { + "patch_dice": 0.6968826503970063, + "patch_iou": 0.5873942958945824, + "patch_precision": 0.7688665778942382, + "patch_recall": 0.7397190870694399, + "patch_vi": 0.434540009497679, + "n_patches": 2883, + "cal_ece": 0.013984583116174568, + "cal_nll": 0.20291160477716605, + "cal_brier": 0.05947698933410016, + "cal_mean_confidence": 0.933616504075245, + "cal_accuracy": 0.9196319209590704, + "cal_predicted_prevalence": 0.20512520865851544, + "cal_true_prevalence": 0.23013442215140478, + "cal_mean_probability": 0.2089750948953597 + }, + "test": { + "patch_dice": 0.7515223139947016, + "patch_iou": 0.6502919854628868, + "patch_precision": 0.8459618413535717, + "patch_recall": 0.7458955232078299, + "patch_vi": 0.4085141725179584, + "n_patches": 2883, + "image_dice": 0.8126819379613308, + "image_iou": 0.6878766774473583, + "image_precision": 0.8865888050426772, + "image_recall": 0.7540789897163429, + "image_vi": 0.4586308388710216, + "n_images": 3, + "cal_ece": 0.0014887245718914603, + "cal_nll": 0.1819273293225221, + "cal_brier": 0.05446771271927853, + "cal_mean_confidence": 0.926763834764255, + "cal_accuracy": 0.9254770692421089, + "cal_predicted_prevalence": 0.2046103369970517, + "cal_true_prevalence": 0.23013442215140478, + "cal_mean_probability": 0.2107195412252572 + }, + "eval_seconds": 44.7, + "total_seconds": 1547.5, + "resources": { + "phases": { + "setup": { + "seconds": 2.22, + "peak_cpu_rss_mb": 1028.3, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 33.69, + "peak_cpu_rss_mb": 1800.9, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 346.12, + "peak_cpu_rss_mb": 3693.7, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4618.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 1118.32, + "peak_cpu_rss_mb": 3693.7, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4642.0, + "rounds": 6 + }, + "inference": { + "seconds": 44.73, + "peak_cpu_rss_mb": 3693.7, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4644.0, + "patches": 2883, + "images": 3, + "patches_per_second": 64.5, + "seconds_per_image": 14.91 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 3693.7, + "total_seconds": 1545.1 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..ae24b8c667abc9146fe3b6a1ce3257de2d820d0f --- /dev/null +++ b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,72,49.4,219.7,"[0.34699131258491106, 0.15400938239211243, 0.17063725881990668, 0.29518215496468825, 0.1907363491430849, 0.40286997571879657, 0.402560670886263, 0.4126936934594237]",,891480,58,True,kept,0.22785475732277613,0.4126936934594237,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.1573376001938336, 0.311542520327699, 0.5311198794784675]",top3,"{'mixture': 0.4486622376468838, 'uniform': 0.257123442242594, 'top1': 0.4126936934594237, 'top2': 0.427720538104064, 'top3': 0.5579440810273272}",0.6070700831590716,0.6184691251569843,0.05175704877036804,,,,, +2,144,26.0,67.2,"[0.30315355038755776, 0.12001301230858169, 0.1465265228793779, 0.2254534366318072, 0.11921934308719918, 0.3172824812217642, 0.33741894167974107, 0.3498583982698843]",,1729464,100,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.13042238801348177, 0.31683400374116205, 0.5527436082453562]",top3,"{'mixture': 0.3530103977499102, 'uniform': 0.17731911177941523, 'top1': 0.3498583982698843, 'top2': 0.37390484882696645, 'top3': 0.4072446824388192}",0.6110047127116678,0.638117576109154,0.006539609706106881,0.5102164768161838,0.31257957043181106,student,0.034, +3,216,25.9,167.2,"[0.3819753440736413, 0.14061082822423704, 0.16821673752804672, 0.2537591878167585, 0.13870534762009895, 0.3973398975358051, 0.40137266311189335, 0.44969605142945246]","[0.42386395699475177, 0.15099438303350263, 0.17495050231510595, 0.279604574493834, 0.17744932514923112, 0.43695860860524693, 0.4493108258230955, 0.49185366434198335]",2913575,165,False,,,,"[0.07229111007511609, 0.019827741191740342, 0.0017180507822275705, 0.1646392588259067, 0.0049496923673013095, 0.08946576052936685, 0.18501163344613708, 0.4620967527822041]",mixture,"{'mixture': 0.5030819805858586, 'uniform': 0.2265744778786027, 'top1': 0.44969605142945246, 'top2': 0.4635763198897255, 'top3': 0.477477686886879}",0.5930079496634324,0.5588676820879914,0.020844457318665577,0.591602265736447,0.5539129798210711,student,0.305, +4,288,26.0,175.6,"[0.3659048985997267, 0.15353753119751318, 0.16844964751011626, 0.2908847684179655, 0.15789017264557856, 0.4047283674903942, 0.3966932197117854, 0.4321963382581655]","[0.3476838243870192, 0.19217188601799204, 0.172233344062994, 0.3611551305335358, 0.2396295638748975, 0.4852648082757493, 0.4149907624250994, 0.4142943549311604]",4568760,233,False,,,,"[0.062100170113087695, 0.022762195720855558, 0.0017371945807277208, 0.24075133517549557, 0.0060491979750396845, 0.09717298181633877, 0.17810553945189095, 0.391321385166564]",mixture,"{'mixture': 0.4866048628809274, 'uniform': 0.2619714021531271, 'top1': 0.4321963382581655, 'top2': 0.44509639703054504, 'top3': 0.49060202517976464}",0.6411766760615955,0.5584266727212388,0.03166756076429024,0.5995607250850519,0.4281460952256102,student,0.332, +5,360,26.0,184.9,"[0.3953085237250244, 0.13647534350877416, 0.17612099976432438, 0.2628817625648954, 0.1419262187673558, 0.3879575357978311, 0.4141298405020889, 0.46600203134969875]","[0.3986317767680898, 0.16823970417904313, 0.1841090592603666, 0.3212210790491769, 0.21687275845407342, 0.4787500737459872, 0.43429163241666896, 0.4643731595703601]",5969708,301,False,,,,"[0.0734540416368638, 0.016917594038813894, 0.0016534433165064843, 0.16039018219487342, 0.004545606243436821, 0.0724328074489532, 0.18690817398135207, 0.48369815113920034]",mixture,"{'mixture': 0.5052101415491518, 'uniform': 0.22916977644195366, 'top1': 0.46600203134969875, 'top2': 0.4822501640003084, 'top3': 0.48496910511121183}",0.64392883350122,0.5519165497753227,0.005270940102244961,0.6430858944634501,0.44390733840638275,student,0.477, +6,432,26.0,194.0,"[0.4015779769021923, 0.13371681261033305, 0.18321938102607802, 0.250881543038054, 0.12972555025287827, 0.37503542861674366, 0.40930616637479933, 0.4782651715898207]","[0.3952798389140022, 0.15953434509246006, 0.1854799560811665, 0.30070633663258745, 0.19935759134853634, 0.44519469823025515, 0.4222142629817597, 0.4765445169330087]",7439104,370,False,,,,"[0.0758345688067689, 0.01595813326405757, 0.001721237600976133, 0.13793838128455513, 0.0039014731144281904, 0.06172185298498267, 0.17270423199249021, 0.5302201209517412]",mixture,"{'mixture': 0.4911926135221738, 'uniform': 0.21810257800847072, 'top1': 0.4782651715898207, 'top2': 0.4907837766601343, 'top3': 0.49343918422359895}",0.6561008947304966,0.5270340585580242,0.00698206159380761,0.6814277622500998,0.49926128889806726,student,0.605, +7,432,0.0,94.4,,,,,,,,,,,,0.6658333978014943,0.5143445223532903,0.02122605101113116,,,,,True diff --git a/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..eea413c5ee5fbfbab5a1ea04755b42cee5eb7609 --- /dev/null +++ b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt @@ -0,0 +1,103 @@ +[06:17:56][ 0s][START] target=dhanyasi-P14-mouse-cerebellum variant=macs-v11-norestart-acq-random budget=5.0% seed=2 | {"task": "membrane", "target": "dhanyasi-P14-mouse-cerebellum", "variant": "macs-v11-norestart-acq-random", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "random", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 0, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[06:17:56][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[06:17:56][ 0s][DATA] dhanyasi-P14-mouse-cerebellum: train=8649 val=1922 test=2883 patches +[06:17:57][ 1s][CALIB] source-val temperatures: c-elegans=2.58, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, oct-sfl=2.64, snemi=3.55, mouse-brain=4.08 +[06:17:58][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[06:18:03][ 8s][WEIGHTS] mixture: c-elegans=0.051, flywire=0.156, berghia=0.010, microns=0.417, oct-glia=0.040, oct-sfl=0.054, snemi=0.107, mouse-brain=0.165 | eff.sources=5.34 | MMD2 mix=0.0662 uniform=0.1039 +[06:18:03][ 8s][WEIGHTS] per-source: c-elegans=0.183, flywire=0.001, berghia=0.071, microns=0.290, oct-glia=0.195, oct-sfl=0.144, snemi=0.112, mouse-brain=0.004 | min-MMD source (NeuroADDA rule) = microns +[06:18:27][ 32s][TEACHERS] target-val Dice: c-elegans=0.413, flywire=0.107, berghia=0.158, microns=0.223, oct-glia=0.110, oct-sfl=0.291, snemi=0.297, mouse-brain=0.448 +[06:18:27][ 32s][TEACHERS] ensemble target-val Dice: mixture=0.210, mixture@raw=0.220, mixture@logit=0.189, per_source=0.193, per_source@raw=0.209, per_source@logit=0.156, uniform=0.192, uniform@raw=0.210, uniform@logit=0.154 +[06:18:27][ 32s][WEIGHTS] using mixture: c-elegans=0.051, flywire=0.156, berghia=0.010, microns=0.417, oct-glia=0.040, oct-sfl=0.054, snemi=0.107, mouse-brain=0.165 +[06:18:27][ 32s][INIT] student initialised from teacher microns (weight 0.417) +[06:18:30][ 34s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[06:18:30][ 34s][KD] distilling the mixture ensemble (prob space) into the student +[06:19:38][ 102s][KD] soft labels for 6000 patches from 8 teachers in 69s +[06:20:32][ 157s][KD] epoch 1/5 | patches=6000 | loss=0.5835 | 54s +[06:21:27][ 212s][KD] epoch 2/5 | patches=6000 | loss=0.4531 | 55s +[06:22:22][ 267s][KD] epoch 3/5 | patches=6000 | loss=0.4309 | 55s +[06:23:16][ 321s][KD] epoch 4/5 | patches=6000 | loss=0.4249 | 54s +[06:24:09][ 374s][KD] epoch 5/5 | patches=6000 | loss=0.4230 | 53s +[06:24:12][ 376s][ROUND0] before any target label: val_dice=0.1712 val_vi=0.5021 +[06:24:12][ 376s][AL] patches=8649 (pool 8647 after dropping blank; free blank negatives: True) | budget=432 (5.0%) | rounds=6 | per_round=72 | acquisition=random | first_round=stratified +[06:24:24][ 388s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': -0.01683200768757291, 'auroc_top20_error': 0.5015251953125} +[06:24:47][ 412s][AL] round 1: stratified cold start: 36 patches from the membrane-rich half (ensemble prevalence >= 0.184) + 36 from the membrane-poor half of 2000 candidates +[06:24:55][ 419s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.449, uniform=0.257, top1=0.413, top2=0.428, top3=0.558 -> using top3 +[06:24:57][ 421s][WEIGHTS] round 1: student kept (labeled Dice 0.216 vs best teacher mouse-brain 0.413); distilling the corrected ensemble into it +[06:24:57][ 421s][WEIGHTS] round 1: teacher Dice on 72 labeled patches + 2 blank (891480 fg px): c-elegans=0.347, flywire=0.154, berghia=0.171, microns=0.295, oct-glia=0.191, oct-sfl=0.403, snemi=0.403, mouse-brain=0.413 | weights: c-elegans=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.157, snemi=0.312, mouse-brain=0.531 +[06:25:23][ 447s][KD] soft labels for 6000 patches from 8 teachers in 26s +[06:26:18][ 502s][KD] epoch 1/2 | patches=6000 | loss=0.5647 | 55s +[06:27:14][ 558s][KD] epoch 2/2 | patches=6000 | loss=0.5552 | 56s +[06:27:23][ 567s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 9s +[06:27:26][ 570s][FT] round 1/6 epoch 1/5 | patches=74 | loss=1.3886 +[06:27:29][ 574s][FT] round 1/6 epoch 2/5 | patches=74 | loss=1.3259 +[06:27:32][ 577s][FT] round 1/6 epoch 3/5 | patches=74 | loss=1.3300 +[06:27:35][ 580s][FT] round 1/6 epoch 4/5 | patches=74 | loss=1.3050 +[06:27:38][ 582s][FT] round 1/6 epoch 5/5 | patches=74 | loss=1.2636 +[06:27:41][ 585s][AL] round 1/6 | labeled=72 | selection_seconds=35.6000 round_seconds=206.6000 restart_student_dice=0.2157 restart_teacher_dice=0.4127 val_dice=0.5873 val_vi=0.6600 val_ece=0.0484 +[06:27:51][ 595s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.511, uniform=0.254, top1=0.449, top2=0.457, top3=0.492 -> using mixture +[06:27:51][ 595s][WEIGHTS] round 2: teacher Dice on 144 labeled patches + 2 blank (2038859 fg px): c-elegans=0.388, flywire=0.149, berghia=0.170, microns=0.291, oct-glia=0.178, oct-sfl=0.419, snemi=0.405, mouse-brain=0.449 | weights: c-elegans=0.069, flywire=0.019, berghia=0.002, microns=0.216, oct-glia=0.007, oct-sfl=0.100, snemi=0.172, mouse-brain=0.415 +[06:28:02][ 606s][CONS] round 2: on the 72 new patches student Dice=0.772 vs ensemble 0.499 -> pseudo-labels from the student | consistency x0.91 +[06:28:06][ 611s][FT] round 2/6 epoch 1/5 | patches=146 | loss=1.0746 +[06:28:11][ 615s][FT] round 2/6 epoch 2/5 | patches=146 | loss=1.0634 +[06:28:15][ 620s][FT] round 2/6 epoch 3/5 | patches=146 | loss=1.0433 +[06:28:20][ 624s][FT] round 2/6 epoch 4/5 | patches=146 | loss=1.0409 +[06:28:24][ 629s][FT] round 2/6 epoch 5/5 | patches=146 | loss=1.0049 +[06:28:27][ 631s][AL] round 2/6 | labeled=144 | selection_seconds=0.2000 round_seconds=43.8000 pseudo_student_dice=0.7723 pseudo_ensemble_dice=0.4989 consistency_scale=0.9080 val_dice=0.6586 val_vi=0.5740 val_ece=0.0054 +[06:29:21][ 685s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.442, flywire=0.155, berghia=0.189, microns=0.287, oct-glia=0.186, oct-sfl=0.457, snemi=0.439, mouse-brain=0.516 | blended with labeled Dice at 0.50 +[06:30:11][ 735s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.545, uniform=0.266, top1=0.485, top2=0.491, top3=0.510 -> using mixture +[06:30:11][ 735s][WEIGHTS] round 3: teacher Dice on 216 labeled patches + 2 blank (3013404 fg px): c-elegans=0.414, flywire=0.154, berghia=0.180, microns=0.291, oct-glia=0.183, oct-sfl=0.442, snemi=0.418, mouse-brain=0.485 | weights: c-elegans=0.072, flywire=0.016, berghia=0.001, microns=0.172, oct-glia=0.006, oct-sfl=0.101, snemi=0.158, mouse-brain=0.474 +[06:30:21][ 746s][CONS] round 3: on the 72 new patches student Dice=0.800 vs ensemble 0.507 -> pseudo-labels from the student | consistency x1.00 +[06:30:27][ 751s][FT] round 3/6 epoch 1/5 | patches=218 | loss=0.9337 +[06:30:32][ 757s][FT] round 3/6 epoch 2/5 | patches=218 | loss=0.9435 +[06:30:38][ 762s][FT] round 3/6 epoch 3/5 | patches=218 | loss=0.8940 +[06:30:43][ 768s][FT] round 3/6 epoch 4/5 | patches=218 | loss=0.9051 +[06:30:49][ 773s][FT] round 3/6 epoch 5/5 | patches=218 | loss=0.8813 +[06:30:51][ 776s][AL] round 3/6 | labeled=216 | selection_seconds=0.2000 round_seconds=142.6000 pseudo_student_dice=0.7999 pseudo_ensemble_dice=0.5074 consistency_scale=1.0000 val_dice=0.6489 val_vi=0.5501 val_ece=0.0084 +[06:31:46][ 831s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.393, flywire=0.183, berghia=0.188, microns=0.344, oct-glia=0.231, oct-sfl=0.497, snemi=0.420, mouse-brain=0.455 | blended with labeled Dice at 0.50 +[06:32:38][ 882s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.493, uniform=0.298, top1=0.466, top2=0.469, top3=0.578 -> using top3 +[06:32:38][ 882s][WEIGHTS] round 4: teacher Dice on 288 labeled patches + 2 blank (4032333 fg px): c-elegans=0.389, flywire=0.168, berghia=0.180, microns=0.323, oct-glia=0.203, oct-sfl=0.466, snemi=0.419, mouse-brain=0.450 | weights: c-elegans=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.206, snemi=0.255, mouse-brain=0.539 +[06:32:49][ 893s][CONS] round 4: on the 72 new patches student Dice=0.783 vs ensemble 0.571 -> pseudo-labels from the student | consistency x0.94 +[06:32:56][ 900s][FT] round 4/6 epoch 1/5 | patches=290 | loss=0.8305 +[06:33:03][ 907s][FT] round 4/6 epoch 2/5 | patches=290 | loss=0.8281 +[06:33:10][ 914s][FT] round 4/6 epoch 3/5 | patches=290 | loss=0.7818 +[06:33:16][ 921s][FT] round 4/6 epoch 4/5 | patches=290 | loss=0.7958 +[06:33:23][ 927s][FT] round 4/6 epoch 5/5 | patches=290 | loss=0.8050 +[06:33:25][ 930s][AL] round 4/6 | labeled=288 | selection_seconds=0.2000 round_seconds=151.8000 pseudo_student_dice=0.7826 pseudo_ensemble_dice=0.5707 consistency_scale=0.9420 val_dice=0.6588 val_vi=0.5911 val_ece=0.0125 +[06:34:22][ 987s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.455, flywire=0.146, berghia=0.189, microns=0.268, oct-glia=0.170, oct-sfl=0.425, snemi=0.427, mouse-brain=0.522 | blended with labeled Dice at 0.50 +[06:35:16][ 1040s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.528, uniform=0.258, top1=0.480, top2=0.483, top3=0.500 -> using mixture +[06:35:16][ 1040s][WEIGHTS] round 5: teacher Dice on 360 labeled patches + 2 blank (4939842 fg px): c-elegans=0.419, flywire=0.150, berghia=0.182, microns=0.284, oct-glia=0.175, oct-sfl=0.432, snemi=0.422, mouse-brain=0.480 | weights: c-elegans=0.078, flywire=0.016, berghia=0.001, microns=0.167, oct-glia=0.005, oct-sfl=0.095, snemi=0.170, mouse-brain=0.467 +[06:35:27][ 1051s][CONS] round 5: on the 72 new patches student Dice=0.803 vs ensemble 0.467 -> pseudo-labels from the student | consistency x1.00 +[06:35:35][ 1059s][FT] round 5/6 epoch 1/5 | patches=362 | loss=0.8099 +[06:35:43][ 1067s][FT] round 5/6 epoch 2/5 | patches=362 | loss=0.8055 +[06:35:51][ 1075s][FT] round 5/6 epoch 3/5 | patches=362 | loss=0.7721 +[06:35:59][ 1083s][FT] round 5/6 epoch 4/5 | patches=362 | loss=0.7884 +[06:36:07][ 1091s][FT] round 5/6 epoch 5/5 | patches=362 | loss=0.7350 +[06:36:09][ 1093s][AL] round 5/6 | labeled=360 | selection_seconds=0.2000 round_seconds=161.3000 pseudo_student_dice=0.8026 pseudo_ensemble_dice=0.4667 consistency_scale=1.0000 val_dice=0.6882 val_vi=0.5241 val_ece=0.0292 +[06:37:07][ 1151s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.405, flywire=0.161, berghia=0.183, microns=0.297, oct-glia=0.199, oct-sfl=0.448, snemi=0.416, mouse-brain=0.481 | blended with labeled Dice at 0.50 +[06:38:02][ 1206s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.516, uniform=0.277, top1=0.456, top2=0.466, top3=0.531 -> using top3 +[06:38:02][ 1206s][WEIGHTS] round 6: teacher Dice on 432 labeled patches + 2 blank (5814272 fg px): c-elegans=0.392, flywire=0.159, berghia=0.179, microns=0.301, oct-glia=0.190, oct-sfl=0.449, snemi=0.417, mouse-brain=0.456 | weights: c-elegans=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.175, snemi=0.253, mouse-brain=0.572 +[06:38:13][ 1217s][CONS] round 6: on the 72 new patches student Dice=0.804 vs ensemble 0.511 -> pseudo-labels from the student | consistency x1.00 +[06:38:22][ 1226s][FT] round 6/6 epoch 1/5 | patches=434 | loss=0.6374 +[06:38:31][ 1235s][FT] round 6/6 epoch 2/5 | patches=434 | loss=0.6052 +[06:38:40][ 1244s][FT] round 6/6 epoch 3/5 | patches=434 | loss=0.6068 +[06:38:49][ 1253s][FT] round 6/6 epoch 4/5 | patches=434 | loss=0.5832 +[06:38:58][ 1262s][FT] round 6/6 epoch 5/5 | patches=434 | loss=0.6386 +[06:39:00][ 1265s][AL] round 6/6 | labeled=432 | selection_seconds=0.2000 round_seconds=169.2000 pseudo_student_dice=0.8042 pseudo_ensemble_dice=0.5110 consistency_scale=1.0000 val_dice=0.6883 val_vi=0.5339 val_ece=0.0105 +[06:39:04][ 1268s][CONS] final consolidation: pseudo-labels refreshed from the student +[06:39:13][ 1277s][FT] final 1/10 epoch 1/1 | patches=434 | loss=0.6010 +[06:39:22][ 1286s][FT] final 2/10 epoch 1/1 | patches=434 | loss=0.5822 +[06:39:31][ 1296s][FT] final 3/10 epoch 1/1 | patches=434 | loss=0.5734 +[06:39:41][ 1305s][FT] final 4/10 epoch 1/1 | patches=434 | loss=0.5643 +[06:39:50][ 1314s][FT] final 5/10 epoch 1/1 | patches=434 | loss=0.5516 +[06:39:59][ 1323s][FT] final 6/10 epoch 1/1 | patches=434 | loss=0.5410 +[06:40:08][ 1332s][FT] final 7/10 epoch 1/1 | patches=434 | loss=0.5213 +[06:40:17][ 1341s][FT] final 8/10 epoch 1/1 | patches=434 | loss=0.5250 +[06:40:26][ 1351s][FT] final 9/10 epoch 1/1 | patches=434 | loss=0.5119 +[06:40:36][ 1360s][FT] final 10/10 epoch 1/1 | patches=434 | loss=0.5281 +[06:40:38][ 1362s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=95.4000 val_dice=0.7122 val_vi=0.5155 val_ece=0.0282 +[06:40:38][ 1362s][WEIGHTS] final: c-elegans=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.175, snemi=0.253, mouse-brain=0.572 +[06:41:17][ 1401s][TEST] raw student (no EMA, no TTA): patch_dice=0.7470 patch_vi=0.4291 +[06:41:22][ 1407s][TEST] final model without TTA: patch_dice=0.7470 patch_vi=0.4291 +[06:41:22][ 1407s][RESOURCES] {"phases": {"setup": {"seconds": 1.22, "peak_cpu_rss_mb": 1028.3, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 32.54, "peak_cpu_rss_mb": 1801.3, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 339.71, "peak_cpu_rss_mb": 3693.9, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 986.15, "peak_cpu_rss_mb": 3693.9, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4642.0, "rounds": 6}, "inference": {"seconds": 44.7, "peak_cpu_rss_mb": 3693.9, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4644.0, "patches": 2883, "images": 3, "patches_per_second": 64.5, "seconds_per_image": 14.9}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 3693.9, "total_seconds": 1404.3} +[06:41:22][ 1407s][TEST] patch_dice=0.7784 patch_iou=0.6821 patch_precision=0.8233 patch_recall=0.8023 patch_vi=0.4046 image_dice=0.8283 image_iou=0.7096 image_precision=0.8551 image_recall=0.8058 image_vi=0.4575 cal_ece=0.0084 cal_nll=0.1759 cal_brier=0.0526 cal_mean_confidence=0.9360 cal_accuracy=0.9277 cal_predicted_prevalence=0.2215 cal_true_prevalence=0.2301 cal_mean_probability=0.2258 +[06:41:23][ 1407s][DONE] dhanyasi-P14-mouse-cerebellum macs-v11-norestart-acq-random b5 s2 | test patch Dice=0.7784 VI=0.4046 | image Dice=0.8283 VI=0.4575 | 1407s diff --git a/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/result.json b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..eca865e181848c3005cae6ff1b7eebb21a4211f3 --- /dev/null +++ b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/result.json @@ -0,0 +1,785 @@ +{ + "target": "dhanyasi-P14-mouse-cerebellum", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 2, + "config": { + "task": "membrane", + "target": "dhanyasi-P14-mouse-cerebellum", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "random", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 0, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "dhanyasi-P14-mouse-cerebellum", + "--variants", + "macs-v11-norestart-acq-random", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d", + "whole-mouse-brain" + ], + "alphas": [ + 0.05083387092965836, + 0.15580245740591697, + 0.010243446832791585, + 0.41728942804507835, + 0.03964191869924105, + 0.053950781313640436, + 0.10715830600341246, + 0.16507979077026083 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d", + "whole-mouse-brain" + ], + "mixture": [ + 0.05083387092965836, + 0.15580245740591697, + 0.010243446832791585, + 0.41728942804507835, + 0.03964191869924105, + 0.053950781313640436, + 0.10715830600341246, + 0.16507979077026083 + ], + "per_source": [ + 0.1831792908635141, + 0.0008117602305311557, + 0.07130039492253423, + 0.2901333759297942, + 0.19473854254781, + 0.14427929446206111, + 0.11165168345469406, + 0.0039056575890611817 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.04944674080252824, + "mmd2_mixture": 0.06619580859842833, + "mmd2_uniform": 0.10385692805437269, + "mmd2_single_source": [ + 0.35157092449944216, + 0.1914113737092883, + 0.2296477365874025, + 0.08478080568303437, + 0.272622477181822, + 0.2463089382682188, + 0.15004801624266395, + 0.17311813097471462 + ], + "median_squared_distance": 0.18230687630230707, + "gammas": [ + 1.3713141548508683, + 2.7426283097017365, + 5.485256619403473, + 10.970513238806946, + 21.941026477613892 + ], + "max_weight": 0.41728942804507835, + "entropy": 1.6749067795900086, + "effective_sources": 5.338297487596983, + "min_weight": 0.010243446832791585 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.24983749105579034, + 0.6076273510548802, + 0.31213616057205607, + 0.2194742775498013, + 0.24579726709659655, + 0.26559858325874997, + 0.28252517774848496, + 0.5039038205278614 + ], + "standardized": [ + -0.6514591536335823, + 2.0580486812336294, + -0.17967738442258652, + -0.8813968184025964, + -0.6820553769924043, + -0.5321019342588011, + -0.4039184793457027, + 1.2725604658220449 + ], + "tau": 0.5, + "max_weight": 0.2901333759297942, + "entropy": 1.7283708993213025, + "effective_sources": 5.631472196315185, + "min_weight": 0.0008117602305311557 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "microns_basil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.41286983026930635, + 0.10651095633970442, + 0.15770527050491268, + 0.22267858774357308, + 0.11006956393332956, + 0.2911632103623596, + 0.2970295092802324, + 0.447529566690438 + ], + "ensemble_val_dice": { + "mixture": 0.20976331913409396, + "mixture@raw": 0.2201044706093569, + "mixture@logit": 0.1889936769494014, + "per_source": 0.19317883340253061, + "per_source@raw": 0.20883607926566583, + "per_source@logit": 0.15551849586998046, + "uniform": 0.19238025487255683, + "uniform@raw": 0.21034175472151873, + "uniform@logit": 0.15367464892063687 + }, + "val_patches": 1024 + }, + "temperatures": [ + 2.575807571411133, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 2.6421806812286377, + 3.5461363792419434, + 4.075904369354248 + ], + "top_source": "microns_basil", + "weight_stats": { + "max_weight": 0.41728942804507835, + "entropy": 1.6749067795900086, + "effective_sources": 5.338297487596983, + "min_weight": 0.010243446832791585 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.2380952380952381, + "spearman_weight_taxonomy": 0.7324670207647144, + "top_source": "microns_basil", + "top_source_is_best_teacher": false + }, + "init_teacher": "microns_basil", + "student_init": "top-source", + "blank_fraction": 0.00023124060585038732, + "kd_seconds": 339.7, + "round0_val": { + "val_dice": 0.1711621108784323, + "val_vi": 0.5020588886973317, + "val_ece": 0.14820936102204899 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 72, + "selection_seconds": 35.6, + "round_seconds": 206.6, + "teacher_dice_labeled": [ + 0.34699131258491106, + 0.15400938239211243, + 0.17063725881990668, + 0.29518215496468825, + 0.1907363491430849, + 0.40286997571879657, + 0.402560670886263, + 0.4126936934594237 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 891480, + "evidence_nonempty_patches": 58, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.21571648537522917, + "restart_teacher_dice": 0.4126936934594237, + "weights": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.1573376001938336, + 0.311542520327699, + 0.5311198794784675 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.4486622376468838, + "uniform": 0.257123442242594, + "top1": 0.4126936934594237, + "top2": 0.427720538104064, + "top3": 0.5579440810273272 + }, + "val_dice": 0.587305604106643, + "val_vi": 0.6599535991066656, + "val_ece": 0.04835740491699966 + }, + { + "round": 2, + "n_labeled": 144, + "selection_seconds": 0.2, + "round_seconds": 43.8, + "teacher_dice_labeled": [ + 0.3877112948623053, + 0.14909626611196175, + 0.17005234419964096, + 0.29110798999593296, + 0.17764307442922855, + 0.41866300287842756, + 0.40459903528437674, + 0.4492049901441661 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 2038859, + "evidence_nonempty_patches": 117, + "first_correction": false, + "restart": "", + "weights": [ + 0.06909489706776745, + 0.019479382824218353, + 0.0015792762266002333, + 0.21586754477923684, + 0.006593754437182248, + 0.09993372841912157, + 0.17244918911249582, + 0.4150022271333776 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5114063614291914, + "uniform": 0.2542809724207618, + "top1": 0.4492049901441661, + "top2": 0.4573143432435651, + "top3": 0.4917913428812348 + }, + "pseudo_student_dice": 0.7723456432892528, + "pseudo_ensemble_dice": 0.4989113809299943, + "pseudo_source": "student", + "consistency_scale": 0.908, + "val_dice": 0.6585940233579239, + "val_vi": 0.5740162973538123, + "val_ece": 0.005417245034970651 + }, + { + "round": 3, + "n_labeled": 216, + "selection_seconds": 0.2, + "round_seconds": 142.6, + "teacher_dice_labeled": [ + 0.4140682111607643, + 0.1536399927755801, + 0.1799205572303296, + 0.29112126324332566, + 0.18334845976830494, + 0.4423851087364882, + 0.41803091046534124, + 0.4848902568732807 + ], + "teacher_dice_pool": [ + 0.4416778556160547, + 0.15487823535765688, + 0.18914476821851361, + 0.28655053365525646, + 0.18583099402223757, + 0.4567643200946957, + 0.43939763997607106, + 0.5156445163650382 + ], + "evidence_fg_pixels": 3013404, + "evidence_nonempty_patches": 174, + "first_correction": false, + "restart": "", + "weights": [ + 0.0718401217611655, + 0.016284063956197966, + 0.001392418083738922, + 0.17246439507692587, + 0.0055765473665446265, + 0.10120229003831627, + 0.157561280540448, + 0.4736788831766628 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5454374977215515, + "uniform": 0.2664851046953125, + "top1": 0.4848902568732807, + "top2": 0.49108429525345076, + "top3": 0.5097392015266642 + }, + "pseudo_student_dice": 0.7999457618489225, + "pseudo_ensemble_dice": 0.5073909859154929, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.6488990698124532, + "val_vi": 0.5500686787278544, + "val_ece": 0.008365029822864706 + }, + { + "round": 4, + "n_labeled": 288, + "selection_seconds": 0.2, + "round_seconds": 151.8, + "teacher_dice_labeled": [ + 0.38945019463626285, + 0.1679853583375917, + 0.18044124015986018, + 0.32282208176596117, + 0.2034846994902197, + 0.4660624113903503, + 0.41870020084547344, + 0.4503311467877327 + ], + "teacher_dice_pool": [ + 0.3928618027754761, + 0.18289820943264143, + 0.1883542828074731, + 0.3437577001051289, + 0.2306743626695978, + 0.4966166642733409, + 0.4197338102857259, + 0.45470006684976394 + ], + "evidence_fg_pixels": 4032333, + "evidence_nonempty_patches": 233, + "first_correction": false, + "restart": "", + "weights": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.2061290137522376, + 0.25496217742297783, + 0.5389088088247845 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.4927856038827534, + "uniform": 0.29765161943165275, + "top1": 0.4660624113903503, + "top2": 0.4691393882366579, + "top3": 0.5775226351451215 + }, + "pseudo_student_dice": 0.7826412650210633, + "pseudo_ensemble_dice": 0.5706861006865913, + "pseudo_source": "student", + "consistency_scale": 0.942, + "val_dice": 0.658762843387845, + "val_vi": 0.5911395322997408, + "val_ece": 0.012483891875880187 + }, + { + "round": 5, + "n_labeled": 360, + "selection_seconds": 0.2, + "round_seconds": 161.3, + "teacher_dice_labeled": [ + 0.41854435812646784, + 0.1495941701206981, + 0.1822344165135339, + 0.28407492428974873, + 0.17477511072235818, + 0.43229461743105685, + 0.42189655355246203, + 0.47989858564637067 + ], + "teacher_dice_pool": [ + 0.45500621878025627, + 0.14580300142111405, + 0.18938505602311717, + 0.2676734414185573, + 0.1697405400963424, + 0.4251377383150291, + 0.4270884498370188, + 0.5221131621647941 + ], + "evidence_fg_pixels": 4939842, + "evidence_nonempty_patches": 289, + "first_correction": false, + "restart": "", + "weights": [ + 0.0779412985155291, + 0.01622381303412828, + 0.001478357728619292, + 0.1667471831433008, + 0.005309978119330455, + 0.09491367972415017, + 0.16990207946327957, + 0.4674836102716623 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5283083464616156, + "uniform": 0.2575204836185174, + "top1": 0.47989858564637067, + "top2": 0.4829881803552951, + "top3": 0.5002578227349939 + }, + "pseudo_student_dice": 0.8026457077424155, + "pseudo_ensemble_dice": 0.46667988670073507, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.6882328190174084, + "val_vi": 0.5241002137565222, + "val_ece": 0.029186931568365077 + }, + { + "round": 6, + "n_labeled": 432, + "selection_seconds": 0.2, + "round_seconds": 169.2, + "teacher_dice_labeled": [ + 0.39221785990090063, + 0.15853969100727025, + 0.1789619795127443, + 0.3011489393039044, + 0.18995564470764165, + 0.4492783861051983, + 0.4173859453694295, + 0.45590431446624324 + ], + "teacher_dice_pool": [ + 0.4051408302204626, + 0.16061494373172766, + 0.1831960622696714, + 0.2969592804592786, + 0.19929429611917968, + 0.44823407067395027, + 0.41566445570090216, + 0.4808073840817319 + ], + "evidence_fg_pixels": 5814272, + "evidence_nonempty_patches": 348, + "first_correction": false, + "restart": "", + "weights": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.17503087706727477, + 0.2527176996415204, + 0.5722514232912048 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.5162106450575661, + "uniform": 0.27684872084241097, + "top1": 0.45590431446624324, + "top2": 0.46629336580424974, + "top3": 0.5313853044604864 + }, + "pseudo_student_dice": 0.8041609915056682, + "pseudo_ensemble_dice": 0.5110455493001167, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.6883015407537225, + "val_vi": 0.5339455101163537, + "val_ece": 0.010485531571301124 + }, + { + "round": 7, + "n_labeled": 432, + "selection_seconds": 0.0, + "round_seconds": 95.4, + "final_consolidation": true, + "val_dice": 0.7122290178272843, + "val_vi": 0.515518646087249, + "val_ece": 0.0282238460581894 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.01683200768757291, + "auroc_top20_error": 0.5015251953125 + }, + "al_seconds": 986.1, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.17503087706727477, + 0.2527176996415204, + 0.5722514232912048 + ], + "weight_stats_final": { + "max_weight": 0.5722514232912048, + "entropy": 0.9720687572709196, + "effective_sources": 2.643407374909532, + "min_weight": 0.0 + }, + "top_source_final": "whole-mouse-brain", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7470388735427181, + "patch_iou": 0.6460663585703925, + "patch_precision": 0.7807578668019558, + "patch_recall": 0.8005368189549578, + "patch_vi": 0.42906332242828876, + "n_patches": 2883, + "cal_ece": 0.02143620715811456, + "cal_nll": 0.20047653371661475, + "cal_brier": 0.05796857954842853, + "cal_mean_confidence": 0.9434937833591204, + "cal_accuracy": 0.9220575762010059, + "cal_predicted_prevalence": 0.2271959276361429, + "cal_true_prevalence": 0.23013442215140478, + "cal_mean_probability": 0.23093329968843 + }, + "test_notta": { + "patch_dice": 0.7470388735427181, + "patch_iou": 0.6460663585703925, + "patch_precision": 0.7807578668019558, + "patch_recall": 0.8005368189549578, + "patch_vi": 0.42906332242828876, + "n_patches": 2883, + "cal_ece": 0.02143620715811456, + "cal_nll": 0.20047653371661475, + "cal_brier": 0.05796857954842853, + "cal_mean_confidence": 0.9434937833591204, + "cal_accuracy": 0.9220575762010059, + "cal_predicted_prevalence": 0.2271959276361429, + "cal_true_prevalence": 0.23013442215140478, + "cal_mean_probability": 0.23093329968843 + }, + "test": { + "patch_dice": 0.7783869137129839, + "patch_iou": 0.6820751619244556, + "patch_precision": 0.8232589978608481, + "patch_recall": 0.8022611640300222, + "patch_vi": 0.4045632298190714, + "n_patches": 2883, + "image_dice": 0.8282705378907229, + "image_iou": 0.7096447795902364, + "image_precision": 0.8550598096304546, + "image_recall": 0.8057972121205764, + "image_vi": 0.45746639872723177, + "n_images": 3, + "cal_ece": 0.008398856496640952, + "cal_nll": 0.17592274256698134, + "cal_brier": 0.05262819081835747, + "cal_mean_confidence": 0.9360410674965641, + "cal_accuracy": 0.927740472164412, + "cal_predicted_prevalence": 0.22152996824054805, + "cal_true_prevalence": 0.23013442215140478, + "cal_mean_probability": 0.2258376879442032 + }, + "eval_seconds": 44.7, + "total_seconds": 1406.8, + "resources": { + "phases": { + "setup": { + "seconds": 1.22, + "peak_cpu_rss_mb": 1028.3, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 32.54, + "peak_cpu_rss_mb": 1801.3, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 339.71, + "peak_cpu_rss_mb": 3693.9, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4618.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 986.15, + "peak_cpu_rss_mb": 3693.9, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4642.0, + "rounds": 6 + }, + "inference": { + "seconds": 44.7, + "peak_cpu_rss_mb": 3693.9, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4644.0, + "patches": 2883, + "images": 3, + "patches_per_second": 64.5, + "seconds_per_image": 14.9 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 3693.9, + "total_seconds": 1404.3 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..8cb50520b15a5d3bbcd62cff65bbf4e1a791b3b2 --- /dev/null +++ b/macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,72,35.6,206.6,"[0.34699131258491106, 0.15400938239211243, 0.17063725881990668, 0.29518215496468825, 0.1907363491430849, 0.40286997571879657, 0.402560670886263, 0.4126936934594237]",,891480,58,True,kept,0.21571648537522917,0.4126936934594237,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.1573376001938336, 0.311542520327699, 0.5311198794784675]",top3,"{'mixture': 0.4486622376468838, 'uniform': 0.257123442242594, 'top1': 0.4126936934594237, 'top2': 0.427720538104064, 'top3': 0.5579440810273272}",0.587305604106643,0.6599535991066656,0.04835740491699966,,,,, +2,144,0.2,43.8,"[0.3877112948623053, 0.14909626611196175, 0.17005234419964096, 0.29110798999593296, 0.17764307442922855, 0.41866300287842756, 0.40459903528437674, 0.4492049901441661]",,2038859,117,False,,,,"[0.06909489706776745, 0.019479382824218353, 0.0015792762266002333, 0.21586754477923684, 0.006593754437182248, 0.09993372841912157, 0.17244918911249582, 0.4150022271333776]",mixture,"{'mixture': 0.5114063614291914, 'uniform': 0.2542809724207618, 'top1': 0.4492049901441661, 'top2': 0.4573143432435651, 'top3': 0.4917913428812348}",0.6585940233579239,0.5740162973538123,0.005417245034970651,0.7723456432892528,0.4989113809299943,student,0.908, +3,216,0.2,142.6,"[0.4140682111607643, 0.1536399927755801, 0.1799205572303296, 0.29112126324332566, 0.18334845976830494, 0.4423851087364882, 0.41803091046534124, 0.4848902568732807]","[0.4416778556160547, 0.15487823535765688, 0.18914476821851361, 0.28655053365525646, 0.18583099402223757, 0.4567643200946957, 0.43939763997607106, 0.5156445163650382]",3013404,174,False,,,,"[0.0718401217611655, 0.016284063956197966, 0.001392418083738922, 0.17246439507692587, 0.0055765473665446265, 0.10120229003831627, 0.157561280540448, 0.4736788831766628]",mixture,"{'mixture': 0.5454374977215515, 'uniform': 0.2664851046953125, 'top1': 0.4848902568732807, 'top2': 0.49108429525345076, 'top3': 0.5097392015266642}",0.6488990698124532,0.5500686787278544,0.008365029822864706,0.7999457618489225,0.5073909859154929,student,1.0, +4,288,0.2,151.8,"[0.38945019463626285, 0.1679853583375917, 0.18044124015986018, 0.32282208176596117, 0.2034846994902197, 0.4660624113903503, 0.41870020084547344, 0.4503311467877327]","[0.3928618027754761, 0.18289820943264143, 0.1883542828074731, 0.3437577001051289, 0.2306743626695978, 0.4966166642733409, 0.4197338102857259, 0.45470006684976394]",4032333,233,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.2061290137522376, 0.25496217742297783, 0.5389088088247845]",top3,"{'mixture': 0.4927856038827534, 'uniform': 0.29765161943165275, 'top1': 0.4660624113903503, 'top2': 0.4691393882366579, 'top3': 0.5775226351451215}",0.658762843387845,0.5911395322997408,0.012483891875880187,0.7826412650210633,0.5706861006865913,student,0.942, +5,360,0.2,161.3,"[0.41854435812646784, 0.1495941701206981, 0.1822344165135339, 0.28407492428974873, 0.17477511072235818, 0.43229461743105685, 0.42189655355246203, 0.47989858564637067]","[0.45500621878025627, 0.14580300142111405, 0.18938505602311717, 0.2676734414185573, 0.1697405400963424, 0.4251377383150291, 0.4270884498370188, 0.5221131621647941]",4939842,289,False,,,,"[0.0779412985155291, 0.01622381303412828, 0.001478357728619292, 0.1667471831433008, 0.005309978119330455, 0.09491367972415017, 0.16990207946327957, 0.4674836102716623]",mixture,"{'mixture': 0.5283083464616156, 'uniform': 0.2575204836185174, 'top1': 0.47989858564637067, 'top2': 0.4829881803552951, 'top3': 0.5002578227349939}",0.6882328190174084,0.5241002137565222,0.029186931568365077,0.8026457077424155,0.46667988670073507,student,1.0, +6,432,0.2,169.2,"[0.39221785990090063, 0.15853969100727025, 0.1789619795127443, 0.3011489393039044, 0.18995564470764165, 0.4492783861051983, 0.4173859453694295, 0.45590431446624324]","[0.4051408302204626, 0.16061494373172766, 0.1831960622696714, 0.2969592804592786, 0.19929429611917968, 0.44823407067395027, 0.41566445570090216, 0.4808073840817319]",5814272,348,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.17503087706727477, 0.2527176996415204, 0.5722514232912048]",top3,"{'mixture': 0.5162106450575661, 'uniform': 0.27684872084241097, 'top1': 0.45590431446624324, 'top2': 0.46629336580424974, 'top3': 0.5313853044604864}",0.6883015407537225,0.5339455101163537,0.010485531571301124,0.8041609915056682,0.5110455493001167,student,1.0, +7,432,0.0,95.4,,,,,,,,,,,,0.7122290178272843,0.515518646087249,0.0282238460581894,,,,,True diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/log.txt b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..105026f673f537cbb4b7404da234b2a6e5f2f166 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/log.txt @@ -0,0 +1,107 @@ +[06:56:31][ 0s][START] target=octopus-vulgaris-vertical-lobe-sfltract variant=macs-v11-norestart-acq-entropy budget=5.0% seed=2 | {"task": "membrane", "target": "octopus-vulgaris-vertical-lobe-sfltract", "variant": "macs-v11-norestart-acq-entropy", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "entropy", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[06:56:31][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[06:56:31][ 0s][DATA] octopus-vulgaris-vertical-lobe-sfltract: train=3584 val=768 test=768 patches +[06:56:33][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, snemi=3.55, mouse-brain=4.08 +[06:56:33][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[06:56:39][ 7s][WEIGHTS] mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 | eff.sources=3.41 | MMD2 mix=0.0128 uniform=0.0970 +[06:56:39][ 7s][WEIGHTS] per-source: c-elegans=0.320, cerebellum=0.017, flywire=0.001, berghia=0.175, microns=0.027, oct-glia=0.436, snemi=0.017, mouse-brain=0.007 | min-MMD source (NeuroADDA rule) = oct-glia +[06:56:57][ 26s][TEACHERS] target-val Dice: c-elegans=0.476, cerebellum=0.633, flywire=0.142, berghia=0.467, microns=0.708, oct-glia=0.845, snemi=0.229, mouse-brain=0.115 +[06:56:57][ 26s][TEACHERS] ensemble target-val Dice: mixture=0.849, mixture@raw=0.847, mixture@logit=0.850, per_source=0.843, per_source@raw=0.619, per_source@logit=0.848, uniform=0.718, uniform@raw=0.677, uniform@logit=0.807 +[06:56:57][ 26s][WEIGHTS] using mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 +[06:56:57][ 26s][INIT] student initialised from teacher oct-glia (weight 0.636) +[06:56:58][ 27s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[06:56:58][ 27s][KD] distilling the mixture ensemble (prob space) into the student +[06:57:39][ 68s][KD] soft labels for 3584 patches from 8 teachers in 41s +[06:58:12][ 101s][KD] epoch 1/5 | patches=3584 | loss=0.6002 | 33s +[06:58:45][ 133s][KD] epoch 2/5 | patches=3584 | loss=0.4596 | 33s +[06:59:17][ 166s][KD] epoch 3/5 | patches=3584 | loss=0.4055 | 33s +[06:59:50][ 199s][KD] epoch 4/5 | patches=3584 | loss=0.3664 | 33s +[07:00:23][ 232s][KD] epoch 5/5 | patches=3584 | loss=0.3377 | 33s +[07:00:24][ 233s][ROUND0] before any target label: val_dice=0.8442 val_vi=0.1463 +[07:00:24][ 233s][AL] patches=3584 (pool 2338 after dropping blank; free blank negatives: True) | budget=179 (5.0%) | rounds=6 | per_round=29 | acquisition=entropy | first_round=stratified +[07:00:28][ 237s][AL] round-1 uncertainty quality on 2338 patches: {'spearman_uncertainty_error': 0.4491343304947581, 'auroc_top20_error': 0.6088995155171626} +[07:00:51][ 260s][AL] round 1: stratified cold start: 15 patches from the membrane-rich half (ensemble prevalence >= 0.192) + 14 from the membrane-poor half of 2000 candidates +[07:00:54][ 263s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.765, uniform=0.741, top1=0.765, top2=0.764, top3=0.765 -> using mixture +[07:00:55][ 263s][WEIGHTS] round 1: student kept (labeled Dice 0.748 vs best teacher oct-glia 0.765); distilling the corrected ensemble into it +[07:00:55][ 263s][WEIGHTS] round 1: teacher Dice on 29 labeled patches + 29 blank (179039 fg px): c-elegans=0.257, cerebellum=0.569, flywire=0.169, berghia=0.276, microns=0.581, oct-glia=0.765, snemi=0.110, mouse-brain=0.203 | weights: c-elegans=0.001, cerebellum=0.006, flywire=0.000, berghia=0.002, microns=0.008, oct-glia=0.983, snemi=0.000, mouse-brain=0.000 +[07:01:21][ 290s][KD] soft labels for 2309 patches from 8 teachers in 26s +[07:01:42][ 311s][KD] epoch 1/2 | patches=2309 | loss=0.1395 | 21s +[07:02:03][ 332s][KD] epoch 2/2 | patches=2309 | loss=0.1332 | 21s +[07:02:26][ 355s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[07:02:28][ 356s][FT] round 1/6 epoch 1/5 | patches=58 | loss=0.8560 +[07:02:29][ 358s][FT] round 1/6 epoch 2/5 | patches=58 | loss=0.5224 +[07:02:30][ 359s][FT] round 1/6 epoch 3/5 | patches=58 | loss=0.4384 +[07:02:32][ 361s][FT] round 1/6 epoch 4/5 | patches=58 | loss=0.3711 +[07:02:33][ 362s][FT] round 1/6 epoch 5/5 | patches=58 | loss=0.3721 +[07:02:35][ 364s][AL] round 1/6 | labeled=29 | selection_seconds=26.9000 round_seconds=129.3000 restart_student_dice=0.7483 restart_teacher_dice=0.7648 val_dice=0.8506 val_vi=0.1502 val_ece=0.0197 +[07:02:39][ 367s][AL] round 2: 595/2309 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:02:44][ 373s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.578, uniform=0.590, top1=0.578, top2=0.577, top3=0.577 -> using uniform +[07:02:44][ 373s][WEIGHTS] round 2: teacher Dice on 58 labeled patches + 58 blank (666295 fg px): c-elegans=0.305, cerebellum=0.438, flywire=0.166, berghia=0.275, microns=0.457, oct-glia=0.578, snemi=0.187, mouse-brain=0.290 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[07:02:47][ 376s][CONS] round 2: on the 29 new patches student Dice=0.527 vs ensemble 0.520 -> pseudo-labels from the ensemble | consistency x1.00 +[07:02:49][ 378s][FT] round 2/6 epoch 1/5 | patches=116 | loss=0.9634 +[07:02:52][ 381s][FT] round 2/6 epoch 2/5 | patches=116 | loss=0.8654 +[07:02:54][ 383s][FT] round 2/6 epoch 3/5 | patches=116 | loss=0.8580 +[07:02:56][ 385s][FT] round 2/6 epoch 4/5 | patches=116 | loss=0.7631 +[07:02:59][ 388s][FT] round 2/6 epoch 5/5 | patches=116 | loss=0.6561 +[07:03:00][ 389s][AL] round 2/6 | labeled=58 | selection_seconds=3.7000 round_seconds=23.9000 pseudo_student_dice=0.5266 pseudo_ensemble_dice=0.5199 consistency_scale=1.0000 val_dice=0.8430 val_vi=0.1733 val_ece=0.0119 +[07:03:04][ 393s][AL] round 3: 550/2280 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:03:13][ 401s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.425, uniform=0.447, top1=0.423, top2=0.424, top3=0.423 -> using uniform +[07:03:13][ 401s][WEIGHTS] round 3: teacher Dice on 87 labeled patches + 87 blank (1340584 fg px): c-elegans=0.257, cerebellum=0.358, flywire=0.144, berghia=0.219, microns=0.340, oct-glia=0.423, snemi=0.249, mouse-brain=0.287 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[07:03:18][ 407s][CONS] round 3: on the 29 new patches student Dice=0.484 vs ensemble 0.251 -> pseudo-labels from the student | consistency x0.00 +[07:03:20][ 409s][FT] round 3/6 epoch 1/5 | patches=174 | loss=0.8398 +[07:03:22][ 411s][FT] round 3/6 epoch 2/5 | patches=174 | loss=0.7916 +[07:03:24][ 413s][FT] round 3/6 epoch 3/5 | patches=174 | loss=1.0183 +[07:03:26][ 414s][FT] round 3/6 epoch 4/5 | patches=174 | loss=0.8409 +[07:03:27][ 416s][FT] round 3/6 epoch 5/5 | patches=174 | loss=0.8439 +[07:03:29][ 418s][AL] round 3/6 | labeled=87 | selection_seconds=3.9000 round_seconds=26.9000 pseudo_student_dice=0.4843 pseudo_ensemble_dice=0.2510 consistency_scale=0.0000 val_dice=0.7727 val_vi=0.1726 val_ece=0.0198 +[07:03:32][ 421s][AL] round 4: 574/2251 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:04:24][ 473s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.286, cerebellum=0.521, flywire=0.234, berghia=0.315, microns=0.554, oct-glia=0.793, snemi=0.220, mouse-brain=0.190 | blended with labeled Dice at 0.50 +[07:05:13][ 522s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.630, uniform=0.587, top1=0.630, top2=0.630, top3=0.630 -> using mixture +[07:05:13][ 522s][WEIGHTS] round 4: teacher Dice on 116 labeled patches + 116 blank (1701624 fg px): c-elegans=0.294, cerebellum=0.456, flywire=0.192, berghia=0.285, microns=0.465, oct-glia=0.630, snemi=0.225, mouse-brain=0.248 | weights: c-elegans=0.005, cerebellum=0.008, flywire=0.000, berghia=0.007, microns=0.009, oct-glia=0.969, snemi=0.001, mouse-brain=0.001 +[07:05:19][ 527s][CONS] round 4: on the 29 new patches student Dice=0.674 vs ensemble 0.607 -> pseudo-labels from the student | consistency x0.58 +[07:05:23][ 532s][FT] round 4/6 epoch 1/5 | patches=232 | loss=0.6445 +[07:05:27][ 536s][FT] round 4/6 epoch 2/5 | patches=232 | loss=0.6586 +[07:05:32][ 540s][FT] round 4/6 epoch 3/5 | patches=232 | loss=0.5856 +[07:05:36][ 545s][FT] round 4/6 epoch 4/5 | patches=232 | loss=0.5562 +[07:05:40][ 549s][FT] round 4/6 epoch 5/5 | patches=232 | loss=0.4736 +[07:05:42][ 551s][AL] round 4/6 | labeled=116 | selection_seconds=3.8000 round_seconds=131.4000 pseudo_student_dice=0.6742 pseudo_ensemble_dice=0.6070 consistency_scale=0.5810 val_dice=0.8448 val_vi=0.1761 val_ece=0.0075 +[07:05:45][ 554s][AL] round 5: 536/2222 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:06:38][ 607s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.409, cerebellum=0.640, flywire=0.238, berghia=0.380, microns=0.605, oct-glia=0.820, snemi=0.314, mouse-brain=0.283 | blended with labeled Dice at 0.50 +[07:07:28][ 657s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.657, uniform=0.657, top1=0.656, top2=0.657, top3=0.656 -> using mixture +[07:07:28][ 657s][WEIGHTS] round 5: teacher Dice on 145 labeled patches + 145 blank (1973658 fg px): c-elegans=0.353, cerebellum=0.514, flywire=0.195, berghia=0.320, microns=0.494, oct-glia=0.656, snemi=0.263, mouse-brain=0.290 | weights: c-elegans=0.007, cerebellum=0.010, flywire=0.000, berghia=0.008, microns=0.009, oct-glia=0.963, snemi=0.001, mouse-brain=0.001 +[07:07:34][ 663s][CONS] round 5: on the 29 new patches student Dice=0.689 vs ensemble 0.631 -> pseudo-labels from the student | consistency x0.63 +[07:07:39][ 668s][FT] round 5/6 epoch 1/5 | patches=290 | loss=0.4823 +[07:07:44][ 673s][FT] round 5/6 epoch 2/5 | patches=290 | loss=0.4024 +[07:07:50][ 679s][FT] round 5/6 epoch 3/5 | patches=290 | loss=0.4317 +[07:07:55][ 684s][FT] round 5/6 epoch 4/5 | patches=290 | loss=0.3859 +[07:08:00][ 689s][FT] round 5/6 epoch 5/5 | patches=290 | loss=0.4092 +[07:08:02][ 691s][AL] round 5/6 | labeled=145 | selection_seconds=3.8000 round_seconds=138.5000 pseudo_student_dice=0.6891 pseudo_ensemble_dice=0.6310 consistency_scale=0.6300 val_dice=0.5182 val_vi=0.1551 val_ece=0.0061 +[07:08:05][ 694s][AL] round 6: 551/2193 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:08:59][ 748s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.376, cerebellum=0.625, flywire=0.243, berghia=0.368, microns=0.609, oct-glia=0.861, snemi=0.290, mouse-brain=0.253 | blended with labeled Dice at 0.50 +[07:09:51][ 799s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.682, uniform=0.662, top1=0.682, top2=0.682, top3=0.682 -> using mixture +[07:09:51][ 799s][WEIGHTS] round 6: teacher Dice on 174 labeled patches + 174 blank (2295762 fg px): c-elegans=0.338, cerebellum=0.514, flywire=0.198, berghia=0.319, microns=0.501, oct-glia=0.682, snemi=0.248, mouse-brain=0.275 | weights: c-elegans=0.005, cerebellum=0.008, flywire=0.000, berghia=0.006, microns=0.008, oct-glia=0.972, snemi=0.000, mouse-brain=0.001 +[07:09:56][ 805s][CONS] round 6: on the 29 new patches student Dice=0.729 vs ensemble 0.557 -> pseudo-labels from the student | consistency x0.76 +[07:10:02][ 811s][FT] round 6/6 epoch 1/5 | patches=348 | loss=0.5120 +[07:10:09][ 817s][FT] round 6/6 epoch 2/5 | patches=348 | loss=0.4880 +[07:10:15][ 824s][FT] round 6/6 epoch 3/5 | patches=348 | loss=0.4572 +[07:10:21][ 830s][FT] round 6/6 epoch 4/5 | patches=348 | loss=0.4679 +[07:10:27][ 836s][FT] round 6/6 epoch 5/5 | patches=348 | loss=0.4644 +[07:10:29][ 838s][AL] round 6/6 | labeled=174 | selection_seconds=3.6000 round_seconds=145.5000 pseudo_student_dice=0.7290 pseudo_ensemble_dice=0.5568 consistency_scale=0.7630 val_dice=0.8730 val_vi=0.1522 val_ece=0.0088 +[07:10:32][ 841s][CONS] final consolidation: pseudo-labels refreshed from the student +[07:10:38][ 847s][FT] final 1/10 epoch 1/1 | patches=348 | loss=0.3857 +[07:10:45][ 853s][FT] final 2/10 epoch 1/1 | patches=348 | loss=0.3635 +[07:10:51][ 860s][FT] final 3/10 epoch 1/1 | patches=348 | loss=0.3861 +[07:10:57][ 866s][FT] final 4/10 epoch 1/1 | patches=348 | loss=0.3872 +[07:11:03][ 872s][FT] final 5/10 epoch 1/1 | patches=348 | loss=0.3739 +[07:11:10][ 879s][FT] final 6/10 epoch 1/1 | patches=348 | loss=0.3606 +[07:11:16][ 885s][FT] final 7/10 epoch 1/1 | patches=348 | loss=0.3436 +[07:11:23][ 891s][FT] final 8/10 epoch 1/1 | patches=348 | loss=0.3332 +[07:11:29][ 898s][FT] final 9/10 epoch 1/1 | patches=348 | loss=0.3277 +[07:11:35][ 904s][FT] final 10/10 epoch 1/1 | patches=348 | loss=0.3268 +[07:11:37][ 906s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=66.2000 val_dice=0.8848 val_vi=0.1468 val_ece=0.0087 +[07:11:37][ 906s][WEIGHTS] final: c-elegans=0.005, cerebellum=0.008, flywire=0.000, berghia=0.006, microns=0.008, oct-glia=0.972, snemi=0.000, mouse-brain=0.001 +[07:11:47][ 916s][TEST] raw student (no EMA, no TTA): patch_dice=0.8539 patch_vi=0.1791 +[07:11:49][ 918s][TEST] final model without TTA: patch_dice=0.8539 patch_vi=0.1791 +[07:11:49][ 918s][RESOURCES] {"phases": {"setup": {"seconds": 1.13, "peak_cpu_rss_mb": 1027.8, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 25.8, "peak_cpu_rss_mb": 1809.7, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 204.46, "peak_cpu_rss_mb": 2938.7, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 672.49, "peak_cpu_rss_mb": 2938.7, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4626.0, "rounds": 6}, "inference": {"seconds": 12.28, "peak_cpu_rss_mb": 2938.7, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4626.0, "patches": 768, "images": 6, "patches_per_second": 62.5, "seconds_per_image": 2.047}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 2938.7, "total_seconds": 916.2} +[07:11:49][ 918s][TEST] patch_dice=0.8592 patch_iou=0.7973 patch_precision=0.8997 patch_recall=0.8627 patch_vi=0.1753 image_dice=0.7942 image_iou=0.6669 image_precision=0.7997 image_recall=0.7956 image_vi=0.1990 cal_ece=0.0074 cal_nll=0.0770 cal_brier=0.0223 cal_mean_confidence=0.9774 cal_accuracy=0.9701 cal_predicted_prevalence=0.0725 cal_true_prevalence=0.0718 cal_mean_probability=0.0736 +[07:11:49][ 918s][DONE] octopus-vulgaris-vertical-lobe-sfltract macs-v11-norestart-acq-entropy b5 s2 | test patch Dice=0.8592 VI=0.1753 | image Dice=0.7942 VI=0.1990 | 918s diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/result.json b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..9674dfeabf3e916d5604d5c8f7e5b7f203217715 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/result.json @@ -0,0 +1,776 @@ +{ + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 2, + "config": { + "task": "membrane", + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "entropy", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "octopus-vulgaris-vertical-lobe-sfltract", + "--variants", + "macs-v11-norestart-acq-entropy", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "alphas": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "mixture": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "per_source": [ + 0.3196472209911767, + 0.016524281783989066, + 0.0007715704214095669, + 0.17539935780265475, + 0.02748152710843107, + 0.4363201963447303, + 0.016643455180579362, + 0.00721239036702913 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.00048684441851293454, + "mmd2_mixture": 0.01275567083190321, + "mmd2_uniform": 0.09698324592915836, + "mmd2_single_source": [ + 0.07244541101040552, + 0.256897395483041, + 0.3929088032810887, + 0.05287172109384486, + 0.26498470596000767, + 0.011934168264888134, + 0.339258942897097, + 0.2684433232010991 + ], + "median_squared_distance": 0.18344226468067681, + "gammas": [ + 1.3628266116054668, + 2.7256532232109336, + 5.451306446421867, + 10.902612892843734, + 21.80522578568747 + ], + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.0496200334528889, + 0.42663682749168497, + 0.8166058321365515, + 0.12600018979150718, + 0.3618977926472007, + 0.01001955344998784, + 0.4257222632020228, + 0.5321458020642489 + ], + "standardized": [ + -1.1548905481172125, + 0.3263029673133632, + 1.8583820949064265, + -0.8548142807860197, + 0.07196140979373515, + -1.3104697560349354, + 0.32270990006354566, + 0.7408182128610976 + ], + "tau": 0.5, + "max_weight": 0.4363201963447303, + "entropy": 1.3076016834873658, + "effective_sources": 3.6972957848927845, + "min_weight": 0.0007715704214095669 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.4757138324136084, + 0.6329416506363946, + 0.14191107641715917, + 0.4669227387129098, + 0.7079858630413108, + 0.8450575342195966, + 0.2290226697023551, + 0.11473197497037783 + ], + "ensemble_val_dice": { + "mixture": 0.8487996328814233, + "mixture@raw": 0.8468783370062004, + "mixture@logit": 0.8498165961112344, + "per_source": 0.8430704623597217, + "per_source@raw": 0.6192918885999461, + "per_source@logit": 0.8479512470904803, + "uniform": 0.7183095932451996, + "uniform@raw": 0.6771602818911759, + "uniform@logit": 0.8065076510127938 + }, + "val_patches": 768 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 3.5461363792419434, + 4.075904369354248 + ], + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "weight_stats": { + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.5952380952380953, + "spearman_weight_taxonomy": 0.6257415691925263, + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "top_source_is_best_teacher": true + }, + "init_teacher": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "student_init": "top-source", + "blank_fraction": 0.34765625, + "kd_seconds": 204.5, + "round0_val": { + "val_dice": 0.844188048800421, + "val_vi": 0.14634638944976133, + "val_ece": 0.06461472771169184 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 29, + "selection_seconds": 26.9, + "round_seconds": 129.3, + "teacher_dice_labeled": [ + 0.2571777942968238, + 0.5690605504275248, + 0.1686588604013407, + 0.2758037194420113, + 0.5811237001396865, + 0.7648114620563284, + 0.10985703762520847, + 0.20342659662893037 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 179039, + "evidence_nonempty_patches": 21, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.7483403957093632, + "restart_teacher_dice": 0.7648114620563284, + "weights": [ + 0.0009536063313568787, + 0.00622160049410085, + 2.7365280978604556e-05, + 0.0017150542671718894, + 0.00778389682673956, + 0.9830816831898324, + 4.206174632514249e-05, + 0.0001747318634946766 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.7646874735949377, + "uniform": 0.7405359432386459, + "top1": 0.7648114620563284, + "top2": 0.7643850371191413, + "top3": 0.7646318754401064 + }, + "val_dice": 0.850588566843053, + "val_vi": 0.15020738136525003, + "val_ece": 0.019654480413019926 + }, + { + "round": 2, + "n_labeled": 58, + "selection_seconds": 3.7, + "round_seconds": 23.9, + "teacher_dice_labeled": [ + 0.3051898457957521, + 0.43829723274180427, + 0.1662645122120252, + 0.2748940361871823, + 0.4567545091150516, + 0.5777898396180228, + 0.18661986847273826, + 0.28953794592991694 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 666295, + "evidence_nonempty_patches": 49, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.5780045228581705, + "uniform": 0.5900107461950028, + "top1": 0.5777898396180228, + "top2": 0.5769733017351059, + "top3": 0.5772074201516861 + }, + "pseudo_student_dice": 0.5266264913726709, + "pseudo_ensemble_dice": 0.5198939717868823, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8430299891923164, + "val_vi": 0.17334322904515395, + "val_ece": 0.011880019773361832 + }, + { + "round": 3, + "n_labeled": 87, + "selection_seconds": 3.9, + "round_seconds": 26.9, + "teacher_dice_labeled": [ + 0.256554203215484, + 0.3584625882888147, + 0.14446923270617784, + 0.21940902140115776, + 0.3404308510983355, + 0.423359355028126, + 0.24896099647764894, + 0.28709608031045697 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1340584, + "evidence_nonempty_patches": 77, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.42523293426148684, + "uniform": 0.4468307673264077, + "top1": 0.423359355028126, + "top2": 0.42387354326914495, + "top3": 0.42262162899245326 + }, + "pseudo_student_dice": 0.48432161207532876, + "pseudo_ensemble_dice": 0.2509745273916862, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.7727355398927459, + "val_vi": 0.17261770973435211, + "val_ece": 0.019825972598603598 + }, + { + "round": 4, + "n_labeled": 116, + "selection_seconds": 3.8, + "round_seconds": 131.4, + "teacher_dice_labeled": [ + 0.2936284700962486, + 0.4555138797855507, + 0.1916935869025312, + 0.28483595860735256, + 0.4645854181906192, + 0.6302061070335462, + 0.22524482639783452, + 0.24801376422596666 + ], + "teacher_dice_pool": [ + 0.2856535313876258, + 0.5207219061423793, + 0.2336147684929623, + 0.3149336309940731, + 0.5542851398624332, + 0.7928285164628092, + 0.21995953296421208, + 0.1900572617220739 + ], + "evidence_fg_pixels": 1701624, + "evidence_nonempty_patches": 105, + "first_correction": false, + "restart": "", + "weights": [ + 0.005201196091910545, + 0.007571920116834874, + 0.000130517662110731, + 0.007111074208082487, + 0.009194086198533695, + 0.9692522174719312, + 0.0005051739265953853, + 0.0010338143240010826 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6304836448276889, + "uniform": 0.5866893800713544, + "top1": 0.6302061070335462, + "top2": 0.6301844454387036, + "top3": 0.6303030425059223 + }, + "pseudo_student_dice": 0.6741909994732245, + "pseudo_ensemble_dice": 0.6069983324752639, + "pseudo_source": "student", + "consistency_scale": 0.581, + "val_dice": 0.8447836167150625, + "val_vi": 0.17608145575249626, + "val_ece": 0.007456402815686945 + }, + { + "round": 5, + "n_labeled": 145, + "selection_seconds": 3.8, + "round_seconds": 138.5, + "teacher_dice_labeled": [ + 0.3525403253235777, + 0.5143178603488624, + 0.19521558185516774, + 0.32040272699958583, + 0.4937034634276607, + 0.6562794950132421, + 0.26321708956655626, + 0.28986782022526414 + ], + "teacher_dice_pool": [ + 0.4086587040387924, + 0.6404567254485388, + 0.2383637366387085, + 0.38032477471780685, + 0.6049985309865628, + 0.8196890401954791, + 0.3140789070656653, + 0.28254675899248766 + ], + "evidence_fg_pixels": 1973658, + "evidence_nonempty_patches": 133, + "first_correction": false, + "restart": "", + "weights": [ + 0.007178796284775949, + 0.010439649829683862, + 0.00010352913395009788, + 0.007771362881353382, + 0.009420286261216514, + 0.9633177306284751, + 0.0005655224320038362, + 0.001203122548541244 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6567929858161429, + "uniform": 0.656789974783481, + "top1": 0.6562794950132421, + "top2": 0.6567521716447765, + "top3": 0.6561504363680574 + }, + "pseudo_student_dice": 0.6890727445515457, + "pseudo_ensemble_dice": 0.6309668156812759, + "pseudo_source": "student", + "consistency_scale": 0.63, + "val_dice": 0.5182098282101325, + "val_vi": 0.15514475495746, + "val_ece": 0.006062149701860885 + }, + { + "round": 6, + "n_labeled": 174, + "selection_seconds": 3.6, + "round_seconds": 145.5, + "teacher_dice_labeled": [ + 0.337566897659429, + 0.5141171511256178, + 0.19790145540894027, + 0.31851610101400474, + 0.5005018480617079, + 0.6817950271148931, + 0.24806949112657262, + 0.27509876690905666 + ], + "teacher_dice_pool": [ + 0.3757462025226483, + 0.6248672855702827, + 0.2432417622493409, + 0.3677013462721566, + 0.6094810118682077, + 0.8612422762866326, + 0.28957574781839723, + 0.2533942182588554 + ], + "evidence_fg_pixels": 2295762, + "evidence_nonempty_patches": 158, + "first_correction": false, + "restart": "", + "weights": [ + 0.004831314409281969, + 0.008144355165130322, + 8.313226695241612e-05, + 0.00596137055309223, + 0.007881911449567177, + 0.9719066277004534, + 0.00037993306747879105, + 0.0008113553880436958 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6821446395117668, + "uniform": 0.6616735807081089, + "top1": 0.6817950271148931, + "top2": 0.6820726476890172, + "top3": 0.6817917541113985 + }, + "pseudo_student_dice": 0.7289755472355686, + "pseudo_ensemble_dice": 0.5567674137146332, + "pseudo_source": "student", + "consistency_scale": 0.763, + "val_dice": 0.872970073630813, + "val_vi": 0.1522411484043612, + "val_ece": 0.008795662186711174 + }, + { + "round": 7, + "n_labeled": 174, + "selection_seconds": 0.0, + "round_seconds": 66.2, + "final_consolidation": true, + "val_dice": 0.8847641017050304, + "val_vi": 0.14680541323636923, + "val_ece": 0.008719392809047056 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.4491343304947581, + "auroc_top20_error": 0.6088995155171626 + }, + "al_seconds": 672.5, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.004831314409281969, + 0.008144355165130322, + 8.313226695241612e-05, + 0.00596137055309223, + 0.007881911449567177, + 0.9719066277004534, + 0.00037993306747879105, + 0.0008113553880436958 + ], + "weight_stats_final": { + "max_weight": 0.9719066277004534, + "entropy": 0.17089436807831188, + "effective_sources": 1.1863654243430872, + "min_weight": 8.313226695241612e-05 + }, + "top_source_final": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8539403069986387, + "patch_iou": 0.7897468209479852, + "patch_precision": 0.8916481287479515, + "patch_recall": 0.8602379909823856, + "patch_vi": 0.17912111520513588, + "n_patches": 768, + "cal_ece": 0.010201323030848362, + "cal_nll": 0.08445373201829658, + "cal_brier": 0.023090665519235807, + "cal_mean_confidence": 0.9794432489222348, + "cal_accuracy": 0.9692433675130209, + "cal_predicted_prevalence": 0.07187652587890625, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.0721161566746557 + }, + "test_notta": { + "patch_dice": 0.8539403069986387, + "patch_iou": 0.7897468209479852, + "patch_precision": 0.8916481287479515, + "patch_recall": 0.8602379909823856, + "patch_vi": 0.17912111520513588, + "n_patches": 768, + "cal_ece": 0.010201323030848362, + "cal_nll": 0.08445373201829658, + "cal_brier": 0.023090665519235807, + "cal_mean_confidence": 0.9794432489222348, + "cal_accuracy": 0.9692433675130209, + "cal_predicted_prevalence": 0.07187652587890625, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.0721161566746557 + }, + "test": { + "patch_dice": 0.8591772266029952, + "patch_iou": 0.7973100976285888, + "patch_precision": 0.8997360376998985, + "patch_recall": 0.8626797510061494, + "patch_vi": 0.17529183800507697, + "n_patches": 768, + "image_dice": 0.794234595546598, + "image_iou": 0.6669425716713131, + "image_precision": 0.7997486828309645, + "image_recall": 0.7956493079051613, + "image_vi": 0.19896338686997997, + "n_images": 6, + "cal_ece": 0.0073982110200828165, + "cal_nll": 0.07701464436765242, + "cal_brier": 0.022278359769095674, + "cal_mean_confidence": 0.9774351348043228, + "cal_accuracy": 0.9700902303059896, + "cal_predicted_prevalence": 0.07253519694010417, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.07355339704691839 + }, + "eval_seconds": 12.3, + "total_seconds": 917.8, + "resources": { + "phases": { + "setup": { + "seconds": 1.13, + "peak_cpu_rss_mb": 1027.8, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 25.8, + "peak_cpu_rss_mb": 1809.7, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 204.46, + "peak_cpu_rss_mb": 2938.7, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4618.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 672.49, + "peak_cpu_rss_mb": 2938.7, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4626.0, + "rounds": 6 + }, + "inference": { + "seconds": 12.28, + "peak_cpu_rss_mb": 2938.7, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4626.0, + "patches": 768, + "images": 6, + "patches_per_second": 62.5, + "seconds_per_image": 2.047 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 2938.7, + "total_seconds": 916.2 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..1a87a46a0882d9207f0693500f29274ba3c0030f --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,29,26.9,129.3,"[0.2571777942968238, 0.5690605504275248, 0.1686588604013407, 0.2758037194420113, 0.5811237001396865, 0.7648114620563284, 0.10985703762520847, 0.20342659662893037]",,179039,21,True,kept,0.7483403957093632,0.7648114620563284,"[0.0009536063313568787, 0.00622160049410085, 2.7365280978604556e-05, 0.0017150542671718894, 0.00778389682673956, 0.9830816831898324, 4.206174632514249e-05, 0.0001747318634946766]",mixture,"{'mixture': 0.7646874735949377, 'uniform': 0.7405359432386459, 'top1': 0.7648114620563284, 'top2': 0.7643850371191413, 'top3': 0.7646318754401064}",0.850588566843053,0.15020738136525003,0.019654480413019926,,,,, +2,58,3.7,23.9,"[0.3051898457957521, 0.43829723274180427, 0.1662645122120252, 0.2748940361871823, 0.4567545091150516, 0.5777898396180228, 0.18661986847273826, 0.28953794592991694]",,666295,49,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.5780045228581705, 'uniform': 0.5900107461950028, 'top1': 0.5777898396180228, 'top2': 0.5769733017351059, 'top3': 0.5772074201516861}",0.8430299891923164,0.17334322904515395,0.011880019773361832,0.5266264913726709,0.5198939717868823,ensemble,1.0, +3,87,3.9,26.9,"[0.256554203215484, 0.3584625882888147, 0.14446923270617784, 0.21940902140115776, 0.3404308510983355, 0.423359355028126, 0.24896099647764894, 0.28709608031045697]",,1340584,77,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.42523293426148684, 'uniform': 0.4468307673264077, 'top1': 0.423359355028126, 'top2': 0.42387354326914495, 'top3': 0.42262162899245326}",0.7727355398927459,0.17261770973435211,0.019825972598603598,0.48432161207532876,0.2509745273916862,student,0.0, +4,116,3.8,131.4,"[0.2936284700962486, 0.4555138797855507, 0.1916935869025312, 0.28483595860735256, 0.4645854181906192, 0.6302061070335462, 0.22524482639783452, 0.24801376422596666]","[0.2856535313876258, 0.5207219061423793, 0.2336147684929623, 0.3149336309940731, 0.5542851398624332, 0.7928285164628092, 0.21995953296421208, 0.1900572617220739]",1701624,105,False,,,,"[0.005201196091910545, 0.007571920116834874, 0.000130517662110731, 0.007111074208082487, 0.009194086198533695, 0.9692522174719312, 0.0005051739265953853, 0.0010338143240010826]",mixture,"{'mixture': 0.6304836448276889, 'uniform': 0.5866893800713544, 'top1': 0.6302061070335462, 'top2': 0.6301844454387036, 'top3': 0.6303030425059223}",0.8447836167150625,0.17608145575249626,0.007456402815686945,0.6741909994732245,0.6069983324752639,student,0.581, +5,145,3.8,138.5,"[0.3525403253235777, 0.5143178603488624, 0.19521558185516774, 0.32040272699958583, 0.4937034634276607, 0.6562794950132421, 0.26321708956655626, 0.28986782022526414]","[0.4086587040387924, 0.6404567254485388, 0.2383637366387085, 0.38032477471780685, 0.6049985309865628, 0.8196890401954791, 0.3140789070656653, 0.28254675899248766]",1973658,133,False,,,,"[0.007178796284775949, 0.010439649829683862, 0.00010352913395009788, 0.007771362881353382, 0.009420286261216514, 0.9633177306284751, 0.0005655224320038362, 0.001203122548541244]",mixture,"{'mixture': 0.6567929858161429, 'uniform': 0.656789974783481, 'top1': 0.6562794950132421, 'top2': 0.6567521716447765, 'top3': 0.6561504363680574}",0.5182098282101325,0.15514475495746,0.006062149701860885,0.6890727445515457,0.6309668156812759,student,0.63, +6,174,3.6,145.5,"[0.337566897659429, 0.5141171511256178, 0.19790145540894027, 0.31851610101400474, 0.5005018480617079, 0.6817950271148931, 0.24806949112657262, 0.27509876690905666]","[0.3757462025226483, 0.6248672855702827, 0.2432417622493409, 0.3677013462721566, 0.6094810118682077, 0.8612422762866326, 0.28957574781839723, 0.2533942182588554]",2295762,158,False,,,,"[0.004831314409281969, 0.008144355165130322, 8.313226695241612e-05, 0.00596137055309223, 0.007881911449567177, 0.9719066277004534, 0.00037993306747879105, 0.0008113553880436958]",mixture,"{'mixture': 0.6821446395117668, 'uniform': 0.6616735807081089, 'top1': 0.6817950271148931, 'top2': 0.6820726476890172, 'top3': 0.6817917541113985}",0.872970073630813,0.1522411484043612,0.008795662186711174,0.7289755472355686,0.5567674137146332,student,0.763, +7,174,0.0,66.2,,,,,,,,,,,,0.8847641017050304,0.14680541323636923,0.008719392809047056,,,,,True diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..53d0600bc224113f3ce3a8a1f0f1387de9e296a5 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt @@ -0,0 +1,107 @@ +[05:42:29][ 0s][START] target=octopus-vulgaris-vertical-lobe-sfltract variant=macs-v11-norestart-acq-laplace budget=5.0% seed=2 | {"task": "membrane", "target": "octopus-vulgaris-vertical-lobe-sfltract", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "laplace", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[05:42:29][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[05:42:29][ 0s][DATA] octopus-vulgaris-vertical-lobe-sfltract: train=3584 val=768 test=768 patches +[05:42:30][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, snemi=3.55, mouse-brain=4.08 +[05:42:30][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[05:42:36][ 7s][WEIGHTS] mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 | eff.sources=3.41 | MMD2 mix=0.0128 uniform=0.0970 +[05:42:36][ 7s][WEIGHTS] per-source: c-elegans=0.320, cerebellum=0.017, flywire=0.001, berghia=0.175, microns=0.027, oct-glia=0.436, snemi=0.017, mouse-brain=0.007 | min-MMD source (NeuroADDA rule) = oct-glia +[05:42:55][ 26s][TEACHERS] target-val Dice: c-elegans=0.476, cerebellum=0.633, flywire=0.142, berghia=0.467, microns=0.708, oct-glia=0.845, snemi=0.229, mouse-brain=0.115 +[05:42:55][ 26s][TEACHERS] ensemble target-val Dice: mixture=0.849, mixture@raw=0.847, mixture@logit=0.850, per_source=0.843, per_source@raw=0.619, per_source@logit=0.848, uniform=0.718, uniform@raw=0.677, uniform@logit=0.807 +[05:42:55][ 26s][WEIGHTS] using mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 +[05:42:55][ 26s][INIT] student initialised from teacher oct-glia (weight 0.636) +[05:42:56][ 27s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[05:42:56][ 27s][KD] distilling the mixture ensemble (prob space) into the student +[05:43:37][ 68s][KD] soft labels for 3584 patches from 8 teachers in 41s +[05:44:09][ 100s][KD] epoch 1/5 | patches=3584 | loss=0.5967 | 32s +[05:44:40][ 131s][KD] epoch 2/5 | patches=3584 | loss=0.4536 | 31s +[05:45:11][ 162s][KD] epoch 3/5 | patches=3584 | loss=0.4007 | 31s +[05:45:42][ 193s][KD] epoch 4/5 | patches=3584 | loss=0.3637 | 31s +[05:46:14][ 225s][KD] epoch 5/5 | patches=3584 | loss=0.3392 | 32s +[05:46:15][ 227s][ROUND0] before any target label: val_dice=0.8527 val_vi=0.1456 +[05:46:15][ 227s][AL] patches=3584 (pool 2338 after dropping blank; free blank negatives: True) | budget=179 (5.0%) | rounds=6 | per_round=29 | acquisition=laplace | first_round=stratified +[05:46:23][ 235s][AL] round-1 uncertainty quality on 2338 patches: {'spearman_uncertainty_error': 0.4468831233044695, 'auroc_top20_error': 0.618147538735774} +[05:46:46][ 258s][AL] round 1: stratified cold start: 15 patches from the membrane-rich half (ensemble prevalence >= 0.192) + 14 from the membrane-poor half of 2000 candidates +[05:46:49][ 261s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.765, uniform=0.741, top1=0.765, top2=0.764, top3=0.765 -> using mixture +[05:46:50][ 261s][WEIGHTS] round 1: student kept (labeled Dice 0.760 vs best teacher oct-glia 0.765); distilling the corrected ensemble into it +[05:46:50][ 261s][WEIGHTS] round 1: teacher Dice on 29 labeled patches + 29 blank (179039 fg px): c-elegans=0.257, cerebellum=0.569, flywire=0.169, berghia=0.276, microns=0.581, oct-glia=0.765, snemi=0.110, mouse-brain=0.203 | weights: c-elegans=0.001, cerebellum=0.006, flywire=0.000, berghia=0.002, microns=0.008, oct-glia=0.983, snemi=0.000, mouse-brain=0.000 +[05:47:16][ 288s][KD] soft labels for 2309 patches from 8 teachers in 26s +[05:47:37][ 308s][KD] epoch 1/2 | patches=2309 | loss=0.1388 | 21s +[05:47:58][ 329s][KD] epoch 2/2 | patches=2309 | loss=0.1333 | 21s +[05:48:21][ 352s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[05:48:22][ 354s][FT] round 1/6 epoch 1/5 | patches=58 | loss=0.8715 +[05:48:24][ 355s][FT] round 1/6 epoch 2/5 | patches=58 | loss=0.5653 +[05:48:25][ 357s][FT] round 1/6 epoch 3/5 | patches=58 | loss=0.5161 +[05:48:27][ 358s][FT] round 1/6 epoch 4/5 | patches=58 | loss=0.5170 +[05:48:28][ 360s][FT] round 1/6 epoch 5/5 | patches=58 | loss=0.4795 +[05:48:30][ 361s][AL] round 1/6 | labeled=29 | selection_seconds=30.9000 round_seconds=132.8000 restart_student_dice=0.7603 restart_teacher_dice=0.7648 val_dice=0.8582 val_vi=0.1520 val_ece=0.0197 +[05:48:37][ 368s][AL] round 2: 590/2309 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:48:43][ 374s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.598, uniform=0.658, top1=0.599, top2=0.597, top3=0.597 -> using uniform +[05:48:43][ 374s][WEIGHTS] round 2: teacher Dice on 58 labeled patches + 58 blank (638880 fg px): c-elegans=0.359, cerebellum=0.528, flywire=0.175, berghia=0.310, microns=0.562, oct-glia=0.599, snemi=0.177, mouse-brain=0.325 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[05:48:49][ 380s][CONS] round 2: on the 29 new patches student Dice=0.661 vs ensemble 0.616 -> pseudo-labels from the student | consistency x0.54 +[05:48:51][ 383s][FT] round 2/6 epoch 1/5 | patches=116 | loss=0.5240 +[05:48:54][ 385s][FT] round 2/6 epoch 2/5 | patches=116 | loss=0.5249 +[05:48:56][ 387s][FT] round 2/6 epoch 3/5 | patches=116 | loss=0.4267 +[05:48:58][ 390s][FT] round 2/6 epoch 4/5 | patches=116 | loss=0.4202 +[05:49:01][ 392s][FT] round 2/6 epoch 5/5 | patches=116 | loss=0.4486 +[05:49:02][ 393s][AL] round 2/6 | labeled=58 | selection_seconds=7.5000 round_seconds=30.8000 pseudo_student_dice=0.6608 pseudo_ensemble_dice=0.6160 consistency_scale=0.5360 val_dice=0.8519 val_vi=0.1695 val_ece=0.0174 +[05:49:09][ 401s][AL] round 3: 567/2280 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:50:00][ 451s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.433, cerebellum=0.661, flywire=0.229, berghia=0.388, microns=0.577, oct-glia=0.797, snemi=0.336, mouse-brain=0.309 | blended with labeled Dice at 0.50 +[05:50:47][ 498s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.708, uniform=0.708, top1=0.708, top2=0.709, top3=0.708 -> using mixture +[05:50:47][ 498s][WEIGHTS] round 3: teacher Dice on 87 labeled patches + 87 blank (816708 fg px): c-elegans=0.355, cerebellum=0.544, flywire=0.200, berghia=0.336, microns=0.539, oct-glia=0.708, snemi=0.247, mouse-brain=0.285 | weights: c-elegans=0.004, cerebellum=0.008, flywire=0.000, berghia=0.005, microns=0.009, oct-glia=0.972, snemi=0.000, mouse-brain=0.001 +[05:50:50][ 501s][CONS] round 3: on the 29 new patches student Dice=0.542 vs ensemble 0.682 -> pseudo-labels from the ensemble | consistency x1.00 +[05:50:53][ 505s][FT] round 3/6 epoch 1/5 | patches=174 | loss=0.5448 +[05:50:57][ 508s][FT] round 3/6 epoch 2/5 | patches=174 | loss=0.5017 +[05:51:00][ 511s][FT] round 3/6 epoch 3/5 | patches=174 | loss=0.5088 +[05:51:04][ 515s][FT] round 3/6 epoch 4/5 | patches=174 | loss=0.4270 +[05:51:07][ 518s][FT] round 3/6 epoch 5/5 | patches=174 | loss=0.4082 +[05:51:08][ 520s][AL] round 3/6 | labeled=87 | selection_seconds=7.6000 round_seconds=124.8000 pseudo_student_dice=0.5422 pseudo_ensemble_dice=0.6818 consistency_scale=1.0000 val_dice=0.8693 val_vi=0.1472 val_ece=0.0117 +[05:51:16][ 527s][AL] round 4: 584/2251 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:51:26][ 538s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.472, uniform=0.491, top1=0.471, top2=0.470, top3=0.470 -> using uniform +[05:51:26][ 538s][WEIGHTS] round 4: teacher Dice on 116 labeled patches + 116 blank (1509201 fg px): c-elegans=0.281, cerebellum=0.374, flywire=0.150, berghia=0.245, microns=0.402, oct-glia=0.471, snemi=0.209, mouse-brain=0.290 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[05:51:32][ 544s][CONS] round 4: on the 29 new patches student Dice=0.526 vs ensemble 0.265 -> pseudo-labels from the student | consistency x0.09 +[05:51:36][ 548s][FT] round 4/6 epoch 1/5 | patches=232 | loss=0.6832 +[05:51:41][ 552s][FT] round 4/6 epoch 2/5 | patches=232 | loss=0.9581 +[05:51:45][ 556s][FT] round 4/6 epoch 3/5 | patches=232 | loss=0.5798 +[05:51:49][ 561s][FT] round 4/6 epoch 4/5 | patches=232 | loss=0.3980 +[05:51:54][ 565s][FT] round 4/6 epoch 5/5 | patches=232 | loss=0.4769 +[05:51:55][ 567s][AL] round 4/6 | labeled=116 | selection_seconds=7.6000 round_seconds=45.4000 pseudo_student_dice=0.5262 pseudo_ensemble_dice=0.2655 consistency_scale=0.0870 val_dice=0.8838 val_vi=0.1529 val_ece=0.0111 +[05:52:03][ 574s][AL] round 5: 569/2222 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:52:55][ 627s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.420, cerebellum=0.661, flywire=0.239, berghia=0.388, microns=0.594, oct-glia=0.833, snemi=0.317, mouse-brain=0.291 | blended with labeled Dice at 0.50 +[05:53:45][ 676s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.657, uniform=0.667, top1=0.656, top2=0.657, top3=0.656 -> using uniform +[05:53:45][ 676s][WEIGHTS] round 5: teacher Dice on 145 labeled patches + 145 blank (1935317 fg px): c-elegans=0.358, cerebellum=0.529, flywire=0.197, berghia=0.324, microns=0.497, oct-glia=0.656, snemi=0.265, mouse-brain=0.294 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[05:53:51][ 682s][CONS] round 5: on the 29 new patches student Dice=0.723 vs ensemble 0.558 -> pseudo-labels from the student | consistency x0.74 +[05:53:56][ 687s][FT] round 5/6 epoch 1/5 | patches=290 | loss=0.5566 +[05:54:01][ 693s][FT] round 5/6 epoch 2/5 | patches=290 | loss=0.5073 +[05:54:07][ 698s][FT] round 5/6 epoch 3/5 | patches=290 | loss=0.4316 +[05:54:12][ 703s][FT] round 5/6 epoch 4/5 | patches=290 | loss=0.4643 +[05:54:17][ 709s][FT] round 5/6 epoch 5/5 | patches=290 | loss=0.6065 +[05:54:19][ 710s][AL] round 5/6 | labeled=145 | selection_seconds=7.4000 round_seconds=141.9000 pseudo_student_dice=0.7225 pseudo_ensemble_dice=0.5582 consistency_scale=0.7420 val_dice=0.8797 val_vi=0.1583 val_ece=0.0099 +[05:54:26][ 717s][AL] round 6: 561/2193 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:55:20][ 772s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.419, cerebellum=0.654, flywire=0.238, berghia=0.389, microns=0.593, oct-glia=0.826, snemi=0.320, mouse-brain=0.293 | blended with labeled Dice at 0.50 +[05:56:11][ 823s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.665, uniform=0.674, top1=0.664, top2=0.665, top3=0.664 -> using uniform +[05:56:11][ 823s][WEIGHTS] round 6: teacher Dice on 174 labeled patches + 174 blank (2258511 fg px): c-elegans=0.360, cerebellum=0.536, flywire=0.199, berghia=0.334, microns=0.501, oct-glia=0.664, snemi=0.265, mouse-brain=0.294 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[05:56:17][ 828s][CONS] round 6: on the 29 new patches student Dice=0.767 vs ensemble 0.632 -> pseudo-labels from the student | consistency x0.89 +[05:56:23][ 835s][FT] round 6/6 epoch 1/5 | patches=348 | loss=0.5293 +[05:56:29][ 841s][FT] round 6/6 epoch 2/5 | patches=348 | loss=0.4097 +[05:56:36][ 847s][FT] round 6/6 epoch 3/5 | patches=348 | loss=0.3725 +[05:56:42][ 853s][FT] round 6/6 epoch 4/5 | patches=348 | loss=0.3776 +[05:56:48][ 860s][FT] round 6/6 epoch 5/5 | patches=348 | loss=0.3708 +[05:56:50][ 861s][AL] round 6/6 | labeled=174 | selection_seconds=7.5000 round_seconds=149.7000 pseudo_student_dice=0.7673 pseudo_ensemble_dice=0.6316 consistency_scale=0.8910 val_dice=0.8887 val_vi=0.1513 val_ece=0.0094 +[05:56:53][ 864s][CONS] final consolidation: pseudo-labels refreshed from the student +[05:57:00][ 871s][FT] final 1/10 epoch 1/1 | patches=348 | loss=0.3354 +[05:57:06][ 877s][FT] final 2/10 epoch 1/1 | patches=348 | loss=0.3671 +[05:57:12][ 884s][FT] final 3/10 epoch 1/1 | patches=348 | loss=0.4036 +[05:57:18][ 890s][FT] final 4/10 epoch 1/1 | patches=348 | loss=0.4250 +[05:57:25][ 896s][FT] final 5/10 epoch 1/1 | patches=348 | loss=0.3658 +[05:57:31][ 903s][FT] final 6/10 epoch 1/1 | patches=348 | loss=0.3438 +[05:57:37][ 909s][FT] final 7/10 epoch 1/1 | patches=348 | loss=0.3156 +[05:57:44][ 915s][FT] final 8/10 epoch 1/1 | patches=348 | loss=0.3556 +[05:57:50][ 921s][FT] final 9/10 epoch 1/1 | patches=348 | loss=0.3240 +[05:57:56][ 928s][FT] final 10/10 epoch 1/1 | patches=348 | loss=0.3354 +[05:57:58][ 929s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=66.3000 val_dice=0.8887 val_vi=0.1468 val_ece=0.0108 +[05:57:58][ 929s][WEIGHTS] final: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[05:58:08][ 940s][TEST] raw student (no EMA, no TTA): patch_dice=0.8648 patch_vi=0.1775 +[05:58:10][ 941s][TEST] final model without TTA: patch_dice=0.8648 patch_vi=0.1775 +[05:58:10][ 941s][RESOURCES] {"phases": {"setup": {"seconds": 1.14, "peak_cpu_rss_mb": 1027.2, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 25.95, "peak_cpu_rss_mb": 1808.4, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 197.94, "peak_cpu_rss_mb": 2938.2, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 702.31, "peak_cpu_rss_mb": 2938.2, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4626.0, "rounds": 6}, "inference": {"seconds": 12.0, "peak_cpu_rss_mb": 2938.2, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4626.0, "patches": 768, "images": 6, "patches_per_second": 64.0, "seconds_per_image": 2.0}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 2938.2, "total_seconds": 939.3} +[05:58:10][ 941s][TEST] patch_dice=0.8647 patch_iou=0.8033 patch_precision=0.9035 patch_recall=0.8667 patch_vi=0.1753 image_dice=0.7921 image_iou=0.6652 image_precision=0.7986 image_recall=0.7938 image_vi=0.1992 cal_ece=0.0066 cal_nll=0.0771 cal_brier=0.0219 cal_mean_confidence=0.9765 cal_accuracy=0.9701 cal_predicted_prevalence=0.0729 cal_true_prevalence=0.0718 cal_mean_probability=0.0727 +[05:58:10][ 941s][DONE] octopus-vulgaris-vertical-lobe-sfltract macs-v11-norestart-acq-laplace b5 s2 | test patch Dice=0.8647 VI=0.1753 | image Dice=0.7921 VI=0.1992 | 941s diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..d319b5dedec18d32f61a1187a06670a77ed40178 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json @@ -0,0 +1,776 @@ +{ + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 2, + "config": { + "task": "membrane", + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "laplace", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "octopus-vulgaris-vertical-lobe-sfltract", + "--variants", + "macs-v11-norestart-acq-laplace", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "alphas": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "mixture": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "per_source": [ + 0.3196472209911767, + 0.016524281783989066, + 0.0007715704214095669, + 0.17539935780265475, + 0.02748152710843107, + 0.4363201963447303, + 0.016643455180579362, + 0.00721239036702913 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.00048684441851293454, + "mmd2_mixture": 0.01275567083190321, + "mmd2_uniform": 0.09698324592915836, + "mmd2_single_source": [ + 0.07244541101040552, + 0.256897395483041, + 0.3929088032810887, + 0.05287172109384486, + 0.26498470596000767, + 0.011934168264888134, + 0.339258942897097, + 0.2684433232010991 + ], + "median_squared_distance": 0.18344226468067681, + "gammas": [ + 1.3628266116054668, + 2.7256532232109336, + 5.451306446421867, + 10.902612892843734, + 21.80522578568747 + ], + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.0496200334528889, + 0.42663682749168497, + 0.8166058321365515, + 0.12600018979150718, + 0.3618977926472007, + 0.01001955344998784, + 0.4257222632020228, + 0.5321458020642489 + ], + "standardized": [ + -1.1548905481172125, + 0.3263029673133632, + 1.8583820949064265, + -0.8548142807860197, + 0.07196140979373515, + -1.3104697560349354, + 0.32270990006354566, + 0.7408182128610976 + ], + "tau": 0.5, + "max_weight": 0.4363201963447303, + "entropy": 1.3076016834873658, + "effective_sources": 3.6972957848927845, + "min_weight": 0.0007715704214095669 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.4757138324136084, + 0.6329416506363946, + 0.14191107641715917, + 0.4669227387129098, + 0.7079858630413108, + 0.8450575342195966, + 0.2290226697023551, + 0.11473197497037783 + ], + "ensemble_val_dice": { + "mixture": 0.8487996328814233, + "mixture@raw": 0.8468783370062004, + "mixture@logit": 0.8498165961112344, + "per_source": 0.8430704623597217, + "per_source@raw": 0.6192918885999461, + "per_source@logit": 0.8479512470904803, + "uniform": 0.7183095932451996, + "uniform@raw": 0.6771602818911759, + "uniform@logit": 0.8065076510127938 + }, + "val_patches": 768 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 3.5461363792419434, + 4.075904369354248 + ], + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "weight_stats": { + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.5952380952380953, + "spearman_weight_taxonomy": 0.6257415691925263, + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "top_source_is_best_teacher": true + }, + "init_teacher": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "student_init": "top-source", + "blank_fraction": 0.34765625, + "kd_seconds": 197.9, + "round0_val": { + "val_dice": 0.8527240517674085, + "val_vi": 0.1455715366868596, + "val_ece": 0.06194851765646773 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 29, + "selection_seconds": 30.9, + "round_seconds": 132.8, + "teacher_dice_labeled": [ + 0.2571777942968238, + 0.5690605504275248, + 0.1686588604013407, + 0.2758037194420113, + 0.5811237001396865, + 0.7648114620563284, + 0.10985703762520847, + 0.20342659662893037 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 179039, + "evidence_nonempty_patches": 21, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.7603000274646353, + "restart_teacher_dice": 0.7648114620563284, + "weights": [ + 0.0009536063313568787, + 0.00622160049410085, + 2.7365280978604556e-05, + 0.0017150542671718894, + 0.00778389682673956, + 0.9830816831898324, + 4.206174632514249e-05, + 0.0001747318634946766 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.7646874735949377, + "uniform": 0.7405359432386459, + "top1": 0.7648114620563284, + "top2": 0.7643850371191413, + "top3": 0.7646318754401064 + }, + "val_dice": 0.8582032598395823, + "val_vi": 0.15202137350161224, + "val_ece": 0.01967065256522741 + }, + { + "round": 2, + "n_labeled": 58, + "selection_seconds": 7.5, + "round_seconds": 30.8, + "teacher_dice_labeled": [ + 0.3588600320416357, + 0.5283610665271623, + 0.1754836982934096, + 0.30965104253978754, + 0.5621811921846875, + 0.5986021775475412, + 0.17715468712343485, + 0.3248640901664514 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 638880, + "evidence_nonempty_patches": 50, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.5983930359391161, + "uniform": 0.6576421407682509, + "top1": 0.5986021775475412, + "top2": 0.5968083368993796, + "top3": 0.597320828329163 + }, + "pseudo_student_dice": 0.6608478637558942, + "pseudo_ensemble_dice": 0.6159762447621395, + "pseudo_source": "student", + "consistency_scale": 0.536, + "val_dice": 0.8519298021852384, + "val_vi": 0.1695486292888474, + "val_ece": 0.01740237649855266 + }, + { + "round": 3, + "n_labeled": 87, + "selection_seconds": 7.6, + "round_seconds": 124.8, + "teacher_dice_labeled": [ + 0.3550508758955952, + 0.5444899469348708, + 0.20038206997152114, + 0.33595390889582355, + 0.5390476896390397, + 0.7080628920438248, + 0.24712166474669486, + 0.28541810272502866 + ], + "teacher_dice_pool": [ + 0.4328659013976558, + 0.661135411122184, + 0.22903361986981352, + 0.38842096758112005, + 0.5765943393419145, + 0.797088399227804, + 0.33584992040025263, + 0.30945976554298166 + ], + "evidence_fg_pixels": 816708, + "evidence_nonempty_patches": 74, + "first_correction": false, + "restart": "", + "weights": [ + 0.00442401041120939, + 0.008483654377623146, + 6.551815847839877e-05, + 0.005456277398952017, + 0.008909491152141185, + 0.9716801215408971, + 0.00028934076331636185, + 0.0006915861973823141 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.708058559923522, + "uniform": 0.7080861303326307, + "top1": 0.7080628920438248, + "top2": 0.708505089247222, + "top3": 0.7076425236551863 + }, + "pseudo_student_dice": 0.5422034726585485, + "pseudo_ensemble_dice": 0.6817766491064452, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8692568562603569, + "val_vi": 0.14722995535203087, + "val_ece": 0.011732638054617456 + }, + { + "round": 4, + "n_labeled": 116, + "selection_seconds": 7.6, + "round_seconds": 45.4, + "teacher_dice_labeled": [ + 0.2812613113577766, + 0.37428606024354677, + 0.15015731686738024, + 0.24517187706258378, + 0.4022338376611055, + 0.471145297805343, + 0.2088359438902864, + 0.29038033021372894 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1509201, + "evidence_nonempty_patches": 103, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.47207579835642194, + "uniform": 0.4913551341147341, + "top1": 0.471145297805343, + "top2": 0.4699058743924868, + "top3": 0.4702756933037702 + }, + "pseudo_student_dice": 0.526227817416522, + "pseudo_ensemble_dice": 0.26545333677506955, + "pseudo_source": "student", + "consistency_scale": 0.087, + "val_dice": 0.8837533492723129, + "val_vi": 0.15290057656135012, + "val_ece": 0.011086664785801689 + }, + { + "round": 5, + "n_labeled": 145, + "selection_seconds": 7.4, + "round_seconds": 141.9, + "teacher_dice_labeled": [ + 0.3575857705768084, + 0.5288894927276441, + 0.1972882651699285, + 0.3237538784944006, + 0.4972437412072611, + 0.6563423781453516, + 0.2654844034258358, + 0.2937566814955266 + ], + "teacher_dice_pool": [ + 0.4195271845365769, + 0.6613145146970063, + 0.23861478496716995, + 0.38766042413489726, + 0.5937290521963867, + 0.8326240947868595, + 0.31651714330217995, + 0.29086484509438304 + ], + "evidence_fg_pixels": 1935317, + "evidence_nonempty_patches": 132, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.6568639291864343, + "uniform": 0.6666955000371759, + "top1": 0.6563423781453516, + "top2": 0.6570021039931723, + "top3": 0.6561298907989814 + }, + "pseudo_student_dice": 0.7225415124397063, + "pseudo_ensemble_dice": 0.5582015873469427, + "pseudo_source": "student", + "consistency_scale": 0.742, + "val_dice": 0.8797230617226582, + "val_vi": 0.15834655247909965, + "val_ece": 0.00989123741765351 + }, + { + "round": 6, + "n_labeled": 174, + "selection_seconds": 7.5, + "round_seconds": 149.7, + "teacher_dice_labeled": [ + 0.3602505543526682, + 0.5359821774533255, + 0.19879252836890904, + 0.333613065162099, + 0.5009071969940995, + 0.6643916205152369, + 0.26497846806752806, + 0.2937445620594825 + ], + "teacher_dice_pool": [ + 0.41908704552384624, + 0.6539098615025841, + 0.23820759536247726, + 0.38935192574692595, + 0.592662873689235, + 0.8257198302767631, + 0.32037599926452937, + 0.29283550908122574 + ], + "evidence_fg_pixels": 2258511, + "evidence_nonempty_patches": 161, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.6648869229132081, + "uniform": 0.6739570229701966, + "top1": 0.6643916205152369, + "top2": 0.6649912132591502, + "top3": 0.6642192650367976 + }, + "pseudo_student_dice": 0.7673293550156336, + "pseudo_ensemble_dice": 0.6316422578928547, + "pseudo_source": "student", + "consistency_scale": 0.891, + "val_dice": 0.8886952685341428, + "val_vi": 0.1513188957973902, + "val_ece": 0.0094336474487976 + }, + { + "round": 7, + "n_labeled": 174, + "selection_seconds": 0.0, + "round_seconds": 66.3, + "final_consolidation": true, + "val_dice": 0.8886864148471417, + "val_vi": 0.14682279863099465, + "val_ece": 0.010849301665396582 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.4468831233044695, + "auroc_top20_error": 0.618147538735774 + }, + "al_seconds": 702.3, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "weight_stats_final": { + "max_weight": 0.125, + "entropy": 2.0794415416798357, + "effective_sources": 7.999999999999998, + "min_weight": 0.125 + }, + "top_source_final": "c-elegans-dauer-stage", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8647887142358766, + "patch_iou": 0.8012740206376692, + "patch_precision": 0.8955808545122568, + "patch_recall": 0.8680106101740352, + "patch_vi": 0.17752917596284057, + "n_patches": 768, + "cal_ece": 0.008089954957642595, + "cal_nll": 0.08198473665119577, + "cal_brier": 0.0224464631914164, + "cal_mean_confidence": 0.9775250897061557, + "cal_accuracy": 0.9694976806640625, + "cal_predicted_prevalence": 0.07344818115234375, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.07302339563645663 + }, + "test_notta": { + "patch_dice": 0.8647887142358766, + "patch_iou": 0.8012740206376692, + "patch_precision": 0.8955808545122568, + "patch_recall": 0.8680106101740352, + "patch_vi": 0.17752917596284057, + "n_patches": 768, + "cal_ece": 0.008089954957642595, + "cal_nll": 0.08198473665119577, + "cal_brier": 0.0224464631914164, + "cal_mean_confidence": 0.9775250897061557, + "cal_accuracy": 0.9694976806640625, + "cal_predicted_prevalence": 0.07344818115234375, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.07302339563645663 + }, + "test": { + "patch_dice": 0.8647069438389737, + "patch_iou": 0.8033281169756871, + "patch_precision": 0.9034503196996484, + "patch_recall": 0.8667368064369821, + "patch_vi": 0.17527923318371696, + "n_patches": 768, + "image_dice": 0.7920637119427608, + "image_iou": 0.665237657237726, + "image_precision": 0.7986320096435179, + "image_recall": 0.7938427626550224, + "image_vi": 0.1992288512318123, + "n_images": 6, + "cal_ece": 0.006552162411524904, + "cal_nll": 0.07708213635633097, + "cal_brier": 0.021867434336899864, + "cal_mean_confidence": 0.9765031516520883, + "cal_accuracy": 0.9700775146484375, + "cal_predicted_prevalence": 0.07292429606119792, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.0726615540301751 + }, + "eval_seconds": 12.0, + "total_seconds": 941.1, + "resources": { + "phases": { + "setup": { + "seconds": 1.14, + "peak_cpu_rss_mb": 1027.2, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 25.95, + "peak_cpu_rss_mb": 1808.4, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 197.94, + "peak_cpu_rss_mb": 2938.2, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4618.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 702.31, + "peak_cpu_rss_mb": 2938.2, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4626.0, + "rounds": 6 + }, + "inference": { + "seconds": 12.0, + "peak_cpu_rss_mb": 2938.2, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4626.0, + "patches": 768, + "images": 6, + "patches_per_second": 64.0, + "seconds_per_image": 2.0 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 2938.2, + "total_seconds": 939.3 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..2c5e0be71ab629b8d60322032ecab275eac27a78 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,29,30.9,132.8,"[0.2571777942968238, 0.5690605504275248, 0.1686588604013407, 0.2758037194420113, 0.5811237001396865, 0.7648114620563284, 0.10985703762520847, 0.20342659662893037]",,179039,21,True,kept,0.7603000274646353,0.7648114620563284,"[0.0009536063313568787, 0.00622160049410085, 2.7365280978604556e-05, 0.0017150542671718894, 0.00778389682673956, 0.9830816831898324, 4.206174632514249e-05, 0.0001747318634946766]",mixture,"{'mixture': 0.7646874735949377, 'uniform': 0.7405359432386459, 'top1': 0.7648114620563284, 'top2': 0.7643850371191413, 'top3': 0.7646318754401064}",0.8582032598395823,0.15202137350161224,0.01967065256522741,,,,, +2,58,7.5,30.8,"[0.3588600320416357, 0.5283610665271623, 0.1754836982934096, 0.30965104253978754, 0.5621811921846875, 0.5986021775475412, 0.17715468712343485, 0.3248640901664514]",,638880,50,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.5983930359391161, 'uniform': 0.6576421407682509, 'top1': 0.5986021775475412, 'top2': 0.5968083368993796, 'top3': 0.597320828329163}",0.8519298021852384,0.1695486292888474,0.01740237649855266,0.6608478637558942,0.6159762447621395,student,0.536, +3,87,7.6,124.8,"[0.3550508758955952, 0.5444899469348708, 0.20038206997152114, 0.33595390889582355, 0.5390476896390397, 0.7080628920438248, 0.24712166474669486, 0.28541810272502866]","[0.4328659013976558, 0.661135411122184, 0.22903361986981352, 0.38842096758112005, 0.5765943393419145, 0.797088399227804, 0.33584992040025263, 0.30945976554298166]",816708,74,False,,,,"[0.00442401041120939, 0.008483654377623146, 6.551815847839877e-05, 0.005456277398952017, 0.008909491152141185, 0.9716801215408971, 0.00028934076331636185, 0.0006915861973823141]",mixture,"{'mixture': 0.708058559923522, 'uniform': 0.7080861303326307, 'top1': 0.7080628920438248, 'top2': 0.708505089247222, 'top3': 0.7076425236551863}",0.8692568562603569,0.14722995535203087,0.011732638054617456,0.5422034726585485,0.6817766491064452,ensemble,1.0, +4,116,7.6,45.4,"[0.2812613113577766, 0.37428606024354677, 0.15015731686738024, 0.24517187706258378, 0.4022338376611055, 0.471145297805343, 0.2088359438902864, 0.29038033021372894]",,1509201,103,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.47207579835642194, 'uniform': 0.4913551341147341, 'top1': 0.471145297805343, 'top2': 0.4699058743924868, 'top3': 0.4702756933037702}",0.8837533492723129,0.15290057656135012,0.011086664785801689,0.526227817416522,0.26545333677506955,student,0.087, +5,145,7.4,141.9,"[0.3575857705768084, 0.5288894927276441, 0.1972882651699285, 0.3237538784944006, 0.4972437412072611, 0.6563423781453516, 0.2654844034258358, 0.2937566814955266]","[0.4195271845365769, 0.6613145146970063, 0.23861478496716995, 0.38766042413489726, 0.5937290521963867, 0.8326240947868595, 0.31651714330217995, 0.29086484509438304]",1935317,132,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6568639291864343, 'uniform': 0.6666955000371759, 'top1': 0.6563423781453516, 'top2': 0.6570021039931723, 'top3': 0.6561298907989814}",0.8797230617226582,0.15834655247909965,0.00989123741765351,0.7225415124397063,0.5582015873469427,student,0.742, +6,174,7.5,149.7,"[0.3602505543526682, 0.5359821774533255, 0.19879252836890904, 0.333613065162099, 0.5009071969940995, 0.6643916205152369, 0.26497846806752806, 0.2937445620594825]","[0.41908704552384624, 0.6539098615025841, 0.23820759536247726, 0.38935192574692595, 0.592662873689235, 0.8257198302767631, 0.32037599926452937, 0.29283550908122574]",2258511,161,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6648869229132081, 'uniform': 0.6739570229701966, 'top1': 0.6643916205152369, 'top2': 0.6649912132591502, 'top3': 0.6642192650367976}",0.8886952685341428,0.1513188957973902,0.0094336474487976,0.7673293550156336,0.6316422578928547,student,0.891, +7,174,0.0,66.3,,,,,,,,,,,,0.8886864148471417,0.14682279863099465,0.010849301665396582,,,,,True diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..5458bd72229e39dbdb16b84abab014565230ed70 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt @@ -0,0 +1,107 @@ +[10:02:19][ 0s][START] target=octopus-vulgaris-vertical-lobe-sfltract variant=macs-v11-norestart-acq-laplace budget=5.0% seed=3 | {"task": "membrane", "target": "octopus-vulgaris-vertical-lobe-sfltract", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "laplace", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[10:02:19][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[10:02:19][ 0s][DATA] octopus-vulgaris-vertical-lobe-sfltract: train=3584 val=768 test=768 patches +[10:02:20][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, snemi=3.55, mouse-brain=4.08 +[10:02:20][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[10:02:26][ 7s][WEIGHTS] mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 | eff.sources=3.41 | MMD2 mix=0.0128 uniform=0.0970 +[10:02:26][ 7s][WEIGHTS] per-source: c-elegans=0.320, cerebellum=0.017, flywire=0.001, berghia=0.175, microns=0.027, oct-glia=0.436, snemi=0.017, mouse-brain=0.007 | min-MMD source (NeuroADDA rule) = oct-glia +[10:02:45][ 26s][TEACHERS] target-val Dice: c-elegans=0.476, cerebellum=0.633, flywire=0.142, berghia=0.467, microns=0.708, oct-glia=0.845, snemi=0.229, mouse-brain=0.115 +[10:02:45][ 26s][TEACHERS] ensemble target-val Dice: mixture=0.849, mixture@raw=0.847, mixture@logit=0.850, per_source=0.843, per_source@raw=0.619, per_source@logit=0.848, uniform=0.718, uniform@raw=0.677, uniform@logit=0.807 +[10:02:45][ 26s][WEIGHTS] using mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 +[10:02:45][ 26s][INIT] student initialised from teacher oct-glia (weight 0.636) +[10:02:46][ 27s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[10:02:46][ 27s][KD] distilling the mixture ensemble (prob space) into the student +[10:03:27][ 68s][KD] soft labels for 3584 patches from 8 teachers in 41s +[10:03:59][ 100s][KD] epoch 1/5 | patches=3584 | loss=0.5673 | 32s +[10:04:30][ 131s][KD] epoch 2/5 | patches=3584 | loss=0.4392 | 31s +[10:05:03][ 164s][KD] epoch 3/5 | patches=3584 | loss=0.3945 | 33s +[10:05:36][ 197s][KD] epoch 4/5 | patches=3584 | loss=0.3593 | 33s +[10:06:07][ 228s][KD] epoch 5/5 | patches=3584 | loss=0.3311 | 31s +[10:06:08][ 230s][ROUND0] before any target label: val_dice=0.8515 val_vi=0.1453 +[10:06:08][ 230s][AL] patches=3584 (pool 2338 after dropping blank; free blank negatives: True) | budget=179 (5.0%) | rounds=6 | per_round=29 | acquisition=laplace | first_round=stratified +[10:06:16][ 237s][AL] round-1 uncertainty quality on 2338 patches: {'spearman_uncertainty_error': 0.4591574662939882, 'auroc_top20_error': 0.6181795328854153} +[10:06:39][ 260s][AL] round 1: stratified cold start: 15 patches from the membrane-rich half (ensemble prevalence >= 0.192) + 14 from the membrane-poor half of 2000 candidates +[10:06:42][ 264s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.811, uniform=0.768, top1=0.811, top2=0.811, top3=0.811 -> using mixture +[10:06:43][ 264s][WEIGHTS] round 1: student kept (labeled Dice 0.827 vs best teacher oct-glia 0.811); distilling the corrected ensemble into it +[10:06:43][ 264s][WEIGHTS] round 1: teacher Dice on 29 labeled patches + 29 blank (179827 fg px): c-elegans=0.279, cerebellum=0.617, flywire=0.194, berghia=0.316, microns=0.585, oct-glia=0.811, snemi=0.111, mouse-brain=0.210 | weights: c-elegans=0.001, cerebellum=0.006, flywire=0.000, berghia=0.002, microns=0.005, oct-glia=0.986, snemi=0.000, mouse-brain=0.000 +[10:07:09][ 290s][KD] soft labels for 2309 patches from 8 teachers in 26s +[10:07:30][ 311s][KD] epoch 1/2 | patches=2309 | loss=0.1381 | 21s +[10:07:51][ 332s][KD] epoch 2/2 | patches=2309 | loss=0.1323 | 21s +[10:08:14][ 355s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[10:08:16][ 357s][FT] round 1/6 epoch 1/5 | patches=58 | loss=0.8383 +[10:08:17][ 358s][FT] round 1/6 epoch 2/5 | patches=58 | loss=0.8501 +[10:08:18][ 360s][FT] round 1/6 epoch 3/5 | patches=58 | loss=0.6314 +[10:08:20][ 361s][FT] round 1/6 epoch 4/5 | patches=58 | loss=0.4879 +[10:08:21][ 363s][FT] round 1/6 epoch 5/5 | patches=58 | loss=0.4561 +[10:08:23][ 364s][AL] round 1/6 | labeled=29 | selection_seconds=30.8000 round_seconds=133.2000 restart_student_dice=0.8273 restart_teacher_dice=0.8106 val_dice=0.8568 val_vi=0.1554 val_ece=0.0169 +[10:08:31][ 372s][AL] round 2: 586/2309 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:08:36][ 378s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.680, uniform=0.674, top1=0.680, top2=0.680, top3=0.680 -> using mixture +[10:08:36][ 378s][WEIGHTS] round 2: teacher Dice on 58 labeled patches + 58 blank (549483 fg px): c-elegans=0.320, cerebellum=0.519, flywire=0.188, berghia=0.329, microns=0.518, oct-glia=0.680, snemi=0.151, mouse-brain=0.275 | weights: c-elegans=0.004, cerebellum=0.009, flywire=0.000, berghia=0.007, microns=0.010, oct-glia=0.970, snemi=0.000, mouse-brain=0.001 +[10:08:42][ 383s][CONS] round 2: on the 29 new patches student Dice=0.647 vs ensemble 0.603 -> pseudo-labels from the student | consistency x0.49 +[10:08:45][ 386s][FT] round 2/6 epoch 1/5 | patches=116 | loss=0.5509 +[10:08:47][ 388s][FT] round 2/6 epoch 2/5 | patches=116 | loss=0.4116 +[10:08:49][ 391s][FT] round 2/6 epoch 3/5 | patches=116 | loss=0.3350 +[10:08:52][ 393s][FT] round 2/6 epoch 4/5 | patches=116 | loss=0.3681 +[10:08:54][ 395s][FT] round 2/6 epoch 5/5 | patches=116 | loss=0.3817 +[10:08:55][ 397s][AL] round 2/6 | labeled=58 | selection_seconds=7.6000 round_seconds=30.9000 pseudo_student_dice=0.6467 pseudo_ensemble_dice=0.6028 consistency_scale=0.4890 val_dice=0.8618 val_vi=0.1453 val_ece=0.0160 +[10:09:03][ 404s][AL] round 3: 586/2280 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:09:53][ 454s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.363, cerebellum=0.615, flywire=0.246, berghia=0.366, microns=0.621, oct-glia=0.869, snemi=0.273, mouse-brain=0.242 | blended with labeled Dice at 0.50 +[10:10:40][ 501s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.721, uniform=0.700, top1=0.721, top2=0.721, top3=0.721 -> using mixture +[10:10:40][ 501s][WEIGHTS] round 3: teacher Dice on 87 labeled patches + 87 blank (1138003 fg px): c-elegans=0.338, cerebellum=0.551, flywire=0.213, berghia=0.334, microns=0.538, oct-glia=0.721, snemi=0.243, mouse-brain=0.269 | weights: c-elegans=0.003, cerebellum=0.008, flywire=0.000, berghia=0.005, microns=0.008, oct-glia=0.975, snemi=0.000, mouse-brain=0.001 +[10:10:46][ 507s][CONS] round 3: on the 29 new patches student Dice=0.522 vs ensemble 0.450 -> pseudo-labels from the student | consistency x0.07 +[10:10:49][ 511s][FT] round 3/6 epoch 1/5 | patches=174 | loss=0.9099 +[10:10:53][ 514s][FT] round 3/6 epoch 2/5 | patches=174 | loss=0.8412 +[10:10:56][ 517s][FT] round 3/6 epoch 3/5 | patches=174 | loss=0.7315 +[10:10:59][ 521s][FT] round 3/6 epoch 4/5 | patches=174 | loss=0.7279 +[10:11:03][ 524s][FT] round 3/6 epoch 5/5 | patches=174 | loss=0.6884 +[10:11:04][ 526s][AL] round 3/6 | labeled=87 | selection_seconds=7.4000 round_seconds=127.4000 pseudo_student_dice=0.5217 pseudo_ensemble_dice=0.4502 consistency_scale=0.0720 val_dice=0.8723 val_vi=0.1533 val_ece=0.0118 +[10:11:11][ 533s][AL] round 4: 572/2251 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:12:03][ 584s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.379, cerebellum=0.620, flywire=0.246, berghia=0.376, microns=0.620, oct-glia=0.857, snemi=0.282, mouse-brain=0.257 | blended with labeled Dice at 0.50 +[10:12:52][ 633s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.690, uniform=0.682, top1=0.690, top2=0.690, top3=0.690 -> using mixture +[10:12:52][ 633s][WEIGHTS] round 4: teacher Dice on 116 labeled patches + 116 blank (1656088 fg px): c-elegans=0.333, cerebellum=0.541, flywire=0.209, berghia=0.330, microns=0.517, oct-glia=0.690, snemi=0.259, mouse-brain=0.275 | weights: c-elegans=0.004, cerebellum=0.010, flywire=0.000, berghia=0.006, microns=0.009, oct-glia=0.970, snemi=0.000, mouse-brain=0.001 +[10:12:57][ 639s][CONS] round 4: on the 29 new patches student Dice=0.547 vs ensemble 0.395 -> pseudo-labels from the student | consistency x0.16 +[10:13:02][ 643s][FT] round 4/6 epoch 1/5 | patches=232 | loss=0.6824 +[10:13:06][ 647s][FT] round 4/6 epoch 2/5 | patches=232 | loss=0.5460 +[10:13:10][ 652s][FT] round 4/6 epoch 3/5 | patches=232 | loss=0.5290 +[10:13:15][ 656s][FT] round 4/6 epoch 4/5 | patches=232 | loss=0.6514 +[10:13:19][ 660s][FT] round 4/6 epoch 5/5 | patches=232 | loss=0.5690 +[10:13:20][ 662s][AL] round 4/6 | labeled=116 | selection_seconds=7.4000 round_seconds=134.6000 pseudo_student_dice=0.5471 pseudo_ensemble_dice=0.3954 consistency_scale=0.1570 val_dice=0.7533 val_vi=0.3211 val_ece=0.0161 +[10:13:28][ 669s][AL] round 5: 528/2222 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:14:21][ 722s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.557, cerebellum=0.595, flywire=0.207, berghia=0.471, microns=0.463, oct-glia=0.556, snemi=0.429, mouse-brain=0.484 | blended with labeled Dice at 0.50 +[10:15:10][ 772s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.553, uniform=0.606, top1=0.549, top2=0.551, top3=0.549 -> using uniform +[10:15:10][ 772s][WEIGHTS] round 5: teacher Dice on 145 labeled patches + 145 blank (1757811 fg px): c-elegans=0.417, cerebellum=0.525, flywire=0.186, berghia=0.364, microns=0.439, oct-glia=0.549, snemi=0.314, mouse-brain=0.379 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[10:15:13][ 774s][CONS] round 5: on the 29 new patches student Dice=0.269 vs ensemble 0.561 -> pseudo-labels from the ensemble | consistency x1.00 +[10:15:19][ 780s][FT] round 5/6 epoch 1/5 | patches=290 | loss=0.6929 +[10:15:24][ 785s][FT] round 5/6 epoch 2/5 | patches=290 | loss=0.5823 +[10:15:29][ 791s][FT] round 5/6 epoch 3/5 | patches=290 | loss=0.5388 +[10:15:35][ 796s][FT] round 5/6 epoch 4/5 | patches=290 | loss=0.5194 +[10:15:40][ 801s][FT] round 5/6 epoch 5/5 | patches=290 | loss=0.4847 +[10:15:41][ 803s][AL] round 5/6 | labeled=145 | selection_seconds=7.5000 round_seconds=139.6000 pseudo_student_dice=0.2694 pseudo_ensemble_dice=0.5612 consistency_scale=1.0000 val_dice=0.8744 val_vi=0.1512 val_ece=0.0079 +[10:15:49][ 810s][AL] round 6: 573/2193 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:16:04][ 826s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.537, uniform=0.558, top1=0.536, top2=0.537, top3=0.536 -> using uniform +[10:16:04][ 826s][WEIGHTS] round 6: teacher Dice on 174 labeled patches + 174 blank (2188227 fg px): c-elegans=0.288, cerebellum=0.456, flywire=0.166, berghia=0.274, microns=0.415, oct-glia=0.536, snemi=0.205, mouse-brain=0.281 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[10:16:10][ 831s][CONS] round 6: on the 29 new patches student Dice=0.734 vs ensemble 0.563 -> pseudo-labels from the student | consistency x0.78 +[10:16:17][ 838s][FT] round 6/6 epoch 1/5 | patches=348 | loss=0.4139 +[10:16:23][ 844s][FT] round 6/6 epoch 2/5 | patches=348 | loss=0.4055 +[10:16:29][ 850s][FT] round 6/6 epoch 3/5 | patches=348 | loss=0.3732 +[10:16:36][ 857s][FT] round 6/6 epoch 4/5 | patches=348 | loss=0.3716 +[10:16:42][ 863s][FT] round 6/6 epoch 5/5 | patches=348 | loss=0.4127 +[10:16:43][ 865s][AL] round 6/6 | labeled=174 | selection_seconds=7.3000 round_seconds=60.4000 pseudo_student_dice=0.7344 pseudo_ensemble_dice=0.5635 consistency_scale=0.7810 val_dice=0.8617 val_vi=0.1666 val_ece=0.0074 +[10:16:46][ 868s][CONS] final consolidation: pseudo-labels refreshed from the student +[10:16:53][ 874s][FT] final 1/10 epoch 1/1 | patches=348 | loss=0.4970 +[10:16:59][ 881s][FT] final 2/10 epoch 1/1 | patches=348 | loss=0.4235 +[10:17:06][ 887s][FT] final 3/10 epoch 1/1 | patches=348 | loss=0.4503 +[10:17:12][ 893s][FT] final 4/10 epoch 1/1 | patches=348 | loss=0.3917 +[10:17:18][ 900s][FT] final 5/10 epoch 1/1 | patches=348 | loss=0.3825 +[10:17:25][ 906s][FT] final 6/10 epoch 1/1 | patches=348 | loss=0.3685 +[10:17:31][ 912s][FT] final 7/10 epoch 1/1 | patches=348 | loss=0.3657 +[10:17:37][ 919s][FT] final 8/10 epoch 1/1 | patches=348 | loss=0.3473 +[10:17:44][ 925s][FT] final 9/10 epoch 1/1 | patches=348 | loss=0.3830 +[10:17:50][ 931s][FT] final 10/10 epoch 1/1 | patches=348 | loss=0.3511 +[10:17:52][ 933s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=66.5000 val_dice=0.8885 val_vi=0.1459 val_ece=0.0064 +[10:17:52][ 933s][WEIGHTS] final: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[10:18:02][ 944s][TEST] raw student (no EMA, no TTA): patch_dice=0.8706 patch_vi=0.1784 +[10:18:04][ 945s][TEST] final model without TTA: patch_dice=0.8706 patch_vi=0.1784 +[10:18:04][ 945s][RESOURCES] {"phases": {"setup": {"seconds": 1.14, "peak_cpu_rss_mb": 1027.7, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 25.77, "peak_cpu_rss_mb": 1809.1, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 200.91, "peak_cpu_rss_mb": 2938.6, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 703.34, "peak_cpu_rss_mb": 2938.6, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4626.0, "rounds": 6}, "inference": {"seconds": 12.3, "peak_cpu_rss_mb": 2938.6, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4626.0, "patches": 768, "images": 6, "patches_per_second": 62.4, "seconds_per_image": 2.05}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 2938.6, "total_seconds": 943.5} +[10:18:04][ 945s][TEST] patch_dice=0.8722 patch_iou=0.8093 patch_precision=0.8889 patch_recall=0.8824 patch_vi=0.1747 image_dice=0.8012 image_iou=0.6755 image_precision=0.7977 image_recall=0.8106 image_vi=0.1969 cal_ece=0.0051 cal_nll=0.0693 cal_brier=0.0212 cal_mean_confidence=0.9759 cal_accuracy=0.9709 cal_predicted_prevalence=0.0741 cal_true_prevalence=0.0718 cal_mean_probability=0.0749 +[10:18:04][ 945s][DONE] octopus-vulgaris-vertical-lobe-sfltract macs-v11-norestart-acq-laplace b5 s3 | test patch Dice=0.8722 VI=0.1747 | image Dice=0.8012 VI=0.1969 | 945s diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..ae2cd2b59c6d7eb6955a7eecb4d28bd154088c58 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json @@ -0,0 +1,776 @@ +{ + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 3, + "config": { + "task": "membrane", + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "laplace", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "octopus-vulgaris-vertical-lobe-sfltract", + "--variants", + "macs-v11-norestart-acq-laplace", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "alphas": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "mixture": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "per_source": [ + 0.3196472209911767, + 0.016524281783989066, + 0.0007715704214095669, + 0.17539935780265475, + 0.02748152710843107, + 0.4363201963447303, + 0.016643455180579362, + 0.00721239036702913 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.00048684441851293454, + "mmd2_mixture": 0.01275567083190321, + "mmd2_uniform": 0.09698324592915836, + "mmd2_single_source": [ + 0.07244541101040552, + 0.256897395483041, + 0.3929088032810887, + 0.05287172109384486, + 0.26498470596000767, + 0.011934168264888134, + 0.339258942897097, + 0.2684433232010991 + ], + "median_squared_distance": 0.18344226468067681, + "gammas": [ + 1.3628266116054668, + 2.7256532232109336, + 5.451306446421867, + 10.902612892843734, + 21.80522578568747 + ], + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.0496200334528889, + 0.42663682749168497, + 0.8166058321365515, + 0.12600018979150718, + 0.3618977926472007, + 0.01001955344998784, + 0.4257222632020228, + 0.5321458020642489 + ], + "standardized": [ + -1.1548905481172125, + 0.3263029673133632, + 1.8583820949064265, + -0.8548142807860197, + 0.07196140979373515, + -1.3104697560349354, + 0.32270990006354566, + 0.7408182128610976 + ], + "tau": 0.5, + "max_weight": 0.4363201963447303, + "entropy": 1.3076016834873658, + "effective_sources": 3.6972957848927845, + "min_weight": 0.0007715704214095669 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.4757138324136084, + 0.6329416506363946, + 0.14191107641715917, + 0.4669227387129098, + 0.7079858630413108, + 0.8450575342195966, + 0.2290226697023551, + 0.11473197497037783 + ], + "ensemble_val_dice": { + "mixture": 0.8487996328814233, + "mixture@raw": 0.8468783370062004, + "mixture@logit": 0.8498165961112344, + "per_source": 0.8430704623597217, + "per_source@raw": 0.6192918885999461, + "per_source@logit": 0.8479512470904803, + "uniform": 0.7183095932451996, + "uniform@raw": 0.6771602818911759, + "uniform@logit": 0.8065076510127938 + }, + "val_patches": 768 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 3.5461363792419434, + 4.075904369354248 + ], + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "weight_stats": { + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.5952380952380953, + "spearman_weight_taxonomy": 0.6257415691925263, + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "top_source_is_best_teacher": true + }, + "init_teacher": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "student_init": "top-source", + "blank_fraction": 0.34765625, + "kd_seconds": 200.9, + "round0_val": { + "val_dice": 0.8515102797526266, + "val_vi": 0.14528530548020757, + "val_ece": 0.05870248018856551 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 29, + "selection_seconds": 30.8, + "round_seconds": 133.2, + "teacher_dice_labeled": [ + 0.278686973709348, + 0.6172069054477305, + 0.19414410467595433, + 0.3158733635275065, + 0.5849527168014728, + 0.8105764798984268, + 0.11111243697584838, + 0.21043401064256403 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 179827, + "evidence_nonempty_patches": 22, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.8273144858761027, + "restart_teacher_dice": 0.8105764798984268, + "weights": [ + 0.0007503874401662406, + 0.006390021724894883, + 2.240702364497097e-05, + 0.0016248054634756897, + 0.005132506760358811, + 0.9859339072089507, + 2.702979059074383e-05, + 0.0001189345879178869 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8107891012852039, + "uniform": 0.7680840640525722, + "top1": 0.8105764798984268, + "top2": 0.8107881846175238, + "top3": 0.8107436386731551 + }, + "val_dice": 0.8568044591794021, + "val_vi": 0.15536332065951733, + "val_ece": 0.016875869306927235 + }, + { + "round": 2, + "n_labeled": 58, + "selection_seconds": 7.6, + "round_seconds": 30.9, + "teacher_dice_labeled": [ + 0.3197324212585647, + 0.5186989851123923, + 0.18769042400258, + 0.32897712171977045, + 0.5183921599714619, + 0.6796743091805819, + 0.15061281497014214, + 0.2754290735520091 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 549483, + "evidence_nonempty_patches": 50, + "first_correction": false, + "restart": "", + "weights": [ + 0.004119763372614849, + 0.008689933929460799, + 7.650411497395904e-05, + 0.006745884040301596, + 0.009607033535081643, + 0.9697850851871787, + 0.00014612322378190518, + 0.000829672596606807 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6799376360850514, + "uniform": 0.6739134714897281, + "top1": 0.6796743091805819, + "top2": 0.6799873705508308, + "top3": 0.6796347829641398 + }, + "pseudo_student_dice": 0.6466831096930886, + "pseudo_ensemble_dice": 0.6027818783231503, + "pseudo_source": "student", + "consistency_scale": 0.489, + "val_dice": 0.861787266202709, + "val_vi": 0.14531560750334052, + "val_ece": 0.01596588622935088 + }, + { + "round": 3, + "n_labeled": 87, + "selection_seconds": 7.4, + "round_seconds": 127.4, + "teacher_dice_labeled": [ + 0.33816096920786365, + 0.5508747489357715, + 0.2127544630286278, + 0.3342955063859918, + 0.5383557530871647, + 0.7209293122344551, + 0.24296450621113985, + 0.2688145903643604 + ], + "teacher_dice_pool": [ + 0.3631213703221983, + 0.6145786325256796, + 0.24602950896413575, + 0.3664223347709242, + 0.6212679758984406, + 0.8686868739620435, + 0.273348192075486, + 0.24215863231774376 + ], + "evidence_fg_pixels": 1138003, + "evidence_nonempty_patches": 79, + "first_correction": false, + "restart": "", + "weights": [ + 0.0032977847478816197, + 0.007981231489846457, + 6.544137261997147e-05, + 0.004736441621020206, + 0.007809189399856895, + 0.9753479342568299, + 0.0002449702865546584, + 0.0005170068253904648 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.7213260143771558, + "uniform": 0.6999932181400559, + "top1": 0.7209293122344551, + "top2": 0.7211679464098952, + "top3": 0.7211597694240854 + }, + "pseudo_student_dice": 0.5216835236694869, + "pseudo_ensemble_dice": 0.4502389093759044, + "pseudo_source": "student", + "consistency_scale": 0.072, + "val_dice": 0.8722882299942253, + "val_vi": 0.153306299739384, + "val_ece": 0.011781420070394528 + }, + { + "round": 4, + "n_labeled": 116, + "selection_seconds": 7.4, + "round_seconds": 134.6, + "teacher_dice_labeled": [ + 0.33335654980390694, + 0.5414478519401984, + 0.20853802261015897, + 0.3302917388799157, + 0.5167268481721374, + 0.6897356638254668, + 0.2592273871447747, + 0.27469722168508637 + ], + "teacher_dice_pool": [ + 0.37884987150908017, + 0.6200130681762214, + 0.24550768821346203, + 0.37607195076658384, + 0.6200735224606091, + 0.856574798895515, + 0.28189492315590287, + 0.25733860588598917 + ], + "evidence_fg_pixels": 1656088, + "evidence_nonempty_patches": 108, + "first_correction": false, + "restart": "", + "weights": [ + 0.004269758757935813, + 0.009866787505957434, + 8.522892157258763e-05, + 0.006181735356236639, + 0.008545139890425265, + 0.9699149110425108, + 0.0003915512060296217, + 0.0007448873193318116 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6901279025681524, + "uniform": 0.6820152178728482, + "top1": 0.6897356638254668, + "top2": 0.6900723044431973, + "top3": 0.6898417583058918 + }, + "pseudo_student_dice": 0.5470911753511737, + "pseudo_ensemble_dice": 0.39544808173752155, + "pseudo_source": "student", + "consistency_scale": 0.157, + "val_dice": 0.7532872911825909, + "val_vi": 0.3210910036063804, + "val_ece": 0.016072281468791392 + }, + { + "round": 5, + "n_labeled": 145, + "selection_seconds": 7.5, + "round_seconds": 139.6, + "teacher_dice_labeled": [ + 0.41676594581650617, + 0.5246918455092885, + 0.18630663891166677, + 0.3641756496484856, + 0.4389208361506449, + 0.5486765842831056, + 0.31428819745996883, + 0.3791764241664638 + ], + "teacher_dice_pool": [ + 0.5572855135959716, + 0.5945990793338956, + 0.20735111987360755, + 0.47074842385203725, + 0.4634087060527902, + 0.5556566786907592, + 0.4294796873842047, + 0.4841524752871939 + ], + "evidence_fg_pixels": 1757811, + "evidence_nonempty_patches": 135, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.553159440532073, + "uniform": 0.6059717310472152, + "top1": 0.5486765842831056, + "top2": 0.5506095424125969, + "top3": 0.5490036599201282 + }, + "pseudo_student_dice": 0.2693943934850826, + "pseudo_ensemble_dice": 0.5612330996243244, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.874420324316172, + "val_vi": 0.15116773626438473, + "val_ece": 0.00787832594784592 + }, + { + "round": 6, + "n_labeled": 174, + "selection_seconds": 7.3, + "round_seconds": 60.4, + "teacher_dice_labeled": [ + 0.2876735695465173, + 0.45617902944643696, + 0.1661216846844591, + 0.2744202625517011, + 0.41452631979279514, + 0.5362464651534956, + 0.20464379152966422, + 0.28141783931791187 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 2188227, + "evidence_nonempty_patches": 164, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.5372292629570921, + "uniform": 0.5577719248545915, + "top1": 0.5362464651534956, + "top2": 0.5369621472938708, + "top3": 0.536194536894178 + }, + "pseudo_student_dice": 0.73435671070419, + "pseudo_ensemble_dice": 0.56346838228734, + "pseudo_source": "student", + "consistency_scale": 0.781, + "val_dice": 0.8617161226240094, + "val_vi": 0.16661853359663176, + "val_ece": 0.007415384371264444 + }, + { + "round": 7, + "n_labeled": 174, + "selection_seconds": 0.0, + "round_seconds": 66.5, + "final_consolidation": true, + "val_dice": 0.8885012303311527, + "val_vi": 0.14585021657683853, + "val_ece": 0.006406493248002914 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.4591574662939882, + "auroc_top20_error": 0.6181795328854153 + }, + "al_seconds": 703.3, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "weight_stats_final": { + "max_weight": 0.125, + "entropy": 2.0794415416798357, + "effective_sources": 7.999999999999998, + "min_weight": 0.125 + }, + "top_source_final": "c-elegans-dauer-stage", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8706010955494335, + "patch_iou": 0.8062181478807697, + "patch_precision": 0.8878454942987101, + "patch_recall": 0.879627908979745, + "patch_vi": 0.17839662107777113, + "n_patches": 768, + "cal_ece": 0.00687900400319244, + "cal_nll": 0.07303922115646799, + "cal_brier": 0.021915210183211336, + "cal_mean_confidence": 0.9768012147741905, + "cal_accuracy": 0.970001220703125, + "cal_predicted_prevalence": 0.07367706298828125, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.0744689811575781 + }, + "test_notta": { + "patch_dice": 0.8706010955494335, + "patch_iou": 0.8062181478807697, + "patch_precision": 0.8878454942987101, + "patch_recall": 0.879627908979745, + "patch_vi": 0.17839662107777113, + "n_patches": 768, + "cal_ece": 0.00687900400319244, + "cal_nll": 0.07303922115646799, + "cal_brier": 0.021915210183211336, + "cal_mean_confidence": 0.9768012147741905, + "cal_accuracy": 0.970001220703125, + "cal_predicted_prevalence": 0.07367706298828125, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.0744689811575781 + }, + "test": { + "patch_dice": 0.8721652716705893, + "patch_iou": 0.8092648914040117, + "patch_precision": 0.8889102641225229, + "patch_recall": 0.8823612947835796, + "patch_vi": 0.17472178939575436, + "n_patches": 768, + "image_dice": 0.8011970493537958, + "image_iou": 0.6754771032791154, + "image_precision": 0.7977181274448101, + "image_recall": 0.8106349655697888, + "image_vi": 0.19691996364868994, + "n_images": 6, + "cal_ece": 0.005060413319983306, + "cal_nll": 0.06933385008524794, + "cal_brier": 0.0211606984166082, + "cal_mean_confidence": 0.9759225850433184, + "cal_accuracy": 0.9709307352701823, + "cal_predicted_prevalence": 0.0740521748860677, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.07489795879321344 + }, + "eval_seconds": 12.3, + "total_seconds": 945.2, + "resources": { + "phases": { + "setup": { + "seconds": 1.14, + "peak_cpu_rss_mb": 1027.7, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 25.77, + "peak_cpu_rss_mb": 1809.1, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 200.91, + "peak_cpu_rss_mb": 2938.6, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4618.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 703.34, + "peak_cpu_rss_mb": 2938.6, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4626.0, + "rounds": 6 + }, + "inference": { + "seconds": 12.3, + "peak_cpu_rss_mb": 2938.6, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4626.0, + "patches": 768, + "images": 6, + "patches_per_second": 62.4, + "seconds_per_image": 2.05 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 2938.6, + "total_seconds": 943.5 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..c4cec84182a31cd19c214622ab34362244f5b0a1 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,29,30.8,133.2,"[0.278686973709348, 0.6172069054477305, 0.19414410467595433, 0.3158733635275065, 0.5849527168014728, 0.8105764798984268, 0.11111243697584838, 0.21043401064256403]",,179827,22,True,kept,0.8273144858761027,0.8105764798984268,"[0.0007503874401662406, 0.006390021724894883, 2.240702364497097e-05, 0.0016248054634756897, 0.005132506760358811, 0.9859339072089507, 2.702979059074383e-05, 0.0001189345879178869]",mixture,"{'mixture': 0.8107891012852039, 'uniform': 0.7680840640525722, 'top1': 0.8105764798984268, 'top2': 0.8107881846175238, 'top3': 0.8107436386731551}",0.8568044591794021,0.15536332065951733,0.016875869306927235,,,,, +2,58,7.6,30.9,"[0.3197324212585647, 0.5186989851123923, 0.18769042400258, 0.32897712171977045, 0.5183921599714619, 0.6796743091805819, 0.15061281497014214, 0.2754290735520091]",,549483,50,False,,,,"[0.004119763372614849, 0.008689933929460799, 7.650411497395904e-05, 0.006745884040301596, 0.009607033535081643, 0.9697850851871787, 0.00014612322378190518, 0.000829672596606807]",mixture,"{'mixture': 0.6799376360850514, 'uniform': 0.6739134714897281, 'top1': 0.6796743091805819, 'top2': 0.6799873705508308, 'top3': 0.6796347829641398}",0.861787266202709,0.14531560750334052,0.01596588622935088,0.6466831096930886,0.6027818783231503,student,0.489, +3,87,7.4,127.4,"[0.33816096920786365, 0.5508747489357715, 0.2127544630286278, 0.3342955063859918, 0.5383557530871647, 0.7209293122344551, 0.24296450621113985, 0.2688145903643604]","[0.3631213703221983, 0.6145786325256796, 0.24602950896413575, 0.3664223347709242, 0.6212679758984406, 0.8686868739620435, 0.273348192075486, 0.24215863231774376]",1138003,79,False,,,,"[0.0032977847478816197, 0.007981231489846457, 6.544137261997147e-05, 0.004736441621020206, 0.007809189399856895, 0.9753479342568299, 0.0002449702865546584, 0.0005170068253904648]",mixture,"{'mixture': 0.7213260143771558, 'uniform': 0.6999932181400559, 'top1': 0.7209293122344551, 'top2': 0.7211679464098952, 'top3': 0.7211597694240854}",0.8722882299942253,0.153306299739384,0.011781420070394528,0.5216835236694869,0.4502389093759044,student,0.072, +4,116,7.4,134.6,"[0.33335654980390694, 0.5414478519401984, 0.20853802261015897, 0.3302917388799157, 0.5167268481721374, 0.6897356638254668, 0.2592273871447747, 0.27469722168508637]","[0.37884987150908017, 0.6200130681762214, 0.24550768821346203, 0.37607195076658384, 0.6200735224606091, 0.856574798895515, 0.28189492315590287, 0.25733860588598917]",1656088,108,False,,,,"[0.004269758757935813, 0.009866787505957434, 8.522892157258763e-05, 0.006181735356236639, 0.008545139890425265, 0.9699149110425108, 0.0003915512060296217, 0.0007448873193318116]",mixture,"{'mixture': 0.6901279025681524, 'uniform': 0.6820152178728482, 'top1': 0.6897356638254668, 'top2': 0.6900723044431973, 'top3': 0.6898417583058918}",0.7532872911825909,0.3210910036063804,0.016072281468791392,0.5470911753511737,0.39544808173752155,student,0.157, +5,145,7.5,139.6,"[0.41676594581650617, 0.5246918455092885, 0.18630663891166677, 0.3641756496484856, 0.4389208361506449, 0.5486765842831056, 0.31428819745996883, 0.3791764241664638]","[0.5572855135959716, 0.5945990793338956, 0.20735111987360755, 0.47074842385203725, 0.4634087060527902, 0.5556566786907592, 0.4294796873842047, 0.4841524752871939]",1757811,135,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.553159440532073, 'uniform': 0.6059717310472152, 'top1': 0.5486765842831056, 'top2': 0.5506095424125969, 'top3': 0.5490036599201282}",0.874420324316172,0.15116773626438473,0.00787832594784592,0.2693943934850826,0.5612330996243244,ensemble,1.0, +6,174,7.3,60.4,"[0.2876735695465173, 0.45617902944643696, 0.1661216846844591, 0.2744202625517011, 0.41452631979279514, 0.5362464651534956, 0.20464379152966422, 0.28141783931791187]",,2188227,164,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.5372292629570921, 'uniform': 0.5577719248545915, 'top1': 0.5362464651534956, 'top2': 0.5369621472938708, 'top3': 0.536194536894178}",0.8617161226240094,0.16661853359663176,0.007415384371264444,0.73435671070419,0.56346838228734,student,0.781, +7,174,0.0,66.5,,,,,,,,,,,,0.8885012303311527,0.14585021657683853,0.006406493248002914,,,,,True diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..f8adbb9f253f8d6b2baa9fc3d6498ea659f26c27 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt @@ -0,0 +1,106 @@ +[03:54:15][ 0s][START] target=octopus-vulgaris-vertical-lobe-sfltract variant=macs-v11-norestart-acq-margin budget=5.0% seed=1 | {"task": "membrane", "target": "octopus-vulgaris-vertical-lobe-sfltract", "variant": "macs-v11-norestart-acq-margin", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "margin", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[03:54:15][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[03:54:15][ 0s][DATA] octopus-vulgaris-vertical-lobe-sfltract: train=3584 val=768 test=768 patches +[03:54:20][ 5s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, snemi=3.55, mouse-brain=4.08 +[03:54:22][ 6s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 2s +[03:54:27][ 12s][WEIGHTS] mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 | eff.sources=3.41 | MMD2 mix=0.0128 uniform=0.0970 +[03:54:27][ 12s][WEIGHTS] per-source: c-elegans=0.320, cerebellum=0.017, flywire=0.001, berghia=0.175, microns=0.027, oct-glia=0.436, snemi=0.017, mouse-brain=0.007 | min-MMD source (NeuroADDA rule) = oct-glia +[03:54:46][ 30s][TEACHERS] target-val Dice: c-elegans=0.476, cerebellum=0.633, flywire=0.142, berghia=0.467, microns=0.708, oct-glia=0.845, snemi=0.229, mouse-brain=0.115 +[03:54:46][ 30s][TEACHERS] ensemble target-val Dice: mixture=0.849, mixture@raw=0.847, mixture@logit=0.850, per_source=0.843, per_source@raw=0.619, per_source@logit=0.848, uniform=0.718, uniform@raw=0.677, uniform@logit=0.807 +[03:54:46][ 30s][WEIGHTS] using mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 +[03:54:46][ 30s][INIT] student initialised from teacher oct-glia (weight 0.636) +[03:54:47][ 32s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[03:54:47][ 32s][KD] distilling the mixture ensemble (prob space) into the student +[03:55:28][ 72s][KD] soft labels for 3584 patches from 8 teachers in 41s +[03:55:59][ 104s][KD] epoch 1/5 | patches=3584 | loss=0.5719 | 31s +[03:56:32][ 136s][KD] epoch 2/5 | patches=3584 | loss=0.4386 | 32s +[03:57:03][ 168s][KD] epoch 3/5 | patches=3584 | loss=0.4023 | 31s +[03:57:35][ 199s][KD] epoch 4/5 | patches=3584 | loss=0.3566 | 32s +[03:58:08][ 232s][KD] epoch 5/5 | patches=3584 | loss=0.3329 | 33s +[03:58:09][ 234s][ROUND0] before any target label: val_dice=0.8409 val_vi=0.1472 +[03:58:09][ 234s][AL] patches=3584 (pool 2338 after dropping blank; free blank negatives: True) | budget=179 (5.0%) | rounds=6 | per_round=29 | acquisition=margin | first_round=stratified +[03:58:13][ 237s][AL] round-1 uncertainty quality on 2338 patches: {'spearman_uncertainty_error': 0.3286775006068347, 'auroc_top20_error': 0.4926333470451118} +[03:58:36][ 260s][AL] round 1: stratified cold start: 15 patches from the membrane-rich half (ensemble prevalence >= 0.192) + 14 from the membrane-poor half of 2000 candidates +[03:58:39][ 263s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.731, uniform=0.739, top1=0.731, top2=0.731, top3=0.731 -> using uniform +[03:58:39][ 264s][WEIGHTS] round 1: student kept (labeled Dice 0.749 vs best teacher oct-glia 0.731); distilling the corrected ensemble into it +[03:58:39][ 264s][WEIGHTS] round 1: teacher Dice on 29 labeled patches + 29 blank (237130 fg px): c-elegans=0.296, cerebellum=0.620, flywire=0.196, berghia=0.330, microns=0.551, oct-glia=0.731, snemi=0.151, mouse-brain=0.234 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[03:59:06][ 290s][KD] soft labels for 2309 patches from 8 teachers in 26s +[03:59:27][ 311s][KD] epoch 1/2 | patches=2309 | loss=0.7236 | 21s +[03:59:47][ 332s][KD] epoch 2/2 | patches=2309 | loss=0.6172 | 20s +[04:00:10][ 355s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[04:00:12][ 356s][FT] round 1/6 epoch 1/5 | patches=58 | loss=1.4217 +[04:00:13][ 358s][FT] round 1/6 epoch 2/5 | patches=58 | loss=1.3267 +[04:00:15][ 359s][FT] round 1/6 epoch 3/5 | patches=58 | loss=1.3169 +[04:00:16][ 361s][FT] round 1/6 epoch 4/5 | patches=58 | loss=1.2685 +[04:00:18][ 362s][FT] round 1/6 epoch 5/5 | patches=58 | loss=1.3223 +[04:00:19][ 364s][AL] round 1/6 | labeled=29 | selection_seconds=26.7000 round_seconds=128.6000 restart_student_dice=0.7494 restart_teacher_dice=0.7308 val_dice=0.4001 val_vi=0.1954 val_ece=0.0596 +[04:00:23][ 367s][AL] round 2: 536/2309 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:00:29][ 373s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.655, uniform=0.552, top1=0.655, top2=0.655, top3=0.655 -> using mixture +[04:00:29][ 373s][WEIGHTS] round 2: teacher Dice on 58 labeled patches + 58 blank (430675 fg px): c-elegans=0.226, cerebellum=0.352, flywire=0.158, berghia=0.255, microns=0.437, oct-glia=0.655, snemi=0.128, mouse-brain=0.211 | weights: c-elegans=0.002, cerebellum=0.002, flywire=0.000, berghia=0.004, microns=0.006, oct-glia=0.985, snemi=0.000, mouse-brain=0.001 +[04:00:31][ 376s][CONS] round 2: on the 29 new patches student Dice=0.495 vs ensemble 0.550 -> pseudo-labels from the ensemble | consistency x1.00 +[04:00:34][ 378s][FT] round 2/6 epoch 1/5 | patches=116 | loss=1.3258 +[04:00:36][ 381s][FT] round 2/6 epoch 2/5 | patches=116 | loss=1.3389 +[04:00:38][ 383s][FT] round 2/6 epoch 3/5 | patches=116 | loss=1.3048 +[04:00:41][ 385s][FT] round 2/6 epoch 4/5 | patches=116 | loss=1.2951 +[04:00:43][ 388s][FT] round 2/6 epoch 5/5 | patches=116 | loss=1.2841 +[04:00:45][ 389s][AL] round 2/6 | labeled=58 | selection_seconds=3.9000 round_seconds=23.9000 pseudo_student_dice=0.4952 pseudo_ensemble_dice=0.5501 consistency_scale=1.0000 val_dice=0.4091 val_vi=0.1649 val_ece=0.0362 +[04:00:48][ 393s][AL] round 3: 575/2280 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:00:56][ 401s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.419, uniform=0.435, top1=0.417, top2=0.415, top3=0.416 -> using uniform +[04:00:56][ 401s][WEIGHTS] round 3: teacher Dice on 87 labeled patches + 87 blank (1337552 fg px): c-elegans=0.283, cerebellum=0.333, flywire=0.143, berghia=0.223, microns=0.357, oct-glia=0.417, snemi=0.238, mouse-brain=0.314 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[04:01:02][ 406s][CONS] round 3: on the 29 new patches student Dice=0.379 vs ensemble 0.330 -> pseudo-labels from the student | consistency x0.00 +[04:01:04][ 408s][FT] round 3/6 epoch 1/5 | patches=174 | loss=1.1071 +[04:01:05][ 410s][FT] round 3/6 epoch 2/5 | patches=174 | loss=1.0589 +[04:01:07][ 412s][FT] round 3/6 epoch 3/5 | patches=174 | loss=0.9893 +[04:01:09][ 414s][FT] round 3/6 epoch 4/5 | patches=174 | loss=0.9455 +[04:01:11][ 415s][FT] round 3/6 epoch 5/5 | patches=174 | loss=0.8964 +[04:01:12][ 417s][AL] round 3/6 | labeled=87 | selection_seconds=3.6000 round_seconds=26.3000 pseudo_student_dice=0.3792 pseudo_ensemble_dice=0.3304 consistency_scale=0.0000 val_dice=0.4166 val_vi=0.4205 val_ece=0.2388 +[04:01:16][ 420s][AL] round 4: 83/2251 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:02:07][ 472s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.328, cerebellum=0.460, flywire=0.164, berghia=0.272, microns=0.372, oct-glia=0.534, snemi=0.562, mouse-brain=0.246 | blended with labeled Dice at 0.50 +[04:02:56][ 521s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.495, uniform=0.510, top1=0.492, top2=0.493, top3=0.494 -> using uniform +[04:02:56][ 521s][WEIGHTS] round 4: teacher Dice on 116 labeled patches + 116 blank (1499750 fg px): c-elegans=0.297, cerebellum=0.407, flywire=0.154, berghia=0.252, microns=0.373, oct-glia=0.492, snemi=0.384, mouse-brain=0.266 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[04:02:59][ 523s][CONS] round 4: on the 29 new patches student Dice=0.123 vs ensemble 0.671 -> pseudo-labels from the ensemble | consistency x1.00 +[04:03:03][ 528s][FT] round 4/6 epoch 1/5 | patches=232 | loss=1.3456 +[04:03:08][ 532s][FT] round 4/6 epoch 2/5 | patches=232 | loss=1.2946 +[04:03:12][ 536s][FT] round 4/6 epoch 3/5 | patches=232 | loss=1.2901 +[04:03:16][ 541s][FT] round 4/6 epoch 4/5 | patches=232 | loss=1.2260 +[04:03:21][ 545s][FT] round 4/6 epoch 5/5 | patches=232 | loss=1.2357 +[04:03:22][ 547s][AL] round 4/6 | labeled=116 | selection_seconds=3.7000 round_seconds=128.2000 pseudo_student_dice=0.1227 pseudo_ensemble_dice=0.6706 consistency_scale=1.0000 val_dice=0.4529 val_vi=0.1619 val_ece=0.0398 +[04:03:26][ 550s][AL] round 5: 561/2222 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:03:39][ 564s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.451, uniform=0.479, top1=0.450, top2=0.450, top3=0.449 -> using uniform +[04:03:39][ 564s][WEIGHTS] round 5: teacher Dice on 145 labeled patches + 145 blank (1979991 fg px): c-elegans=0.286, cerebellum=0.376, flywire=0.149, berghia=0.242, microns=0.375, oct-glia=0.450, snemi=0.212, mouse-brain=0.304 | weights: c-elegans=0.125, cerebellum=0.125, flywire=0.125, berghia=0.125, microns=0.125, oct-glia=0.125, snemi=0.125, mouse-brain=0.125 +[04:03:45][ 569s][CONS] round 5: on the 29 new patches student Dice=0.657 vs ensemble 0.519 -> pseudo-labels from the student | consistency x0.52 +[04:03:50][ 575s][FT] round 5/6 epoch 1/5 | patches=290 | loss=1.0593 +[04:03:56][ 580s][FT] round 5/6 epoch 2/5 | patches=290 | loss=0.9525 +[04:04:01][ 586s][FT] round 5/6 epoch 3/5 | patches=290 | loss=0.9657 +[04:04:06][ 591s][FT] round 5/6 epoch 4/5 | patches=290 | loss=0.9394 +[04:04:12][ 596s][FT] round 5/6 epoch 5/5 | patches=290 | loss=0.9132 +[04:04:13][ 598s][AL] round 5/6 | labeled=145 | selection_seconds=3.8000 round_seconds=49.6000 pseudo_student_dice=0.6574 pseudo_ensemble_dice=0.5187 consistency_scale=0.5250 val_dice=0.8724 val_vi=0.1473 val_ece=0.0059 +[04:04:17][ 601s][AL] round 6: 569/2193 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:05:11][ 655s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.410, cerebellum=0.662, flywire=0.240, berghia=0.384, microns=0.599, oct-glia=0.865, snemi=0.303, mouse-brain=0.278 | blended with labeled Dice at 0.50 +[04:06:02][ 706s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.656, uniform=0.660, top1=0.656, top2=0.657, top3=0.656 -> using mixture +[04:06:02][ 706s][WEIGHTS] round 6: teacher Dice on 174 labeled patches + 174 blank (2503796 fg px): c-elegans=0.350, cerebellum=0.522, flywire=0.196, berghia=0.317, microns=0.489, oct-glia=0.656, snemi=0.265, mouse-brain=0.290 | weights: c-elegans=0.007, cerebellum=0.011, flywire=0.000, berghia=0.008, microns=0.009, oct-glia=0.963, snemi=0.001, mouse-brain=0.001 +[04:06:08][ 712s][CONS] round 6: on the 29 new patches student Dice=0.618 vs ensemble 0.436 -> pseudo-labels from the student | consistency x0.39 +[04:06:14][ 719s][FT] round 6/6 epoch 1/5 | patches=348 | loss=0.8625 +[04:06:21][ 725s][FT] round 6/6 epoch 2/5 | patches=348 | loss=0.8528 +[04:06:27][ 731s][FT] round 6/6 epoch 3/5 | patches=348 | loss=0.8994 +[04:06:33][ 738s][FT] round 6/6 epoch 4/5 | patches=348 | loss=0.8433 +[04:06:39][ 744s][FT] round 6/6 epoch 5/5 | patches=348 | loss=0.7899 +[04:06:41][ 746s][AL] round 6/6 | labeled=174 | selection_seconds=3.6000 round_seconds=146.1000 pseudo_student_dice=0.6182 pseudo_ensemble_dice=0.4361 consistency_scale=0.3940 val_dice=0.8941 val_vi=0.1443 val_ece=0.0098 +[04:06:44][ 749s][CONS] final consolidation: pseudo-labels refreshed from the student +[04:06:50][ 755s][FT] final 1/10 epoch 1/1 | patches=348 | loss=0.7530 +[04:06:57][ 761s][FT] final 2/10 epoch 1/1 | patches=348 | loss=0.7795 +[04:07:03][ 768s][FT] final 3/10 epoch 1/1 | patches=348 | loss=0.7870 +[04:07:09][ 774s][FT] final 4/10 epoch 1/1 | patches=348 | loss=0.7724 +[04:07:16][ 780s][FT] final 5/10 epoch 1/1 | patches=348 | loss=0.7437 +[04:07:22][ 787s][FT] final 6/10 epoch 1/1 | patches=348 | loss=0.7325 +[04:07:28][ 793s][FT] final 7/10 epoch 1/1 | patches=348 | loss=0.7313 +[04:07:35][ 799s][FT] final 8/10 epoch 1/1 | patches=348 | loss=0.7160 +[04:07:41][ 806s][FT] final 9/10 epoch 1/1 | patches=348 | loss=0.7257 +[04:07:47][ 812s][FT] final 10/10 epoch 1/1 | patches=348 | loss=0.7021 +[04:07:49][ 813s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=66.4000 val_dice=0.8996 val_vi=0.1415 val_ece=0.0054 +[04:07:49][ 813s][WEIGHTS] final: c-elegans=0.007, cerebellum=0.011, flywire=0.000, berghia=0.008, microns=0.009, oct-glia=0.963, snemi=0.001, mouse-brain=0.001 +[04:08:00][ 824s][TEST] raw student (no EMA, no TTA): patch_dice=0.8708 patch_vi=0.1813 +[04:08:01][ 826s][TEST] final model without TTA: patch_dice=0.8708 patch_vi=0.1813 +[04:08:01][ 826s][RESOURCES] {"phases": {"setup": {"seconds": 4.33, "peak_cpu_rss_mb": 1027.1, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 27.04, "peak_cpu_rss_mb": 1808.3, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 200.49, "peak_cpu_rss_mb": 2937.4, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 579.87, "peak_cpu_rss_mb": 2937.4, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4626.0, "rounds": 6}, "inference": {"seconds": 12.25, "peak_cpu_rss_mb": 2937.4, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4626.0, "patches": 768, "images": 6, "patches_per_second": 62.7, "seconds_per_image": 2.042}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 2937.4, "total_seconds": 824.0} +[04:08:01][ 826s][TEST] patch_dice=0.8774 patch_iou=0.8156 patch_precision=0.8777 patch_recall=0.9054 patch_vi=0.1781 image_dice=0.8013 image_iou=0.6749 image_precision=0.7567 image_recall=0.8591 image_vi=0.2061 cal_ece=0.0078 cal_nll=0.0752 cal_brier=0.0227 cal_mean_confidence=0.9765 cal_accuracy=0.9687 cal_predicted_prevalence=0.0818 cal_true_prevalence=0.0718 cal_mean_probability=0.0830 +[04:08:01][ 826s][DONE] octopus-vulgaris-vertical-lobe-sfltract macs-v11-norestart-acq-margin b5 s1 | test patch Dice=0.8774 VI=0.1781 | image Dice=0.8013 VI=0.2061 | 826s diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json new file mode 100644 index 0000000000000000000000000000000000000000..a78d695abae066836fa7825469ed059d8b6c830e --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json @@ -0,0 +1,767 @@ +{ + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 1, + "config": { + "task": "membrane", + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 1, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "margin", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "octopus-vulgaris-vertical-lobe-sfltract", + "--variants", + "macs-v11-norestart-acq-margin", + "--budgets", + "5", + "--seeds", + "1", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "alphas": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "mixture": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "per_source": [ + 0.3196472209911767, + 0.016524281783989066, + 0.0007715704214095669, + 0.17539935780265475, + 0.02748152710843107, + 0.4363201963447303, + 0.016643455180579362, + 0.00721239036702913 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.00048684441851293454, + "mmd2_mixture": 0.01275567083190321, + "mmd2_uniform": 0.09698324592915836, + "mmd2_single_source": [ + 0.07244541101040552, + 0.256897395483041, + 0.3929088032810887, + 0.05287172109384486, + 0.26498470596000767, + 0.011934168264888134, + 0.339258942897097, + 0.2684433232010991 + ], + "median_squared_distance": 0.18344226468067681, + "gammas": [ + 1.3628266116054668, + 2.7256532232109336, + 5.451306446421867, + 10.902612892843734, + 21.80522578568747 + ], + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.0496200334528889, + 0.42663682749168497, + 0.8166058321365515, + 0.12600018979150718, + 0.3618977926472007, + 0.01001955344998784, + 0.4257222632020228, + 0.5321458020642489 + ], + "standardized": [ + -1.1548905481172125, + 0.3263029673133632, + 1.8583820949064265, + -0.8548142807860197, + 0.07196140979373515, + -1.3104697560349354, + 0.32270990006354566, + 0.7408182128610976 + ], + "tau": 0.5, + "max_weight": 0.4363201963447303, + "entropy": 1.3076016834873658, + "effective_sources": 3.6972957848927845, + "min_weight": 0.0007715704214095669 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "seconds": 7.4 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.4757138324136084, + 0.6329416506363946, + 0.14191107641715917, + 0.4669227387129098, + 0.7079858630413108, + 0.8450575342195966, + 0.2290226697023551, + 0.11473197497037783 + ], + "ensemble_val_dice": { + "mixture": 0.8487996328814233, + "mixture@raw": 0.8468783370062004, + "mixture@logit": 0.8498165961112344, + "per_source": 0.8430704623597217, + "per_source@raw": 0.6192918885999461, + "per_source@logit": 0.8479512470904803, + "uniform": 0.7183095932451996, + "uniform@raw": 0.6771602818911759, + "uniform@logit": 0.8065076510127938 + }, + "val_patches": 768 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 3.5461363792419434, + 4.075904369354248 + ], + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "weight_stats": { + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.5952380952380953, + "spearman_weight_taxonomy": 0.6257415691925263, + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "top_source_is_best_teacher": true + }, + "init_teacher": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "student_init": "top-source", + "blank_fraction": 0.34765625, + "kd_seconds": 200.5, + "round0_val": { + "val_dice": 0.8408797944320378, + "val_vi": 0.1471621710815322, + "val_ece": 0.06254463788501025 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 29, + "selection_seconds": 26.7, + "round_seconds": 128.6, + "teacher_dice_labeled": [ + 0.29569636245422026, + 0.6197098722923, + 0.1958949614829199, + 0.3302973386383346, + 0.5508880100223538, + 0.7308314626395732, + 0.1505046807462434, + 0.23392840178802965 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 237130, + "evidence_nonempty_patches": 23, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.749368381186149, + "restart_teacher_dice": 0.7308314626395732, + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.730833809182471, + "uniform": 0.7387444226325794, + "top1": 0.7308314626395732, + "top2": 0.731135280471351, + "top3": 0.7306861293999558 + }, + "val_dice": 0.40014016389188084, + "val_vi": 0.19540568040446374, + "val_ece": 0.05964780378486368 + }, + { + "round": 2, + "n_labeled": 58, + "selection_seconds": 3.9, + "round_seconds": 23.9, + "teacher_dice_labeled": [ + 0.22591072500584414, + 0.3517937333074959, + 0.15802432610526804, + 0.25522763821882105, + 0.43711356544332, + 0.6554180053154208, + 0.127741343329143, + 0.2108749243984569 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 430675, + "evidence_nonempty_patches": 49, + "first_correction": false, + "restart": "", + "weights": [ + 0.0020876127760991476, + 0.0021202950500547766, + 7.36354072139125e-05, + 0.0041781983943484004, + 0.005518753282945707, + 0.9853076052792571, + 0.00015053234578362758, + 0.0005633674642973701 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6553196975148962, + "uniform": 0.551515749726868, + "top1": 0.6554180053154208, + "top2": 0.6551663541812812, + "top3": 0.6551893202438965 + }, + "pseudo_student_dice": 0.4951777392874907, + "pseudo_ensemble_dice": 0.5500707948375794, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.4091470433278109, + "val_vi": 0.16486421629670236, + "val_ece": 0.03624841428508624 + }, + { + "round": 3, + "n_labeled": 87, + "selection_seconds": 3.6, + "round_seconds": 26.3, + "teacher_dice_labeled": [ + 0.2825503726481719, + 0.33345789577819174, + 0.14333429919434593, + 0.22282084470182645, + 0.35654819022203804, + 0.4168395243709581, + 0.23813570786869134, + 0.3137797196004813 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1337552, + "evidence_nonempty_patches": 78, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.418985243294828, + "uniform": 0.4346946450266119, + "top1": 0.4168395243709581, + "top2": 0.41525322405442944, + "top3": 0.41554170758199205 + }, + "pseudo_student_dice": 0.3791860455265543, + "pseudo_ensemble_dice": 0.3304222018760689, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.41663767109360444, + "val_vi": 0.4204583851205233, + "val_ece": 0.2387630061823596 + }, + { + "round": 4, + "n_labeled": 116, + "selection_seconds": 3.7, + "round_seconds": 128.2, + "teacher_dice_labeled": [ + 0.2969813743939188, + 0.4072742803649082, + 0.15403333647361286, + 0.25234984003272964, + 0.37322180151283557, + 0.4923612267773314, + 0.38435115895613686, + 0.2663083992428939 + ], + "teacher_dice_pool": [ + 0.32769681668831147, + 0.4600198472689253, + 0.16416407488008553, + 0.2719897206667647, + 0.37222389220911345, + 0.5336886450915855, + 0.5621863582223808, + 0.24645560702223043 + ], + "evidence_fg_pixels": 1499750, + "evidence_nonempty_patches": 104, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.49466062043967657, + "uniform": 0.5095499189370184, + "top1": 0.4923612267773314, + "top2": 0.49321355675721384, + "top3": 0.494243065939525 + }, + "pseudo_student_dice": 0.12269247283674595, + "pseudo_ensemble_dice": 0.67057551441357, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.4529003077444256, + "val_vi": 0.1618501742375552, + "val_ece": 0.03984077059544547 + }, + { + "round": 5, + "n_labeled": 145, + "selection_seconds": 3.8, + "round_seconds": 49.6, + "teacher_dice_labeled": [ + 0.28635443682328976, + 0.3761179026020876, + 0.14867873534616236, + 0.2422393842583288, + 0.37512147747731395, + 0.4498236328623156, + 0.21215624038179035, + 0.30371129831917815 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1979991, + "evidence_nonempty_patches": 133, + "first_correction": false, + "restart": "", + "weights": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "ensemble_choice": "uniform", + "candidate_dice": { + "mixture": 0.4514495482765114, + "uniform": 0.478574732040192, + "top1": 0.4498236328623156, + "top2": 0.4503073352358141, + "top3": 0.4489798877982555 + }, + "pseudo_student_dice": 0.6573682579006355, + "pseudo_ensemble_dice": 0.5186765316630713, + "pseudo_source": "student", + "consistency_scale": 0.525, + "val_dice": 0.872368749411296, + "val_vi": 0.14727879594978444, + "val_ece": 0.005879442031278861 + }, + { + "round": 6, + "n_labeled": 174, + "selection_seconds": 3.6, + "round_seconds": 146.1, + "teacher_dice_labeled": [ + 0.34966348004284153, + 0.5224991181605332, + 0.1961675156290587, + 0.3170608008365486, + 0.48857806541319204, + 0.6559357726159281, + 0.26461644923199584, + 0.29042007013582627 + ], + "teacher_dice_pool": [ + 0.4100624776609388, + 0.6622054020576088, + 0.24016586508634488, + 0.3841922238786771, + 0.5988061724159435, + 0.8649623049709401, + 0.3026851799005647, + 0.27770776236942235 + ], + "evidence_fg_pixels": 2503796, + "evidence_nonempty_patches": 162, + "first_correction": false, + "restart": "", + "weights": [ + 0.006998517317108208, + 0.011367502944338007, + 0.00010486852735956787, + 0.007541049303920208, + 0.008979521030995922, + 0.9632193067092194, + 0.0005754075693240222, + 0.0012138265977347025 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6564190895903512, + "uniform": 0.6601919630013362, + "top1": 0.6559357726159281, + "top2": 0.6565601036059059, + "top3": 0.6557210754470408 + }, + "pseudo_student_dice": 0.618239192860665, + "pseudo_ensemble_dice": 0.4361382280324493, + "pseudo_source": "student", + "consistency_scale": 0.394, + "val_dice": 0.8941350213311284, + "val_vi": 0.14431789469130596, + "val_ece": 0.009760565396891685 + }, + { + "round": 7, + "n_labeled": 174, + "selection_seconds": 0.0, + "round_seconds": 66.4, + "final_consolidation": true, + "val_dice": 0.8995945889022355, + "val_vi": 0.14149496184449994, + "val_ece": 0.0054247581622388205 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.3286775006068347, + "auroc_top20_error": 0.4926333470451118 + }, + "al_seconds": 579.9, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.006998517317108208, + 0.011367502944338007, + 0.00010486852735956787, + 0.007541049303920208, + 0.008979521030995922, + 0.9632193067092194, + 0.0005754075693240222, + 0.0012138265977347025 + ], + "weight_stats_final": { + "max_weight": 0.9632193067092194, + "entropy": 0.2142932753715035, + "effective_sources": 1.2389859655851392, + "min_weight": 0.00010486852735956787 + }, + "top_source_final": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8708351518479791, + "patch_iou": 0.8067548801798852, + "patch_precision": 0.8696299699165779, + "patch_recall": 0.9032768737050612, + "patch_vi": 0.1812900673335056, + "n_patches": 768, + "cal_ece": 0.009750707704115536, + "cal_nll": 0.08071878577839842, + "cal_brier": 0.02364779874333658, + "cal_mean_confidence": 0.9777733720921362, + "cal_accuracy": 0.9680226643880209, + "cal_predicted_prevalence": 0.08139546712239583, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.08214897919586299 + }, + "test_notta": { + "patch_dice": 0.8708351518479791, + "patch_iou": 0.8067548801798852, + "patch_precision": 0.8696299699165779, + "patch_recall": 0.9032768737050612, + "patch_vi": 0.1812900673335056, + "n_patches": 768, + "cal_ece": 0.009750707704115536, + "cal_nll": 0.08071878577839842, + "cal_brier": 0.02364779874333658, + "cal_mean_confidence": 0.9777733720921362, + "cal_accuracy": 0.9680226643880209, + "cal_predicted_prevalence": 0.08139546712239583, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.08214897919586299 + }, + "test": { + "patch_dice": 0.8773587173152141, + "patch_iou": 0.8155625933215656, + "patch_precision": 0.8777242031910709, + "patch_recall": 0.9053688680403686, + "patch_vi": 0.1780580080486999, + "n_patches": 768, + "image_dice": 0.8013153406391118, + "image_iou": 0.6749187689784778, + "image_precision": 0.756699242658336, + "image_recall": 0.8591299849043321, + "image_vi": 0.2061392803206973, + "n_images": 6, + "cal_ece": 0.007827613516782296, + "cal_nll": 0.07517227874221732, + "cal_brier": 0.02271224630967771, + "cal_mean_confidence": 0.9764888849823414, + "cal_accuracy": 0.9687309265136719, + "cal_predicted_prevalence": 0.08178329467773438, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.08299390230094521 + }, + "eval_seconds": 12.2, + "total_seconds": 825.7, + "resources": { + "phases": { + "setup": { + "seconds": 4.33, + "peak_cpu_rss_mb": 1027.1, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 27.04, + "peak_cpu_rss_mb": 1808.3, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 200.49, + "peak_cpu_rss_mb": 2937.4, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4618.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 579.87, + "peak_cpu_rss_mb": 2937.4, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4626.0, + "rounds": 6 + }, + "inference": { + "seconds": 12.25, + "peak_cpu_rss_mb": 2937.4, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4626.0, + "patches": 768, + "images": 6, + "patches_per_second": 62.7, + "seconds_per_image": 2.042 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 2937.4, + "total_seconds": 824.0 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin_b5_s1.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..1cec50dd6bcc570ad8b83aaccb6a001567c0e4b8 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,29,26.7,128.6,"[0.29569636245422026, 0.6197098722923, 0.1958949614829199, 0.3302973386383346, 0.5508880100223538, 0.7308314626395732, 0.1505046807462434, 0.23392840178802965]",,237130,23,True,kept,0.749368381186149,0.7308314626395732,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.730833809182471, 'uniform': 0.7387444226325794, 'top1': 0.7308314626395732, 'top2': 0.731135280471351, 'top3': 0.7306861293999558}",0.40014016389188084,0.19540568040446374,0.05964780378486368,,,,, +2,58,3.9,23.9,"[0.22591072500584414, 0.3517937333074959, 0.15802432610526804, 0.25522763821882105, 0.43711356544332, 0.6554180053154208, 0.127741343329143, 0.2108749243984569]",,430675,49,False,,,,"[0.0020876127760991476, 0.0021202950500547766, 7.36354072139125e-05, 0.0041781983943484004, 0.005518753282945707, 0.9853076052792571, 0.00015053234578362758, 0.0005633674642973701]",mixture,"{'mixture': 0.6553196975148962, 'uniform': 0.551515749726868, 'top1': 0.6554180053154208, 'top2': 0.6551663541812812, 'top3': 0.6551893202438965}",0.4091470433278109,0.16486421629670236,0.03624841428508624,0.4951777392874907,0.5500707948375794,ensemble,1.0, +3,87,3.6,26.3,"[0.2825503726481719, 0.33345789577819174, 0.14333429919434593, 0.22282084470182645, 0.35654819022203804, 0.4168395243709581, 0.23813570786869134, 0.3137797196004813]",,1337552,78,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.418985243294828, 'uniform': 0.4346946450266119, 'top1': 0.4168395243709581, 'top2': 0.41525322405442944, 'top3': 0.41554170758199205}",0.41663767109360444,0.4204583851205233,0.2387630061823596,0.3791860455265543,0.3304222018760689,student,0.0, +4,116,3.7,128.2,"[0.2969813743939188, 0.4072742803649082, 0.15403333647361286, 0.25234984003272964, 0.37322180151283557, 0.4923612267773314, 0.38435115895613686, 0.2663083992428939]","[0.32769681668831147, 0.4600198472689253, 0.16416407488008553, 0.2719897206667647, 0.37222389220911345, 0.5336886450915855, 0.5621863582223808, 0.24645560702223043]",1499750,104,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.49466062043967657, 'uniform': 0.5095499189370184, 'top1': 0.4923612267773314, 'top2': 0.49321355675721384, 'top3': 0.494243065939525}",0.4529003077444256,0.1618501742375552,0.03984077059544547,0.12269247283674595,0.67057551441357,ensemble,1.0, +5,145,3.8,49.6,"[0.28635443682328976, 0.3761179026020876, 0.14867873534616236, 0.2422393842583288, 0.37512147747731395, 0.4498236328623156, 0.21215624038179035, 0.30371129831917815]",,1979991,133,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.4514495482765114, 'uniform': 0.478574732040192, 'top1': 0.4498236328623156, 'top2': 0.4503073352358141, 'top3': 0.4489798877982555}",0.872368749411296,0.14727879594978444,0.005879442031278861,0.6573682579006355,0.5186765316630713,student,0.525, +6,174,3.6,146.1,"[0.34966348004284153, 0.5224991181605332, 0.1961675156290587, 0.3170608008365486, 0.48857806541319204, 0.6559357726159281, 0.26461644923199584, 0.29042007013582627]","[0.4100624776609388, 0.6622054020576088, 0.24016586508634488, 0.3841922238786771, 0.5988061724159435, 0.8649623049709401, 0.3026851799005647, 0.27770776236942235]",2503796,162,False,,,,"[0.006998517317108208, 0.011367502944338007, 0.00010486852735956787, 0.007541049303920208, 0.008979521030995922, 0.9632193067092194, 0.0005754075693240222, 0.0012138265977347025]",mixture,"{'mixture': 0.6564190895903512, 'uniform': 0.6601919630013362, 'top1': 0.6559357726159281, 'top2': 0.6565601036059059, 'top3': 0.6557210754470408}",0.8941350213311284,0.14431789469130596,0.009760565396891685,0.618239192860665,0.4361382280324493,student,0.394, +7,174,0.0,66.4,,,,,,,,,,,,0.8995945889022355,0.14149496184449994,0.0054247581622388205,,,,,True diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..9df4a699d3771d6ea1e8aea83dde7d7b81886e60 --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt @@ -0,0 +1,102 @@ +[06:41:25][ 0s][START] target=octopus-vulgaris-vertical-lobe-sfltract variant=macs-v11-norestart-acq-random budget=5.0% seed=2 | {"task": "membrane", "target": "octopus-vulgaris-vertical-lobe-sfltract", "variant": "macs-v11-norestart-acq-random", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "random", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 0, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[06:41:25][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[06:41:25][ 0s][DATA] octopus-vulgaris-vertical-lobe-sfltract: train=3584 val=768 test=768 patches +[06:41:26][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, snemi=3.55, mouse-brain=4.08 +[06:41:27][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[06:41:32][ 8s][WEIGHTS] mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 | eff.sources=3.41 | MMD2 mix=0.0128 uniform=0.0970 +[06:41:32][ 8s][WEIGHTS] per-source: c-elegans=0.320, cerebellum=0.017, flywire=0.001, berghia=0.175, microns=0.027, oct-glia=0.436, snemi=0.017, mouse-brain=0.007 | min-MMD source (NeuroADDA rule) = oct-glia +[06:41:51][ 26s][TEACHERS] target-val Dice: c-elegans=0.476, cerebellum=0.633, flywire=0.142, berghia=0.467, microns=0.708, oct-glia=0.845, snemi=0.229, mouse-brain=0.115 +[06:41:51][ 26s][TEACHERS] ensemble target-val Dice: mixture=0.849, mixture@raw=0.847, mixture@logit=0.850, per_source=0.843, per_source@raw=0.619, per_source@logit=0.848, uniform=0.718, uniform@raw=0.677, uniform@logit=0.807 +[06:41:51][ 26s][WEIGHTS] using mixture: c-elegans=0.099, cerebellum=0.029, flywire=0.007, berghia=0.148, microns=0.032, oct-glia=0.636, snemi=0.019, mouse-brain=0.031 +[06:41:51][ 26s][INIT] student initialised from teacher oct-glia (weight 0.636) +[06:41:52][ 27s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[06:41:52][ 27s][KD] distilling the mixture ensemble (prob space) into the student +[06:42:33][ 68s][KD] soft labels for 3584 patches from 8 teachers in 41s +[06:43:05][ 100s][KD] epoch 1/5 | patches=3584 | loss=0.5953 | 31s +[06:43:38][ 133s][KD] epoch 2/5 | patches=3584 | loss=0.4565 | 33s +[06:44:09][ 165s][KD] epoch 3/5 | patches=3584 | loss=0.4018 | 32s +[06:44:41][ 196s][KD] epoch 4/5 | patches=3584 | loss=0.3660 | 32s +[06:45:13][ 228s][KD] epoch 5/5 | patches=3584 | loss=0.3395 | 31s +[06:45:14][ 229s][ROUND0] before any target label: val_dice=0.8459 val_vi=0.1456 +[06:45:14][ 229s][AL] patches=3584 (pool 2338 after dropping blank; free blank negatives: True) | budget=179 (5.0%) | rounds=6 | per_round=29 | acquisition=random | first_round=stratified +[06:45:18][ 233s][AL] round-1 uncertainty quality on 2338 patches: {'spearman_uncertainty_error': 0.00029523523234020787, 'auroc_top20_error': 0.4987465149229856} +[06:45:41][ 256s][AL] round 1: stratified cold start: 15 patches from the membrane-rich half (ensemble prevalence >= 0.192) + 14 from the membrane-poor half of 2000 candidates +[06:45:44][ 259s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.765, uniform=0.741, top1=0.765, top2=0.764, top3=0.765 -> using mixture +[06:45:44][ 260s][WEIGHTS] round 1: student kept (labeled Dice 0.743 vs best teacher oct-glia 0.765); distilling the corrected ensemble into it +[06:45:44][ 260s][WEIGHTS] round 1: teacher Dice on 29 labeled patches + 29 blank (179039 fg px): c-elegans=0.257, cerebellum=0.569, flywire=0.169, berghia=0.276, microns=0.581, oct-glia=0.765, snemi=0.110, mouse-brain=0.203 | weights: c-elegans=0.001, cerebellum=0.006, flywire=0.000, berghia=0.002, microns=0.008, oct-glia=0.983, snemi=0.000, mouse-brain=0.000 +[06:46:11][ 286s][KD] soft labels for 2309 patches from 8 teachers in 26s +[06:46:32][ 307s][KD] epoch 1/2 | patches=2309 | loss=0.1391 | 21s +[06:46:53][ 328s][KD] epoch 2/2 | patches=2309 | loss=0.1328 | 21s +[06:47:16][ 351s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[06:47:17][ 352s][FT] round 1/6 epoch 1/5 | patches=58 | loss=0.8723 +[06:47:19][ 354s][FT] round 1/6 epoch 2/5 | patches=58 | loss=0.6747 +[06:47:20][ 355s][FT] round 1/6 epoch 3/5 | patches=58 | loss=0.6562 +[06:47:21][ 357s][FT] round 1/6 epoch 4/5 | patches=58 | loss=0.4788 +[06:47:23][ 358s][FT] round 1/6 epoch 5/5 | patches=58 | loss=0.4930 +[06:47:24][ 359s][AL] round 1/6 | labeled=29 | selection_seconds=26.6000 round_seconds=128.7000 restart_student_dice=0.7435 restart_teacher_dice=0.7648 val_dice=0.3714 val_vi=0.3523 val_ece=0.0381 +[06:47:30][ 365s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.790, uniform=0.745, top1=0.790, top2=0.790, top3=0.790 -> using mixture +[06:47:30][ 365s][WEIGHTS] round 2: teacher Dice on 58 labeled patches + 58 blank (357977 fg px): c-elegans=0.272, cerebellum=0.565, flywire=0.185, berghia=0.295, microns=0.569, oct-glia=0.790, snemi=0.110, mouse-brain=0.208 | weights: c-elegans=0.001, cerebellum=0.005, flywire=0.000, berghia=0.002, microns=0.005, oct-glia=0.987, snemi=0.000, mouse-brain=0.000 +[06:47:33][ 368s][CONS] round 2: on the 29 new patches student Dice=0.502 vs ensemble 0.814 -> pseudo-labels from the ensemble | consistency x1.00 +[06:47:36][ 371s][FT] round 2/6 epoch 1/5 | patches=116 | loss=0.4292 +[06:47:38][ 373s][FT] round 2/6 epoch 2/5 | patches=116 | loss=0.4555 +[06:47:40][ 375s][FT] round 2/6 epoch 3/5 | patches=116 | loss=0.4129 +[06:47:43][ 378s][FT] round 2/6 epoch 4/5 | patches=116 | loss=0.3810 +[06:47:45][ 380s][FT] round 2/6 epoch 5/5 | patches=116 | loss=0.3797 +[06:47:46][ 382s][AL] round 2/6 | labeled=58 | selection_seconds=0.2000 round_seconds=20.7000 pseudo_student_dice=0.5021 pseudo_ensemble_dice=0.8137 consistency_scale=1.0000 val_dice=0.8627 val_vi=0.1482 val_ece=0.0201 +[06:47:55][ 390s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.777, uniform=0.749, top1=0.777, top2=0.777, top3=0.777 -> using mixture +[06:47:55][ 390s][WEIGHTS] round 3: teacher Dice on 87 labeled patches + 87 blank (538912 fg px): c-elegans=0.268, cerebellum=0.578, flywire=0.187, berghia=0.302, microns=0.575, oct-glia=0.777, snemi=0.110, mouse-brain=0.206 | weights: c-elegans=0.001, cerebellum=0.006, flywire=0.000, berghia=0.002, microns=0.006, oct-glia=0.984, snemi=0.000, mouse-brain=0.000 +[06:48:01][ 396s][CONS] round 3: on the 29 new patches student Dice=0.818 vs ensemble 0.751 -> pseudo-labels from the student | consistency x1.00 +[06:48:04][ 399s][FT] round 3/6 epoch 1/5 | patches=174 | loss=0.3224 +[06:48:07][ 402s][FT] round 3/6 epoch 2/5 | patches=174 | loss=0.2616 +[06:48:11][ 406s][FT] round 3/6 epoch 3/5 | patches=174 | loss=0.2470 +[06:48:14][ 409s][FT] round 3/6 epoch 4/5 | patches=174 | loss=0.2579 +[06:48:17][ 412s][FT] round 3/6 epoch 5/5 | patches=174 | loss=0.2570 +[06:48:19][ 414s][AL] round 3/6 | labeled=87 | selection_seconds=0.2000 round_seconds=30.8000 pseudo_student_dice=0.8182 pseudo_ensemble_dice=0.7506 consistency_scale=1.0000 val_dice=0.8649 val_vi=0.1438 val_ece=0.0222 +[06:49:10][ 465s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.396, cerebellum=0.643, flywire=0.241, berghia=0.380, microns=0.601, oct-glia=0.863, snemi=0.297, mouse-brain=0.267 | blended with labeled Dice at 0.50 +[06:49:59][ 514s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.810, uniform=0.786, top1=0.810, top2=0.810, top3=0.810 -> using mixture +[06:49:59][ 514s][WEIGHTS] round 4: teacher Dice on 116 labeled patches + 116 blank (840112 fg px): c-elegans=0.352, cerebellum=0.620, flywire=0.217, berghia=0.358, microns=0.585, oct-glia=0.810, snemi=0.211, mouse-brain=0.250 | weights: c-elegans=0.002, cerebellum=0.007, flywire=0.000, berghia=0.002, microns=0.005, oct-glia=0.984, snemi=0.000, mouse-brain=0.000 +[06:50:04][ 520s][CONS] round 4: on the 29 new patches student Dice=0.766 vs ensemble 0.719 -> pseudo-labels from the student | consistency x0.89 +[06:50:09][ 524s][FT] round 4/6 epoch 1/5 | patches=232 | loss=0.3260 +[06:50:13][ 528s][FT] round 4/6 epoch 2/5 | patches=232 | loss=0.3081 +[06:50:18][ 533s][FT] round 4/6 epoch 3/5 | patches=232 | loss=0.3103 +[06:50:22][ 537s][FT] round 4/6 epoch 4/5 | patches=232 | loss=0.2948 +[06:50:26][ 541s][FT] round 4/6 epoch 5/5 | patches=232 | loss=0.3194 +[06:50:28][ 543s][AL] round 4/6 | labeled=116 | selection_seconds=0.2000 round_seconds=127.4000 pseudo_student_dice=0.7662 pseudo_ensemble_dice=0.7189 consistency_scale=0.8870 val_dice=0.8627 val_vi=0.1471 val_ece=0.0211 +[06:51:21][ 596s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.387, cerebellum=0.638, flywire=0.245, berghia=0.382, microns=0.597, oct-glia=0.863, snemi=0.286, mouse-brain=0.260 | blended with labeled Dice at 0.50 +[06:52:10][ 646s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.797, uniform=0.768, top1=0.797, top2=0.797, top3=0.797 -> using mixture +[06:52:10][ 646s][WEIGHTS] round 5: teacher Dice on 145 labeled patches + 145 blank (1036034 fg px): c-elegans=0.343, cerebellum=0.602, flywire=0.214, berghia=0.353, microns=0.569, oct-glia=0.797, snemi=0.204, mouse-brain=0.245 | weights: c-elegans=0.002, cerebellum=0.006, flywire=0.000, berghia=0.003, microns=0.005, oct-glia=0.984, snemi=0.000, mouse-brain=0.000 +[06:52:16][ 651s][CONS] round 5: on the 29 new patches student Dice=0.649 vs ensemble 0.604 -> pseudo-labels from the student | consistency x0.50 +[06:52:22][ 657s][FT] round 5/6 epoch 1/5 | patches=290 | loss=0.3937 +[06:52:27][ 662s][FT] round 5/6 epoch 2/5 | patches=290 | loss=0.3946 +[06:52:32][ 668s][FT] round 5/6 epoch 3/5 | patches=290 | loss=0.3277 +[06:52:38][ 673s][FT] round 5/6 epoch 4/5 | patches=290 | loss=0.3587 +[06:52:43][ 678s][FT] round 5/6 epoch 5/5 | patches=290 | loss=0.3612 +[06:52:45][ 680s][AL] round 5/6 | labeled=145 | selection_seconds=0.2000 round_seconds=135.4000 pseudo_student_dice=0.6492 pseudo_ensemble_dice=0.6037 consistency_scale=0.4970 val_dice=0.8781 val_vi=0.1511 val_ece=0.0167 +[06:53:39][ 734s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.402, cerebellum=0.635, flywire=0.239, berghia=0.375, microns=0.587, oct-glia=0.848, snemi=0.308, mouse-brain=0.276 | blended with labeled Dice at 0.50 +[06:54:30][ 785s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.791, uniform=0.769, top1=0.791, top2=0.791, top3=0.791 -> using mixture +[06:54:30][ 785s][WEIGHTS] round 6: teacher Dice on 174 labeled patches + 174 blank (1249546 fg px): c-elegans=0.350, cerebellum=0.605, flywire=0.213, berghia=0.355, microns=0.566, oct-glia=0.791, snemi=0.216, mouse-brain=0.253 | weights: c-elegans=0.002, cerebellum=0.007, flywire=0.000, berghia=0.003, microns=0.005, oct-glia=0.983, snemi=0.000, mouse-brain=0.000 +[06:54:36][ 791s][CONS] round 6: on the 29 new patches student Dice=0.797 vs ensemble 0.744 -> pseudo-labels from the student | consistency x0.99 +[06:54:42][ 797s][FT] round 6/6 epoch 1/5 | patches=348 | loss=0.3454 +[06:54:49][ 804s][FT] round 6/6 epoch 2/5 | patches=348 | loss=0.3065 +[06:54:55][ 810s][FT] round 6/6 epoch 3/5 | patches=348 | loss=0.2850 +[06:55:01][ 816s][FT] round 6/6 epoch 4/5 | patches=348 | loss=0.3737 +[06:55:07][ 823s][FT] round 6/6 epoch 5/5 | patches=348 | loss=0.2806 +[06:55:09][ 824s][AL] round 6/6 | labeled=174 | selection_seconds=0.2000 round_seconds=142.8000 pseudo_student_dice=0.7974 pseudo_ensemble_dice=0.7436 consistency_scale=0.9910 val_dice=0.8896 val_vi=0.1489 val_ece=0.0137 +[06:55:12][ 827s][CONS] final consolidation: pseudo-labels refreshed from the student +[06:55:18][ 833s][FT] final 1/10 epoch 1/1 | patches=348 | loss=0.3100 +[06:55:25][ 840s][FT] final 2/10 epoch 1/1 | patches=348 | loss=0.2765 +[06:55:31][ 846s][FT] final 3/10 epoch 1/1 | patches=348 | loss=0.3011 +[06:55:37][ 852s][FT] final 4/10 epoch 1/1 | patches=348 | loss=0.2956 +[06:55:43][ 859s][FT] final 5/10 epoch 1/1 | patches=348 | loss=0.2610 +[06:55:50][ 865s][FT] final 6/10 epoch 1/1 | patches=348 | loss=0.2400 +[06:55:56][ 871s][FT] final 7/10 epoch 1/1 | patches=348 | loss=0.2573 +[06:56:02][ 878s][FT] final 8/10 epoch 1/1 | patches=348 | loss=0.2287 +[06:56:09][ 884s][FT] final 9/10 epoch 1/1 | patches=348 | loss=0.2246 +[06:56:15][ 890s][FT] final 10/10 epoch 1/1 | patches=348 | loss=0.2208 +[06:56:16][ 892s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=66.1000 val_dice=0.8934 val_vi=0.1478 val_ece=0.0119 +[06:56:16][ 892s][WEIGHTS] final: c-elegans=0.002, cerebellum=0.007, flywire=0.000, berghia=0.003, microns=0.005, oct-glia=0.983, snemi=0.000, mouse-brain=0.000 +[06:56:27][ 902s][TEST] raw student (no EMA, no TTA): patch_dice=0.8718 patch_vi=0.1764 +[06:56:29][ 904s][TEST] final model without TTA: patch_dice=0.8718 patch_vi=0.1764 +[06:56:29][ 904s][RESOURCES] {"phases": {"setup": {"seconds": 1.13, "peak_cpu_rss_mb": 1027.6, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 26.07, "peak_cpu_rss_mb": 1808.4, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 200.36, "peak_cpu_rss_mb": 2938.1, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 662.37, "peak_cpu_rss_mb": 2938.1, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4626.0, "rounds": 6}, "inference": {"seconds": 12.25, "peak_cpu_rss_mb": 2938.1, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4626.0, "patches": 768, "images": 6, "patches_per_second": 62.7, "seconds_per_image": 2.042}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 2938.1, "total_seconds": 902.2} +[06:56:29][ 904s][TEST] patch_dice=0.8733 patch_iou=0.8119 patch_precision=0.8885 patch_recall=0.8888 patch_vi=0.1735 image_dice=0.8026 image_iou=0.6775 image_precision=0.7886 image_recall=0.8228 image_vi=0.1982 cal_ece=0.0085 cal_nll=0.0749 cal_brier=0.0216 cal_mean_confidence=0.9795 cal_accuracy=0.9710 cal_predicted_prevalence=0.0761 cal_true_prevalence=0.0718 cal_mean_probability=0.0759 +[06:56:29][ 904s][DONE] octopus-vulgaris-vertical-lobe-sfltract macs-v11-norestart-acq-random b5 s2 | test patch Dice=0.8733 VI=0.1735 | image Dice=0.8026 VI=0.1982 | 904s diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/result.json b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..8a5371885414f736871527d6e7e7eda1b09336dc --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/result.json @@ -0,0 +1,776 @@ +{ + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 2, + "config": { + "task": "membrane", + "target": "octopus-vulgaris-vertical-lobe-sfltract", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "random", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 0, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "octopus-vulgaris-vertical-lobe-sfltract", + "--variants", + "macs-v11-norestart-acq-random", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "alphas": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "snemi3d", + "whole-mouse-brain" + ], + "mixture": [ + 0.09888773316940314, + 0.02852238358730045, + 0.006877111295002701, + 0.14762496137410624, + 0.031629415990885375, + 0.6364113152878205, + 0.01903121486130303, + 0.03101586443417853 + ], + "per_source": [ + 0.3196472209911767, + 0.016524281783989066, + 0.0007715704214095669, + 0.17539935780265475, + 0.02748152710843107, + 0.4363201963447303, + 0.016643455180579362, + 0.00721239036702913 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.00048684441851293454, + "mmd2_mixture": 0.01275567083190321, + "mmd2_uniform": 0.09698324592915836, + "mmd2_single_source": [ + 0.07244541101040552, + 0.256897395483041, + 0.3929088032810887, + 0.05287172109384486, + 0.26498470596000767, + 0.011934168264888134, + 0.339258942897097, + 0.2684433232010991 + ], + "median_squared_distance": 0.18344226468067681, + "gammas": [ + 1.3628266116054668, + 2.7256532232109336, + 5.451306446421867, + 10.902612892843734, + 21.80522578568747 + ], + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.0496200334528889, + 0.42663682749168497, + 0.8166058321365515, + 0.12600018979150718, + 0.3618977926472007, + 0.01001955344998784, + 0.4257222632020228, + 0.5321458020642489 + ], + "standardized": [ + -1.1548905481172125, + 0.3263029673133632, + 1.8583820949064265, + -0.8548142807860197, + 0.07196140979373515, + -1.3104697560349354, + 0.32270990006354566, + 0.7408182128610976 + ], + "tau": 0.5, + "max_weight": 0.4363201963447303, + "entropy": 1.3076016834873658, + "effective_sources": 3.6972957848927845, + "min_weight": 0.0007715704214095669 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "seconds": 6.2 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.4757138324136084, + 0.6329416506363946, + 0.14191107641715917, + 0.4669227387129098, + 0.7079858630413108, + 0.8450575342195966, + 0.2290226697023551, + 0.11473197497037783 + ], + "ensemble_val_dice": { + "mixture": 0.8487996328814233, + "mixture@raw": 0.8468783370062004, + "mixture@logit": 0.8498165961112344, + "per_source": 0.8430704623597217, + "per_source@raw": 0.6192918885999461, + "per_source@logit": 0.8479512470904803, + "uniform": 0.7183095932451996, + "uniform@raw": 0.6771602818911759, + "uniform@logit": 0.8065076510127938 + }, + "val_patches": 768 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 3.5461363792419434, + 4.075904369354248 + ], + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "weight_stats": { + "max_weight": 0.6364113152878205, + "entropy": 1.2268826413390275, + "effective_sources": 3.410580943129122, + "min_weight": 0.006877111295002701 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.5952380952380953, + "spearman_weight_taxonomy": 0.6257415691925263, + "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "top_source_is_best_teacher": true + }, + "init_teacher": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "student_init": "top-source", + "blank_fraction": 0.34765625, + "kd_seconds": 200.4, + "round0_val": { + "val_dice": 0.8459001660840654, + "val_vi": 0.14561681257014922, + "val_ece": 0.06096620951525809 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 29, + "selection_seconds": 26.6, + "round_seconds": 128.7, + "teacher_dice_labeled": [ + 0.2571777942968238, + 0.5690605504275248, + 0.1686588604013407, + 0.2758037194420113, + 0.5811237001396865, + 0.7648114620563284, + 0.10985703762520847, + 0.20342659662893037 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 179039, + "evidence_nonempty_patches": 21, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.7434767776207417, + "restart_teacher_dice": 0.7648114620563284, + "weights": [ + 0.0009536063313568787, + 0.00622160049410085, + 2.7365280978604556e-05, + 0.0017150542671718894, + 0.00778389682673956, + 0.9830816831898324, + 4.206174632514249e-05, + 0.0001747318634946766 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.7646874735949377, + "uniform": 0.7405359432386459, + "top1": 0.7648114620563284, + "top2": 0.7643850371191413, + "top3": 0.7646318754401064 + }, + "val_dice": 0.3714069295182818, + "val_vi": 0.3523276532410457, + "val_ece": 0.03809157325395153 + }, + { + "round": 2, + "n_labeled": 58, + "selection_seconds": 0.2, + "round_seconds": 20.7, + "teacher_dice_labeled": [ + 0.27233939627165826, + 0.5648992002411293, + 0.18494126905476696, + 0.2950788044938515, + 0.5694094632225994, + 0.7901282813438308, + 0.10961051669571481, + 0.20814065934578377 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 357977, + "evidence_nonempty_patches": 40, + "first_correction": false, + "restart": "", + "weights": [ + 0.0008651839537822588, + 0.004652891610929437, + 2.5107694955276247e-05, + 0.0016213688563193685, + 0.005397791634150245, + 0.9872621406952384, + 3.27122188146369e-05, + 0.00014280333581031675 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.7900516454534338, + "uniform": 0.744561558923848, + "top1": 0.7901282813438308, + "top2": 0.7899526344089269, + "top3": 0.7900473160877718 + }, + "pseudo_student_dice": 0.5020917211634688, + "pseudo_ensemble_dice": 0.8136604464983836, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8627172784603788, + "val_vi": 0.14816620428021832, + "val_ece": 0.020074298774123814 + }, + { + "round": 3, + "n_labeled": 87, + "selection_seconds": 0.2, + "round_seconds": 30.8, + "teacher_dice_labeled": [ + 0.2683289214714117, + 0.577856181422308, + 0.18673087992781356, + 0.3018313761883449, + 0.5746961679042322, + 0.777125265724869, + 0.11037391096234828, + 0.20553076412792937 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 538912, + "evidence_nonempty_patches": 61, + "first_correction": false, + "restart": "", + "weights": [ + 0.0009438330908227715, + 0.006014484421225186, + 2.9025722514262033e-05, + 0.0019697535289780604, + 0.006462194076105672, + 0.9843852962889938, + 3.7430824711988296e-05, + 0.00015798204664822644 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.7771597736566964, + "uniform": 0.7494724702493365, + "top1": 0.777125265724869, + "top2": 0.7773284324921306, + "top3": 0.7771113848505545 + }, + "pseudo_student_dice": 0.8181573209193873, + "pseudo_ensemble_dice": 0.7506062527980898, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8649044370034481, + "val_vi": 0.1438270705034725, + "val_ece": 0.0221553722447091 + }, + { + "round": 4, + "n_labeled": 116, + "selection_seconds": 0.2, + "round_seconds": 127.4, + "teacher_dice_labeled": [ + 0.3523551657918085, + 0.6195955910426016, + 0.21702358946896538, + 0.35817812539551996, + 0.5849809963185317, + 0.8099093730446942, + 0.21149852979362696, + 0.249944372842695 + ], + "teacher_dice_pool": [ + 0.3963038925311522, + 0.643366416580547, + 0.2412226943807228, + 0.3803305564344355, + 0.6006591180223585, + 0.8628850136387899, + 0.2966821474784586, + 0.26739609073689796 + ], + "evidence_fg_pixels": 840112, + "evidence_nonempty_patches": 85, + "first_correction": false, + "restart": "", + "weights": [ + 0.0015748239507074387, + 0.0065748496506743736, + 2.8298151940109134e-05, + 0.0024919434282833783, + 0.00515776924456544, + 0.9839208328705071, + 7.410090370459648e-05, + 0.0001773817996174603 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8100702315447105, + "uniform": 0.7858644717533334, + "top1": 0.8099093730446942, + "top2": 0.8102884590298836, + "top3": 0.8099587844188316 + }, + "pseudo_student_dice": 0.7662125214904288, + "pseudo_ensemble_dice": 0.7188745455139126, + "pseudo_source": "student", + "consistency_scale": 0.887, + "val_dice": 0.8627058602543363, + "val_vi": 0.14709296460397747, + "val_ece": 0.02107210011131645 + }, + { + "round": 5, + "n_labeled": 145, + "selection_seconds": 0.2, + "round_seconds": 135.4, + "teacher_dice_labeled": [ + 0.3434458390283317, + 0.6024990757453136, + 0.2141995135026267, + 0.3527860406324941, + 0.5688136071369587, + 0.7969131606271672, + 0.20414479797548948, + 0.24496127493124809 + ], + "teacher_dice_pool": [ + 0.38700939335882084, + 0.637700200318136, + 0.2451388677680745, + 0.3823338784049224, + 0.5968550856440995, + 0.8627161787334623, + 0.28581341353040957, + 0.26015229074359414 + ], + "evidence_fg_pixels": 1036034, + "evidence_nonempty_patches": 103, + "first_correction": false, + "restart": "", + "weights": [ + 0.0016407488958869057, + 0.006311599420629153, + 3.13325753000206e-05, + 0.0026892023139674236, + 0.004997475263623532, + 0.9840590206740554, + 7.841325407623796e-05, + 0.0001922076024614262 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.7971050077151005, + "uniform": 0.7675423537446693, + "top1": 0.7969131606271672, + "top2": 0.7972941217041166, + "top3": 0.7969877830674019 + }, + "pseudo_student_dice": 0.649230568272293, + "pseudo_ensemble_dice": 0.603722645216952, + "pseudo_source": "student", + "consistency_scale": 0.497, + "val_dice": 0.8780933225724049, + "val_vi": 0.15112189309925872, + "val_ece": 0.016665274923035406 + }, + { + "round": 6, + "n_labeled": 174, + "selection_seconds": 0.2, + "round_seconds": 142.8, + "teacher_dice_labeled": [ + 0.3501263783945856, + 0.6050864237945479, + 0.21269241472651582, + 0.35459013382319676, + 0.5658740646511347, + 0.7906464270163391, + 0.2161101273220039, + 0.25254151656279217 + ], + "teacher_dice_pool": [ + 0.40178570744004233, + 0.635485317205476, + 0.23930938140955102, + 0.37514848519991184, + 0.58683289956519, + 0.8480063602697288, + 0.30812144608107045, + 0.2764468271369436 + ], + "evidence_fg_pixels": 1249546, + "evidence_nonempty_patches": 127, + "first_correction": false, + "restart": "", + "weights": [ + 0.0018652112324314258, + 0.00688729721663382, + 3.281886314378884e-05, + 0.002911596851577237, + 0.00516008754641597, + 0.9828285328627763, + 9.39781582817084e-05, + 0.00022047726873971913 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.7907800643613214, + "uniform": 0.7685180324869638, + "top1": 0.7906464270163391, + "top2": 0.7910706954472992, + "top3": 0.7906376772800607 + }, + "pseudo_student_dice": 0.797396310946261, + "pseudo_ensemble_dice": 0.7435812984466436, + "pseudo_source": "student", + "consistency_scale": 0.991, + "val_dice": 0.8895698960959386, + "val_vi": 0.14886280621004105, + "val_ece": 0.013738702604143803 + }, + { + "round": 7, + "n_labeled": 174, + "selection_seconds": 0.0, + "round_seconds": 66.1, + "final_consolidation": true, + "val_dice": 0.8933540398949001, + "val_vi": 0.1477702542557411, + "val_ece": 0.01187762032777213 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.00029523523234020787, + "auroc_top20_error": 0.4987465149229856 + }, + "al_seconds": 662.4, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.0018652112324314258, + 0.00688729721663382, + 3.281886314378884e-05, + 0.002911596851577237, + 0.00516008754641597, + 0.9828285328627763, + 9.39781582817084e-05, + 0.00022047726873971913 + ], + "weight_stats_final": { + "max_weight": 0.9828285328627763, + "entropy": 0.11027510731915921, + "effective_sources": 1.116585208972361, + "min_weight": 3.281886314378884e-05 + }, + "top_source_final": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8717935908930711, + "patch_iou": 0.8079621865404086, + "patch_precision": 0.8811307975775252, + "patch_recall": 0.8911383126144642, + "patch_vi": 0.1763703127715098, + "n_patches": 768, + "cal_ece": 0.010235559901710106, + "cal_nll": 0.07946818594560806, + "cal_brier": 0.022279491154928013, + "cal_mean_confidence": 0.9805627915840384, + "cal_accuracy": 0.9704310099283854, + "cal_predicted_prevalence": 0.07677459716796875, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.07612501553469163 + }, + "test_notta": { + "patch_dice": 0.8717935908930711, + "patch_iou": 0.8079621865404086, + "patch_precision": 0.8811307975775252, + "patch_recall": 0.8911383126144642, + "patch_vi": 0.1763703127715098, + "n_patches": 768, + "cal_ece": 0.010235559901710106, + "cal_nll": 0.07946818594560806, + "cal_brier": 0.022279491154928013, + "cal_mean_confidence": 0.9805627915840384, + "cal_accuracy": 0.9704310099283854, + "cal_predicted_prevalence": 0.07677459716796875, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.07612501553469163 + }, + "test": { + "patch_dice": 0.8733024516459945, + "patch_iou": 0.8119174567416984, + "patch_precision": 0.8885141746129198, + "patch_recall": 0.8888266529845109, + "patch_vi": 0.1734969541682323, + "n_patches": 768, + "image_dice": 0.8025776625259756, + "image_iou": 0.6774712731283673, + "image_precision": 0.7886269324253391, + "image_recall": 0.8227675597721614, + "image_vi": 0.19818246493296066, + "n_images": 6, + "cal_ece": 0.00853267805767482, + "cal_nll": 0.07485122535739094, + "cal_brier": 0.021577405959943684, + "cal_mean_confidence": 0.9795055914134073, + "cal_accuracy": 0.9710172017415365, + "cal_predicted_prevalence": 0.07612228393554688, + "cal_true_prevalence": 0.07175954182942708, + "cal_mean_probability": 0.07594731116540443 + }, + "eval_seconds": 12.2, + "total_seconds": 903.9, + "resources": { + "phases": { + "setup": { + "seconds": 1.13, + "peak_cpu_rss_mb": 1027.6, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 26.07, + "peak_cpu_rss_mb": 1808.4, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 200.36, + "peak_cpu_rss_mb": 2938.1, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4618.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 662.37, + "peak_cpu_rss_mb": 2938.1, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4626.0, + "rounds": 6 + }, + "inference": { + "seconds": 12.25, + "peak_cpu_rss_mb": 2938.1, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4626.0, + "patches": 768, + "images": 6, + "patches_per_second": 62.7, + "seconds_per_image": 2.042 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 2938.1, + "total_seconds": 902.2 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..4b92bd8e89fbb92d0c69137bdfa7d94f348451fa --- /dev/null +++ b/macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,29,26.6,128.7,"[0.2571777942968238, 0.5690605504275248, 0.1686588604013407, 0.2758037194420113, 0.5811237001396865, 0.7648114620563284, 0.10985703762520847, 0.20342659662893037]",,179039,21,True,kept,0.7434767776207417,0.7648114620563284,"[0.0009536063313568787, 0.00622160049410085, 2.7365280978604556e-05, 0.0017150542671718894, 0.00778389682673956, 0.9830816831898324, 4.206174632514249e-05, 0.0001747318634946766]",mixture,"{'mixture': 0.7646874735949377, 'uniform': 0.7405359432386459, 'top1': 0.7648114620563284, 'top2': 0.7643850371191413, 'top3': 0.7646318754401064}",0.3714069295182818,0.3523276532410457,0.03809157325395153,,,,, +2,58,0.2,20.7,"[0.27233939627165826, 0.5648992002411293, 0.18494126905476696, 0.2950788044938515, 0.5694094632225994, 0.7901282813438308, 0.10961051669571481, 0.20814065934578377]",,357977,40,False,,,,"[0.0008651839537822588, 0.004652891610929437, 2.5107694955276247e-05, 0.0016213688563193685, 0.005397791634150245, 0.9872621406952384, 3.27122188146369e-05, 0.00014280333581031675]",mixture,"{'mixture': 0.7900516454534338, 'uniform': 0.744561558923848, 'top1': 0.7901282813438308, 'top2': 0.7899526344089269, 'top3': 0.7900473160877718}",0.8627172784603788,0.14816620428021832,0.020074298774123814,0.5020917211634688,0.8136604464983836,ensemble,1.0, +3,87,0.2,30.8,"[0.2683289214714117, 0.577856181422308, 0.18673087992781356, 0.3018313761883449, 0.5746961679042322, 0.777125265724869, 0.11037391096234828, 0.20553076412792937]",,538912,61,False,,,,"[0.0009438330908227715, 0.006014484421225186, 2.9025722514262033e-05, 0.0019697535289780604, 0.006462194076105672, 0.9843852962889938, 3.7430824711988296e-05, 0.00015798204664822644]",mixture,"{'mixture': 0.7771597736566964, 'uniform': 0.7494724702493365, 'top1': 0.777125265724869, 'top2': 0.7773284324921306, 'top3': 0.7771113848505545}",0.8649044370034481,0.1438270705034725,0.0221553722447091,0.8181573209193873,0.7506062527980898,student,1.0, +4,116,0.2,127.4,"[0.3523551657918085, 0.6195955910426016, 0.21702358946896538, 0.35817812539551996, 0.5849809963185317, 0.8099093730446942, 0.21149852979362696, 0.249944372842695]","[0.3963038925311522, 0.643366416580547, 0.2412226943807228, 0.3803305564344355, 0.6006591180223585, 0.8628850136387899, 0.2966821474784586, 0.26739609073689796]",840112,85,False,,,,"[0.0015748239507074387, 0.0065748496506743736, 2.8298151940109134e-05, 0.0024919434282833783, 0.00515776924456544, 0.9839208328705071, 7.410090370459648e-05, 0.0001773817996174603]",mixture,"{'mixture': 0.8100702315447105, 'uniform': 0.7858644717533334, 'top1': 0.8099093730446942, 'top2': 0.8102884590298836, 'top3': 0.8099587844188316}",0.8627058602543363,0.14709296460397747,0.02107210011131645,0.7662125214904288,0.7188745455139126,student,0.887, +5,145,0.2,135.4,"[0.3434458390283317, 0.6024990757453136, 0.2141995135026267, 0.3527860406324941, 0.5688136071369587, 0.7969131606271672, 0.20414479797548948, 0.24496127493124809]","[0.38700939335882084, 0.637700200318136, 0.2451388677680745, 0.3823338784049224, 0.5968550856440995, 0.8627161787334623, 0.28581341353040957, 0.26015229074359414]",1036034,103,False,,,,"[0.0016407488958869057, 0.006311599420629153, 3.13325753000206e-05, 0.0026892023139674236, 0.004997475263623532, 0.9840590206740554, 7.841325407623796e-05, 0.0001922076024614262]",mixture,"{'mixture': 0.7971050077151005, 'uniform': 0.7675423537446693, 'top1': 0.7969131606271672, 'top2': 0.7972941217041166, 'top3': 0.7969877830674019}",0.8780933225724049,0.15112189309925872,0.016665274923035406,0.649230568272293,0.603722645216952,student,0.497, +6,174,0.2,142.8,"[0.3501263783945856, 0.6050864237945479, 0.21269241472651582, 0.35459013382319676, 0.5658740646511347, 0.7906464270163391, 0.2161101273220039, 0.25254151656279217]","[0.40178570744004233, 0.635485317205476, 0.23930938140955102, 0.37514848519991184, 0.58683289956519, 0.8480063602697288, 0.30812144608107045, 0.2764468271369436]",1249546,127,False,,,,"[0.0018652112324314258, 0.00688729721663382, 3.281886314378884e-05, 0.002911596851577237, 0.00516008754641597, 0.9828285328627763, 9.39781582817084e-05, 0.00022047726873971913]",mixture,"{'mixture': 0.7907800643613214, 'uniform': 0.7685180324869638, 'top1': 0.7906464270163391, 'top2': 0.7910706954472992, 'top3': 0.7906376772800607}",0.8895698960959386,0.14886280621004105,0.013738702604143803,0.797396310946261,0.7435812984466436,student,0.991, +7,174,0.0,66.1,,,,,,,,,,,,0.8933540398949001,0.1477702542557411,0.01187762032777213,,,,,True diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/log.txt b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..6dfd1d0b9ce297c0d205ccd6bf6290c567a6defe --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/log.txt @@ -0,0 +1,104 @@ +[07:11:51][ 0s][START] target=snemi3d variant=macs-v11-norestart-acq-margin budget=5.0% seed=2 | {"task": "membrane", "target": "snemi3d", "variant": "macs-v11-norestart-acq-margin", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "margin", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[07:11:51][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[07:11:52][ 0s][DATA] snemi3d: train=3430 val=735 test=735 patches +[07:11:53][ 2s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, oct-sfl=2.64, mouse-brain=4.08 +[07:11:53][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[07:11:59][ 8s][WEIGHTS] mixture: c-elegans=0.005, cerebellum=0.058, flywire=0.394, berghia=0.027, microns=0.477, oct-glia=0.007, oct-sfl=0.011, mouse-brain=0.021 | eff.sources=3.24 | MMD2 mix=0.0601 uniform=0.1529 +[07:11:59][ 8s][WEIGHTS] per-source: c-elegans=0.003, cerebellum=0.308, flywire=0.031, berghia=0.005, microns=0.624, oct-glia=0.021, oct-sfl=0.005, mouse-brain=0.002 | min-MMD source (NeuroADDA rule) = microns +[07:12:17][ 26s][TEACHERS] target-val Dice: c-elegans=0.412, cerebellum=0.473, flywire=0.196, berghia=0.310, microns=0.589, oct-glia=0.137, oct-sfl=0.305, mouse-brain=0.350 +[07:12:17][ 26s][TEACHERS] ensemble target-val Dice: mixture=0.421, mixture@raw=0.463, mixture@logit=0.367, per_source=0.589, per_source@raw=0.591, per_source@logit=0.578, uniform=0.411, uniform@raw=0.467, uniform@logit=0.228 +[07:12:17][ 26s][WEIGHTS] using mixture: c-elegans=0.005, cerebellum=0.058, flywire=0.394, berghia=0.027, microns=0.477, oct-glia=0.007, oct-sfl=0.011, mouse-brain=0.021 +[07:12:17][ 26s][INIT] student initialised from teacher microns (weight 0.477) +[07:12:18][ 27s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[07:12:18][ 27s][KD] distilling the mixture ensemble (prob space) into the student +[07:12:57][ 66s][KD] soft labels for 3430 patches from 8 teachers in 39s +[07:13:29][ 97s][KD] epoch 1/5 | patches=3430 | loss=0.4413 | 31s +[07:14:00][ 129s][KD] epoch 2/5 | patches=3430 | loss=0.3718 | 31s +[07:14:31][ 159s][KD] epoch 3/5 | patches=3430 | loss=0.3562 | 31s +[07:15:02][ 190s][KD] epoch 4/5 | patches=3430 | loss=0.3503 | 31s +[07:15:32][ 221s][KD] epoch 5/5 | patches=3430 | loss=0.3474 | 31s +[07:15:34][ 222s][ROUND0] before any target label: val_dice=0.3537 val_vi=0.6275 +[07:15:34][ 222s][AL] patches=3430 (pool 3430 after dropping blank; free blank negatives: True) | budget=171 (5.0%) | rounds=6 | per_round=28 | acquisition=margin | first_round=stratified +[07:15:39][ 228s][AL] round-1 uncertainty quality on 3430 patches: {'spearman_uncertainty_error': 0.3417284798148, 'auroc_top20_error': 0.7648460675398855} +[07:16:02][ 251s][AL] round 1: stratified cold start: 14 patches from the membrane-rich half (ensemble prevalence >= 0.156) + 14 from the membrane-poor half of 2000 candidates +[07:16:04][ 253s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.512, uniform=0.372, top1=0.513, top2=0.514, top3=0.514 -> using mixture +[07:16:04][ 253s][WEIGHTS] round 1: student kept (labeled Dice 0.373 vs best teacher microns 0.513); distilling the corrected ensemble into it +[07:16:04][ 253s][WEIGHTS] round 1: teacher Dice on 28 labeled patches + 0 blank (229858 fg px): c-elegans=0.298, cerebellum=0.401, flywire=0.209, berghia=0.345, microns=0.513, oct-glia=0.175, oct-sfl=0.199, mouse-brain=0.227 | weights: c-elegans=0.001, cerebellum=0.036, flywire=0.036, berghia=0.010, microns=0.913, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.002 +[07:16:43][ 292s][KD] soft labels for 3402 patches from 8 teachers in 39s +[07:17:14][ 322s][KD] epoch 1/2 | patches=3402 | loss=0.2523 | 31s +[07:17:44][ 353s][KD] epoch 2/2 | patches=3402 | loss=0.2493 | 31s +[07:18:07][ 376s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[07:18:08][ 377s][FT] round 1/6 epoch 1/5 | patches=28 | loss=1.0783 +[07:18:09][ 378s][FT] round 1/6 epoch 2/5 | patches=28 | loss=0.9834 +[07:18:10][ 379s][FT] round 1/6 epoch 3/5 | patches=28 | loss=0.9413 +[07:18:11][ 379s][FT] round 1/6 epoch 4/5 | patches=28 | loss=0.9016 +[07:18:12][ 380s][FT] round 1/6 epoch 5/5 | patches=28 | loss=0.8905 +[07:18:13][ 382s][AL] round 1/6 | labeled=28 | selection_seconds=28.4000 round_seconds=158.1000 restart_student_dice=0.3728 restart_teacher_dice=0.5132 val_dice=0.7017 val_vi=0.5973 val_ece=0.0128 +[07:18:19][ 387s][AL] round 2: 2/3402 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:18:22][ 390s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.509, uniform=0.304, top1=0.510, top2=0.512, top3=0.512 -> using mixture +[07:18:22][ 390s][WEIGHTS] round 2: teacher Dice on 56 labeled patches + 0 blank (537619 fg px): c-elegans=0.343, cerebellum=0.380, flywire=0.189, berghia=0.273, microns=0.510, oct-glia=0.119, oct-sfl=0.162, mouse-brain=0.265 | weights: c-elegans=0.002, cerebellum=0.030, flywire=0.031, berghia=0.005, microns=0.927, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[07:18:27][ 395s][CONS] round 2: on the 28 new patches student Dice=0.547 vs ensemble 0.506 -> pseudo-labels from the student | consistency x0.16 +[07:18:28][ 396s][FT] round 2/6 epoch 1/5 | patches=56 | loss=0.8317 +[07:18:29][ 398s][FT] round 2/6 epoch 2/5 | patches=56 | loss=0.7916 +[07:18:30][ 399s][FT] round 2/6 epoch 3/5 | patches=56 | loss=0.7590 +[07:18:32][ 400s][FT] round 2/6 epoch 4/5 | patches=56 | loss=0.7264 +[07:18:33][ 402s][FT] round 2/6 epoch 5/5 | patches=56 | loss=0.7075 +[07:18:35][ 403s][AL] round 2/6 | labeled=56 | selection_seconds=5.5000 round_seconds=19.8000 pseudo_student_dice=0.5472 pseudo_ensemble_dice=0.5061 consistency_scale=0.1570 val_dice=0.7407 val_vi=0.5701 val_ece=0.0149 +[07:19:28][ 457s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.394, cerebellum=0.481, flywire=0.238, berghia=0.381, microns=0.664, oct-glia=0.201, oct-sfl=0.279, mouse-brain=0.316 | blended with labeled Dice at 0.50 +[07:20:14][ 502s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.571, uniform=0.369, top1=0.571, top2=0.572, top3=0.573 -> using mixture +[07:20:14][ 502s][WEIGHTS] round 3: teacher Dice on 84 labeled patches + 0 blank (878752 fg px): c-elegans=0.378, cerebellum=0.438, flywire=0.208, berghia=0.332, microns=0.571, oct-glia=0.162, oct-sfl=0.216, mouse-brain=0.300 | weights: c-elegans=0.001, cerebellum=0.030, flywire=0.020, berghia=0.005, microns=0.939, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[07:20:18][ 507s][CONS] round 3: on the 28 new patches student Dice=0.614 vs ensemble 0.421 -> pseudo-labels from the student | consistency x0.38 +[07:20:20][ 509s][FT] round 3/6 epoch 1/5 | patches=84 | loss=0.8095 +[07:20:22][ 511s][FT] round 3/6 epoch 2/5 | patches=84 | loss=0.8034 +[07:20:24][ 512s][FT] round 3/6 epoch 3/5 | patches=84 | loss=0.7503 +[07:20:26][ 514s][FT] round 3/6 epoch 4/5 | patches=84 | loss=0.7570 +[07:20:27][ 516s][FT] round 3/6 epoch 5/5 | patches=84 | loss=0.7426 +[07:20:29][ 517s][AL] round 3/6 | labeled=84 | selection_seconds=5.3000 round_seconds=112.9000 pseudo_student_dice=0.6139 pseudo_ensemble_dice=0.4207 consistency_scale=0.3800 val_dice=0.7468 val_vi=0.5417 val_ece=0.0109 +[07:21:23][ 571s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.379, cerebellum=0.467, flywire=0.230, berghia=0.373, microns=0.646, oct-glia=0.192, oct-sfl=0.280, mouse-brain=0.304 | blended with labeled Dice at 0.50 +[07:22:09][ 617s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.561, uniform=0.366, top1=0.561, top2=0.562, top3=0.562 -> using mixture +[07:22:09][ 617s][WEIGHTS] round 4: teacher Dice on 112 labeled patches + 0 blank (1174561 fg px): c-elegans=0.371, cerebellum=0.430, flywire=0.202, berghia=0.328, microns=0.561, oct-glia=0.162, oct-sfl=0.221, mouse-brain=0.295 | weights: c-elegans=0.002, cerebellum=0.031, flywire=0.021, berghia=0.005, microns=0.938, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[07:22:14][ 622s][CONS] round 4: on the 28 new patches student Dice=0.645 vs ensemble 0.469 -> pseudo-labels from the student | consistency x0.48 +[07:22:16][ 625s][FT] round 4/6 epoch 1/5 | patches=112 | loss=0.7597 +[07:22:18][ 627s][FT] round 4/6 epoch 2/5 | patches=112 | loss=0.7372 +[07:22:21][ 629s][FT] round 4/6 epoch 3/5 | patches=112 | loss=0.7457 +[07:22:23][ 631s][FT] round 4/6 epoch 4/5 | patches=112 | loss=0.7221 +[07:22:25][ 634s][FT] round 4/6 epoch 5/5 | patches=112 | loss=0.7071 +[07:22:27][ 635s][AL] round 4/6 | labeled=112 | selection_seconds=5.3000 round_seconds=116.2000 pseudo_student_dice=0.6452 pseudo_ensemble_dice=0.4695 consistency_scale=0.4840 val_dice=0.7543 val_vi=0.5300 val_ece=0.0126 +[07:23:21][ 690s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.382, cerebellum=0.470, flywire=0.233, berghia=0.378, microns=0.641, oct-glia=0.190, oct-sfl=0.278, mouse-brain=0.304 | blended with labeled Dice at 0.50 +[07:24:08][ 736s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.557, uniform=0.364, top1=0.558, top2=0.559, top3=0.559 -> using mixture +[07:24:08][ 736s][WEIGHTS] round 5: teacher Dice on 140 labeled patches + 0 blank (1440921 fg px): c-elegans=0.369, cerebellum=0.432, flywire=0.202, berghia=0.328, microns=0.558, oct-glia=0.162, oct-sfl=0.224, mouse-brain=0.293 | weights: c-elegans=0.002, cerebellum=0.032, flywire=0.022, berghia=0.005, microns=0.935, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[07:24:13][ 741s][CONS] round 5: on the 28 new patches student Dice=0.644 vs ensemble 0.466 -> pseudo-labels from the student | consistency x0.48 +[07:24:15][ 744s][FT] round 5/6 epoch 1/5 | patches=140 | loss=0.7219 +[07:24:18][ 747s][FT] round 5/6 epoch 2/5 | patches=140 | loss=0.7203 +[07:24:21][ 749s][FT] round 5/6 epoch 3/5 | patches=140 | loss=0.7187 +[07:24:24][ 752s][FT] round 5/6 epoch 4/5 | patches=140 | loss=0.7061 +[07:24:26][ 755s][FT] round 5/6 epoch 5/5 | patches=140 | loss=0.7081 +[07:24:28][ 756s][AL] round 5/6 | labeled=140 | selection_seconds=5.4000 round_seconds=119.8000 pseudo_student_dice=0.6441 pseudo_ensemble_dice=0.4658 consistency_scale=0.4800 val_dice=0.7540 val_vi=0.5364 val_ece=0.0150 +[07:25:23][ 811s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.391, cerebellum=0.477, flywire=0.236, berghia=0.380, microns=0.644, oct-glia=0.193, oct-sfl=0.280, mouse-brain=0.311 | blended with labeled Dice at 0.50 +[07:26:10][ 859s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.563, uniform=0.370, top1=0.564, top2=0.565, top3=0.565 -> using mixture +[07:26:10][ 859s][WEIGHTS] round 6: teacher Dice on 168 labeled patches + 0 blank (1701154 fg px): c-elegans=0.373, cerebellum=0.436, flywire=0.205, berghia=0.332, microns=0.564, oct-glia=0.166, oct-sfl=0.228, mouse-brain=0.294 | weights: c-elegans=0.002, cerebellum=0.031, flywire=0.021, berghia=0.005, microns=0.937, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[07:26:15][ 863s][CONS] round 6: on the 28 new patches student Dice=0.666 vs ensemble 0.527 -> pseudo-labels from the student | consistency x0.55 +[07:26:18][ 867s][FT] round 6/6 epoch 1/5 | patches=168 | loss=0.7158 +[07:26:21][ 870s][FT] round 6/6 epoch 2/5 | patches=168 | loss=0.6940 +[07:26:24][ 873s][FT] round 6/6 epoch 3/5 | patches=168 | loss=0.6996 +[07:26:28][ 876s][FT] round 6/6 epoch 4/5 | patches=168 | loss=0.6985 +[07:26:31][ 879s][FT] round 6/6 epoch 5/5 | patches=168 | loss=0.6847 +[07:26:32][ 881s][AL] round 6/6 | labeled=168 | selection_seconds=5.2000 round_seconds=123.1000 pseudo_student_dice=0.6663 pseudo_ensemble_dice=0.5269 consistency_scale=0.5540 val_dice=0.7572 val_vi=0.5271 val_ece=0.0199 +[07:26:35][ 884s][CONS] final consolidation: pseudo-labels refreshed from the student +[07:26:39][ 887s][FT] final 1/10 epoch 1/1 | patches=168 | loss=0.6821 +[07:26:42][ 890s][FT] final 2/10 epoch 1/1 | patches=168 | loss=0.6618 +[07:26:45][ 893s][FT] final 3/10 epoch 1/1 | patches=168 | loss=0.6616 +[07:26:48][ 897s][FT] final 4/10 epoch 1/1 | patches=168 | loss=0.6538 +[07:26:51][ 900s][FT] final 5/10 epoch 1/1 | patches=168 | loss=0.6508 +[07:26:54][ 903s][FT] final 6/10 epoch 1/1 | patches=168 | loss=0.6438 +[07:26:58][ 906s][FT] final 7/10 epoch 1/1 | patches=168 | loss=0.6371 +[07:27:01][ 909s][FT] final 8/10 epoch 1/1 | patches=168 | loss=0.6338 +[07:27:04][ 913s][FT] final 9/10 epoch 1/1 | patches=168 | loss=0.6283 +[07:27:07][ 916s][FT] final 10/10 epoch 1/1 | patches=168 | loss=0.6262 +[07:27:09][ 917s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=35.1000 val_dice=0.7544 val_vi=0.5236 val_ece=0.0221 +[07:27:09][ 917s][WEIGHTS] final: c-elegans=0.002, cerebellum=0.031, flywire=0.021, berghia=0.005, microns=0.937, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[07:27:19][ 928s][TEST] raw student (no EMA, no TTA): patch_dice=0.7464 patch_vi=0.5306 +[07:27:21][ 929s][TEST] final model without TTA: patch_dice=0.7464 patch_vi=0.5306 +[07:27:21][ 929s][RESOURCES] {"phases": {"setup": {"seconds": 1.13, "peak_cpu_rss_mb": 1027.0, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 25.21, "peak_cpu_rss_mb": 1809.0, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 194.08, "peak_cpu_rss_mb": 2895.7, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 695.12, "peak_cpu_rss_mb": 2895.7, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4646.0, "rounds": 6}, "inference": {"seconds": 11.65, "peak_cpu_rss_mb": 2895.7, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4646.0, "patches": 735, "images": 15, "patches_per_second": 63.1, "seconds_per_image": 0.777}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 2895.7, "total_seconds": 927.2} +[07:27:21][ 929s][TEST] patch_dice=0.7569 patch_iou=0.6183 patch_precision=0.7430 patch_recall=0.7744 patch_vi=0.5125 image_dice=0.8046 image_iou=0.6736 image_precision=0.7945 image_recall=0.8155 image_vi=0.5193 cal_ece=0.0111 cal_nll=0.1813 cal_brier=0.0572 cal_mean_confidence=0.9282 cal_accuracy=0.9172 cal_predicted_prevalence=0.2129 cal_true_prevalence=0.2059 cal_mean_probability=0.2114 +[07:27:21][ 929s][DONE] snemi3d macs-v11-norestart-acq-margin b5 s2 | test patch Dice=0.7569 VI=0.5125 | image Dice=0.8046 VI=0.5193 | 929s diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/result.json b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..079edbd596a4d53b3935f13af91a2b8cd6766c30 --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/result.json @@ -0,0 +1,785 @@ +{ + "target": "snemi3d", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 2, + "config": { + "task": "membrane", + "target": "snemi3d", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "margin", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "snemi3d", + "--variants", + "macs-v11-norestart-acq-margin", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "whole-mouse-brain" + ], + "alphas": [ + 0.00526521078604608, + 0.05775508166244069, + 0.39371182687582884, + 0.027420285915093805, + 0.47701492758994885, + 0.006835626005792793, + 0.011200163178997314, + 0.020796877985851614 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "whole-mouse-brain" + ], + "mixture": [ + 0.00526521078604608, + 0.05775508166244069, + 0.39371182687582884, + 0.027420285915093805, + 0.47701492758994885, + 0.006835626005792793, + 0.011200163178997314, + 0.020796877985851614 + ], + "per_source": [ + 0.0030099345500300633, + 0.30826329572645517, + 0.030874948568488214, + 0.005310476630139808, + 0.6243402297379833, + 0.021312995363178842, + 0.004803663858606227, + 0.0020844555651184054 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.04836304661062167, + "mmd2_mixture": 0.06012254011242163, + "mmd2_uniform": 0.15285652884935502, + "mmd2_single_source": [ + 0.4522663178230211, + 0.14701205466558376, + 0.12241636968711567, + 0.27265250499014093, + 0.08995147555294303, + 0.3744252998076626, + 0.3395783286723916, + 0.23687871798038196 + ], + "median_squared_distance": 0.18663079059841653, + "gammas": [ + 1.3395431654037109, + 2.6790863308074218, + 5.3581726616148435, + 10.716345323229687, + 21.432690646459374 + ], + "max_weight": 0.47701492758994885, + "entropy": 1.1759493501799958, + "effective_sources": 3.241218535100111, + "min_weight": 0.00526521078604608 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.46244015980269426, + 0.12211161125756897, + 0.2912827113901817, + 0.4206979810988831, + 0.07022524085432302, + 0.3185314378412357, + 0.4280722723409719, + 0.48945233449571823 + ], + "standardized": [ + 0.9323157340081956, + -1.3822022358410426, + -0.2316976633920459, + 0.6484341051420859, + -1.7350728286900672, + -0.0463835976280083, + 0.6985854358714068, + 1.1160210505294779 + ], + "tau": 0.5, + "max_weight": 0.6243402297379833, + "entropy": 0.9300685916208571, + "effective_sources": 2.534683029672754, + "min_weight": 0.0020844555651184054 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "microns_basil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.4124719184272143, + 0.4732143695257148, + 0.1963993253870582, + 0.31013775029960944, + 0.5894614335833763, + 0.13744118666208724, + 0.30544280939432356, + 0.35012843982905756 + ], + "ensemble_val_dice": { + "mixture": 0.4205090273501111, + "mixture@raw": 0.46291647296981747, + "mixture@logit": 0.3665771544587855, + "per_source": 0.5892233046139178, + "per_source@raw": 0.5905333874987907, + "per_source@logit": 0.5781025615679164, + "uniform": 0.41121937389276464, + "uniform@raw": 0.46695544697394986, + "uniform@logit": 0.22779528784246278 + }, + "val_patches": 735 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 2.6421806812286377, + 4.075904369354248 + ], + "top_source": "microns_basil", + "weight_stats": { + "max_weight": 0.47701492758994885, + "entropy": 1.1759493501799958, + "effective_sources": 3.241218535100111, + "min_weight": 0.00526521078604608 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.3571428571428572, + "spearman_weight_taxonomy": 0.50709255283711, + "top_source": "microns_basil", + "top_source_is_best_teacher": true + }, + "init_teacher": "microns_basil", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 194.1, + "round0_val": { + "val_dice": 0.35365409409657816, + "val_vi": 0.627529035005998, + "val_ece": 0.07640721373388315 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 28, + "selection_seconds": 28.4, + "round_seconds": 158.1, + "teacher_dice_labeled": [ + 0.2979926532667371, + 0.40138560205746643, + 0.2091063280267666, + 0.345169021450426, + 0.5132033840553815, + 0.1750097104074813, + 0.1991102608484407, + 0.22650021516746904 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 229858, + "evidence_nonempty_patches": 28, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.3728225794885769, + "restart_teacher_dice": 0.5132033840553815, + "weights": [ + 0.0011717197029955054, + 0.036143320102722025, + 0.03602118748512524, + 0.009780573323141589, + 0.9132470305826798, + 0.0004447112636304281, + 0.0009272386944261949, + 0.002264218845279129 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5124341043740674, + "uniform": 0.371832611058189, + "top1": 0.5132033840553815, + "top2": 0.5140566832535806, + "top3": 0.5140133725772849 + }, + "val_dice": 0.7016877048595026, + "val_vi": 0.5972861693259305, + "val_ece": 0.012805830910195319 + }, + { + "round": 2, + "n_labeled": 56, + "selection_seconds": 5.5, + "round_seconds": 19.8, + "teacher_dice_labeled": [ + 0.3432294869846446, + 0.38007716816165, + 0.1893160272069609, + 0.27325800973621645, + 0.5104215693825217, + 0.11859077059313591, + 0.16161456913007907, + 0.2653818534052973 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 537619, + "evidence_nonempty_patches": 56, + "first_correction": false, + "restart": "", + "weights": [ + 0.001923155375394908, + 0.03049413082456546, + 0.030855984596669918, + 0.004974945473752838, + 0.9273348235211605, + 0.0002641090935606338, + 0.0006653947874493764, + 0.003487456327446416 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.509141392934992, + "uniform": 0.30389289820522647, + "top1": 0.5104215693825217, + "top2": 0.5115149870379297, + "top3": 0.5115711424860857 + }, + "pseudo_student_dice": 0.5472098965968067, + "pseudo_ensemble_dice": 0.506101874934941, + "pseudo_source": "student", + "consistency_scale": 0.157, + "val_dice": 0.7407435071089333, + "val_vi": 0.5701167948241403, + "val_ece": 0.014880619563059446 + }, + { + "round": 3, + "n_labeled": 84, + "selection_seconds": 5.3, + "round_seconds": 112.9, + "teacher_dice_labeled": [ + 0.37771424181735946, + 0.4383924755772515, + 0.20803697104167745, + 0.3322271896816488, + 0.5713981309510543, + 0.1622958007666203, + 0.21629404144420006, + 0.30029101442837863 + ], + "teacher_dice_pool": [ + 0.3939879167492122, + 0.48133117792688046, + 0.2378871834264335, + 0.38115292524663485, + 0.6639196547742888, + 0.2009226453097915, + 0.2786374526977216, + 0.31603711564037745 + ], + "evidence_fg_pixels": 878752, + "evidence_nonempty_patches": 84, + "first_correction": false, + "restart": "", + "weights": [ + 0.0014948062507949434, + 0.030080258640846264, + 0.0204855948778826, + 0.004939617999442016, + 0.9394194674567167, + 0.0002251111641117732, + 0.0006329282351913225, + 0.002722215375014302 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.570805470117565, + "uniform": 0.3686947077957518, + "top1": 0.5713981309510543, + "top2": 0.5724235003227502, + "top3": 0.5725081953043248 + }, + "pseudo_student_dice": 0.6138777505811737, + "pseudo_ensemble_dice": 0.42072085731863107, + "pseudo_source": "student", + "consistency_scale": 0.38, + "val_dice": 0.7468245490384356, + "val_vi": 0.5417238671053858, + "val_ece": 0.010860615428633374 + }, + { + "round": 4, + "n_labeled": 112, + "selection_seconds": 5.3, + "round_seconds": 116.2, + "teacher_dice_labeled": [ + 0.37113155177689233, + 0.4304098004089352, + 0.20188035699616788, + 0.32804902477829634, + 0.5613490894740727, + 0.16195989381006395, + 0.22064107723935433, + 0.2945484326528721 + ], + "teacher_dice_pool": [ + 0.37930059726350224, + 0.46679191904083284, + 0.22973305931652954, + 0.37275095341645953, + 0.646127639407633, + 0.19164165882853965, + 0.2799476672005563, + 0.3042154481257559 + ], + "evidence_fg_pixels": 1174561, + "evidence_nonempty_patches": 112, + "first_correction": false, + "restart": "", + "weights": [ + 0.00154438979007035, + 0.030645980759803813, + 0.021255485098501642, + 0.005227669234419145, + 0.9375132175239673, + 0.0002475702667783434, + 0.00072944689996532, + 0.0028362404264940613 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5607148781393188, + "uniform": 0.3658178463120181, + "top1": 0.5613490894740727, + "top2": 0.5624084345460589, + "top3": 0.5624875840270268 + }, + "pseudo_student_dice": 0.6451945548649818, + "pseudo_ensemble_dice": 0.4694781440108841, + "pseudo_source": "student", + "consistency_scale": 0.484, + "val_dice": 0.7542581674695896, + "val_vi": 0.5300439451809495, + "val_ece": 0.012610876109984098 + }, + { + "round": 5, + "n_labeled": 140, + "selection_seconds": 5.4, + "round_seconds": 119.8, + "teacher_dice_labeled": [ + 0.3686013985558847, + 0.4316764686766728, + 0.2017286174831427, + 0.32834735201508225, + 0.5580438444411788, + 0.16184451751433904, + 0.22400369081247648, + 0.292771023767665 + ], + "teacher_dice_pool": [ + 0.3818583838279288, + 0.4695261790989113, + 0.23319827221739736, + 0.3783978491677929, + 0.6413947113666308, + 0.18985071416968632, + 0.2782163770065046, + 0.30448063459351216 + ], + "evidence_fg_pixels": 1440921, + "evidence_nonempty_patches": 140, + "first_correction": false, + "restart": "", + "weights": [ + 0.0015526486686203665, + 0.03200215204034361, + 0.02188349146114073, + 0.005406401675151992, + 0.9352495152292364, + 0.00025497758356847564, + 0.0007778608562707616, + 0.0028729524856675776 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5573747664340176, + "uniform": 0.36380727550724445, + "top1": 0.5580438444411788, + "top2": 0.5590948972738314, + "top3": 0.5591800198932264 + }, + "pseudo_student_dice": 0.6441366353069468, + "pseudo_ensemble_dice": 0.4658413022561706, + "pseudo_source": "student", + "consistency_scale": 0.48, + "val_dice": 0.7539582315006403, + "val_vi": 0.536392270389954, + "val_ece": 0.01499025266145266 + }, + { + "round": 6, + "n_labeled": 168, + "selection_seconds": 5.2, + "round_seconds": 123.1, + "teacher_dice_labeled": [ + 0.3734322017511206, + 0.4356543154702399, + 0.20464087198641312, + 0.3322555657020442, + 0.563837343525697, + 0.16618033768548526, + 0.22835194680726428, + 0.2938739092673822 + ], + "teacher_dice_pool": [ + 0.39086690892472775, + 0.4766700450989547, + 0.23643556022313506, + 0.3796429239120413, + 0.6442590563252943, + 0.1929323791472062, + 0.28010353436049357, + 0.3105360609364328 + ], + "evidence_fg_pixels": 1701154, + "evidence_nonempty_patches": 168, + "first_correction": false, + "restart": "", + "weights": [ + 0.0015400210244406964, + 0.03147228651218212, + 0.021293052134217047, + 0.005313185861618959, + 0.9366166665504763, + 0.00025165512057501784, + 0.0007678204876118409, + 0.0027453123088780123 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.563181371815908, + "uniform": 0.3701849952532835, + "top1": 0.563837343525697, + "top2": 0.5648706041858641, + "top3": 0.5649549898869545 + }, + "pseudo_student_dice": 0.6663378365932968, + "pseudo_ensemble_dice": 0.526872228608967, + "pseudo_source": "student", + "consistency_scale": 0.554, + "val_dice": 0.7571927107648795, + "val_vi": 0.5271002562939713, + "val_ece": 0.019880075923030854 + }, + { + "round": 7, + "n_labeled": 168, + "selection_seconds": 0.0, + "round_seconds": 35.1, + "final_consolidation": true, + "val_dice": 0.7543813767791763, + "val_vi": 0.5235559405723194, + "val_ece": 0.02210728108908942 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.3417284798148, + "auroc_top20_error": 0.7648460675398855 + }, + "al_seconds": 695.1, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.0015400210244406964, + 0.03147228651218212, + 0.021293052134217047, + 0.005313185861618959, + 0.9366166665504763, + 0.00025165512057501784, + 0.0007678204876118409, + 0.0027453123088780123 + ], + "weight_stats_final": { + "max_weight": 0.9366166665504763, + "entropy": 0.31373234511206094, + "effective_sources": 1.3685233955470057, + "min_weight": 0.00025165512057501784 + }, + "top_source_final": "microns_basil", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7463989317843966, + "patch_iou": 0.6049912614203673, + "patch_precision": 0.7317079484329733, + "patch_recall": 0.7650737901798045, + "patch_vi": 0.5305903317803623, + "n_patches": 735, + "cal_ece": 0.020550589690599023, + "cal_nll": 0.19512247212991776, + "cal_brier": 0.06096353383874734, + "cal_mean_confidence": 0.93365247106815, + "cal_accuracy": 0.913101881377551, + "cal_predicted_prevalence": 0.213156356292517, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.21058684109828682 + }, + "test_notta": { + "patch_dice": 0.7463989317843966, + "patch_iou": 0.6049912614203673, + "patch_precision": 0.7317079484329733, + "patch_recall": 0.7650737901798045, + "patch_vi": 0.5305903317803623, + "n_patches": 735, + "cal_ece": 0.020550589690599023, + "cal_nll": 0.19512247212991776, + "cal_brier": 0.06096353383874734, + "cal_mean_confidence": 0.93365247106815, + "cal_accuracy": 0.913101881377551, + "cal_predicted_prevalence": 0.213156356292517, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.21058684109828682 + }, + "test": { + "patch_dice": 0.7568723733296536, + "patch_iou": 0.6183452706631142, + "patch_precision": 0.7430031357981336, + "patch_recall": 0.7743687919004077, + "patch_vi": 0.5124967708597993, + "n_patches": 735, + "image_dice": 0.8045990482995631, + "image_iou": 0.6735972010434514, + "image_precision": 0.7944607469663268, + "image_recall": 0.8154881555346277, + "image_vi": 0.5193016730431826, + "n_images": 15, + "cal_ece": 0.011084755612734554, + "cal_nll": 0.1812968731235373, + "cal_brier": 0.05720631288259247, + "cal_mean_confidence": 0.9282390392011701, + "cal_accuracy": 0.9171542835884354, + "cal_predicted_prevalence": 0.2129092261904762, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.2114104320651455 + }, + "eval_seconds": 11.6, + "total_seconds": 929.1, + "resources": { + "phases": { + "setup": { + "seconds": 1.13, + "peak_cpu_rss_mb": 1027.0, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 25.21, + "peak_cpu_rss_mb": 1809.0, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 194.08, + "peak_cpu_rss_mb": 2895.7, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4638.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 695.12, + "peak_cpu_rss_mb": 2895.7, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4646.0, + "rounds": 6 + }, + "inference": { + "seconds": 11.65, + "peak_cpu_rss_mb": 2895.7, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4646.0, + "patches": 735, + "images": 15, + "patches_per_second": 63.1, + "seconds_per_image": 0.777 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 2895.7, + "total_seconds": 927.2 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/snemi3d/macs-v11-norestart-acq-margin_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..546757c2db5e60856511f4546e5de3e3edf027f4 --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,28,28.4,158.1,"[0.2979926532667371, 0.40138560205746643, 0.2091063280267666, 0.345169021450426, 0.5132033840553815, 0.1750097104074813, 0.1991102608484407, 0.22650021516746904]",,229858,28,True,kept,0.3728225794885769,0.5132033840553815,"[0.0011717197029955054, 0.036143320102722025, 0.03602118748512524, 0.009780573323141589, 0.9132470305826798, 0.0004447112636304281, 0.0009272386944261949, 0.002264218845279129]",mixture,"{'mixture': 0.5124341043740674, 'uniform': 0.371832611058189, 'top1': 0.5132033840553815, 'top2': 0.5140566832535806, 'top3': 0.5140133725772849}",0.7016877048595026,0.5972861693259305,0.012805830910195319,,,,, +2,56,5.5,19.8,"[0.3432294869846446, 0.38007716816165, 0.1893160272069609, 0.27325800973621645, 0.5104215693825217, 0.11859077059313591, 0.16161456913007907, 0.2653818534052973]",,537619,56,False,,,,"[0.001923155375394908, 0.03049413082456546, 0.030855984596669918, 0.004974945473752838, 0.9273348235211605, 0.0002641090935606338, 0.0006653947874493764, 0.003487456327446416]",mixture,"{'mixture': 0.509141392934992, 'uniform': 0.30389289820522647, 'top1': 0.5104215693825217, 'top2': 0.5115149870379297, 'top3': 0.5115711424860857}",0.7407435071089333,0.5701167948241403,0.014880619563059446,0.5472098965968067,0.506101874934941,student,0.157, +3,84,5.3,112.9,"[0.37771424181735946, 0.4383924755772515, 0.20803697104167745, 0.3322271896816488, 0.5713981309510543, 0.1622958007666203, 0.21629404144420006, 0.30029101442837863]","[0.3939879167492122, 0.48133117792688046, 0.2378871834264335, 0.38115292524663485, 0.6639196547742888, 0.2009226453097915, 0.2786374526977216, 0.31603711564037745]",878752,84,False,,,,"[0.0014948062507949434, 0.030080258640846264, 0.0204855948778826, 0.004939617999442016, 0.9394194674567167, 0.0002251111641117732, 0.0006329282351913225, 0.002722215375014302]",mixture,"{'mixture': 0.570805470117565, 'uniform': 0.3686947077957518, 'top1': 0.5713981309510543, 'top2': 0.5724235003227502, 'top3': 0.5725081953043248}",0.7468245490384356,0.5417238671053858,0.010860615428633374,0.6138777505811737,0.42072085731863107,student,0.38, +4,112,5.3,116.2,"[0.37113155177689233, 0.4304098004089352, 0.20188035699616788, 0.32804902477829634, 0.5613490894740727, 0.16195989381006395, 0.22064107723935433, 0.2945484326528721]","[0.37930059726350224, 0.46679191904083284, 0.22973305931652954, 0.37275095341645953, 0.646127639407633, 0.19164165882853965, 0.2799476672005563, 0.3042154481257559]",1174561,112,False,,,,"[0.00154438979007035, 0.030645980759803813, 0.021255485098501642, 0.005227669234419145, 0.9375132175239673, 0.0002475702667783434, 0.00072944689996532, 0.0028362404264940613]",mixture,"{'mixture': 0.5607148781393188, 'uniform': 0.3658178463120181, 'top1': 0.5613490894740727, 'top2': 0.5624084345460589, 'top3': 0.5624875840270268}",0.7542581674695896,0.5300439451809495,0.012610876109984098,0.6451945548649818,0.4694781440108841,student,0.484, +5,140,5.4,119.8,"[0.3686013985558847, 0.4316764686766728, 0.2017286174831427, 0.32834735201508225, 0.5580438444411788, 0.16184451751433904, 0.22400369081247648, 0.292771023767665]","[0.3818583838279288, 0.4695261790989113, 0.23319827221739736, 0.3783978491677929, 0.6413947113666308, 0.18985071416968632, 0.2782163770065046, 0.30448063459351216]",1440921,140,False,,,,"[0.0015526486686203665, 0.03200215204034361, 0.02188349146114073, 0.005406401675151992, 0.9352495152292364, 0.00025497758356847564, 0.0007778608562707616, 0.0028729524856675776]",mixture,"{'mixture': 0.5573747664340176, 'uniform': 0.36380727550724445, 'top1': 0.5580438444411788, 'top2': 0.5590948972738314, 'top3': 0.5591800198932264}",0.7539582315006403,0.536392270389954,0.01499025266145266,0.6441366353069468,0.4658413022561706,student,0.48, +6,168,5.2,123.1,"[0.3734322017511206, 0.4356543154702399, 0.20464087198641312, 0.3322555657020442, 0.563837343525697, 0.16618033768548526, 0.22835194680726428, 0.2938739092673822]","[0.39086690892472775, 0.4766700450989547, 0.23643556022313506, 0.3796429239120413, 0.6442590563252943, 0.1929323791472062, 0.28010353436049357, 0.3105360609364328]",1701154,168,False,,,,"[0.0015400210244406964, 0.03147228651218212, 0.021293052134217047, 0.005313185861618959, 0.9366166665504763, 0.00025165512057501784, 0.0007678204876118409, 0.0027453123088780123]",mixture,"{'mixture': 0.563181371815908, 'uniform': 0.3701849952532835, 'top1': 0.563837343525697, 'top2': 0.5648706041858641, 'top3': 0.5649549898869545}",0.7571927107648795,0.5271002562939713,0.019880075923030854,0.6663378365932968,0.526872228608967,student,0.554, +7,168,0.0,35.1,,,,,,,,,,,,0.7543813767791763,0.5235559405723194,0.02210728108908942,,,,,True diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/log.txt b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..3d8d469f8bcee88daa524abe4d6a76faa9c202e1 --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/log.txt @@ -0,0 +1,105 @@ +[10:18:06][ 0s][START] target=snemi3d variant=macs-v11-norestart-acq-mcdropout budget=5.0% seed=3 | {"task": "membrane", "target": "snemi3d", "variant": "macs-v11-norestart-acq-mcdropout", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "mcdropout", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[10:18:06][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[10:18:06][ 0s][DATA] snemi3d: train=3430 val=735 test=735 patches +[10:18:08][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, oct-sfl=2.64, mouse-brain=4.08 +[10:18:08][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[10:18:14][ 7s][WEIGHTS] mixture: c-elegans=0.005, cerebellum=0.058, flywire=0.394, berghia=0.027, microns=0.477, oct-glia=0.007, oct-sfl=0.011, mouse-brain=0.021 | eff.sources=3.24 | MMD2 mix=0.0601 uniform=0.1529 +[10:18:14][ 7s][WEIGHTS] per-source: c-elegans=0.003, cerebellum=0.308, flywire=0.031, berghia=0.005, microns=0.624, oct-glia=0.021, oct-sfl=0.005, mouse-brain=0.002 | min-MMD source (NeuroADDA rule) = microns +[10:18:32][ 25s][TEACHERS] target-val Dice: c-elegans=0.412, cerebellum=0.473, flywire=0.196, berghia=0.310, microns=0.589, oct-glia=0.137, oct-sfl=0.305, mouse-brain=0.350 +[10:18:32][ 25s][TEACHERS] ensemble target-val Dice: mixture=0.421, mixture@raw=0.463, mixture@logit=0.367, per_source=0.589, per_source@raw=0.591, per_source@logit=0.578, uniform=0.411, uniform@raw=0.467, uniform@logit=0.228 +[10:18:32][ 25s][WEIGHTS] using mixture: c-elegans=0.005, cerebellum=0.058, flywire=0.394, berghia=0.027, microns=0.477, oct-glia=0.007, oct-sfl=0.011, mouse-brain=0.021 +[10:18:32][ 25s][INIT] student initialised from teacher microns (weight 0.477) +[10:18:33][ 27s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[10:18:33][ 27s][KD] distilling the mixture ensemble (prob space) into the student +[10:19:12][ 66s][KD] soft labels for 3430 patches from 8 teachers in 39s +[10:19:43][ 96s][KD] epoch 1/5 | patches=3430 | loss=0.4446 | 30s +[10:20:14][ 127s][KD] epoch 2/5 | patches=3430 | loss=0.3724 | 31s +[10:20:45][ 158s][KD] epoch 3/5 | patches=3430 | loss=0.3566 | 31s +[10:21:15][ 188s][KD] epoch 4/5 | patches=3430 | loss=0.3518 | 30s +[10:21:45][ 219s][KD] epoch 5/5 | patches=3430 | loss=0.3476 | 31s +[10:21:47][ 220s][ROUND0] before any target label: val_dice=0.3504 val_vi=0.6311 +[10:21:47][ 220s][AL] patches=3430 (pool 3430 after dropping blank; free blank negatives: True) | budget=171 (5.0%) | rounds=6 | per_round=28 | acquisition=mcdropout | first_round=stratified +[10:21:53][ 227s][AL] round-1 uncertainty quality on 3430 patches: {'spearman_uncertainty_error': 0.4795522004497552, 'auroc_top20_error': 0.8258676231842175} +[10:22:16][ 250s][AL] round 1: stratified cold start: 14 patches from the membrane-rich half (ensemble prevalence >= 0.156) + 14 from the membrane-poor half of 2000 candidates +[10:22:18][ 252s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.560, uniform=0.405, top1=0.561, top2=0.563, top3=0.563 -> using mixture +[10:22:18][ 252s][WEIGHTS] round 1: student kept (labeled Dice 0.392 vs best teacher microns 0.561); distilling the corrected ensemble into it +[10:22:18][ 252s][WEIGHTS] round 1: teacher Dice on 28 labeled patches + 0 blank (288982 fg px): c-elegans=0.361, cerebellum=0.451, flywire=0.218, berghia=0.284, microns=0.561, oct-glia=0.186, oct-sfl=0.372, mouse-brain=0.290 | weights: c-elegans=0.001, cerebellum=0.037, flywire=0.025, berghia=0.003, microns=0.927, oct-glia=0.000, oct-sfl=0.003, mouse-brain=0.003 +[10:22:57][ 291s][KD] soft labels for 3402 patches from 8 teachers in 39s +[10:23:28][ 322s][KD] epoch 1/2 | patches=3402 | loss=0.2479 | 31s +[10:23:57][ 351s][KD] epoch 2/2 | patches=3402 | loss=0.2446 | 29s +[10:24:20][ 374s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[10:24:21][ 375s][FT] round 1/6 epoch 1/5 | patches=28 | loss=1.0334 +[10:24:22][ 376s][FT] round 1/6 epoch 2/5 | patches=28 | loss=0.9354 +[10:24:23][ 377s][FT] round 1/6 epoch 3/5 | patches=28 | loss=0.8876 +[10:24:24][ 377s][FT] round 1/6 epoch 4/5 | patches=28 | loss=0.8557 +[10:24:25][ 378s][FT] round 1/6 epoch 5/5 | patches=28 | loss=0.8477 +[10:24:26][ 380s][AL] round 1/6 | labeled=28 | selection_seconds=29.4000 round_seconds=157.9000 restart_student_dice=0.3917 restart_teacher_dice=0.5609 val_dice=0.7324 val_vi=0.5694 val_ece=0.0110 +[10:24:32][ 386s][AL] round 2: 1/3402 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:24:35][ 389s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.444, uniform=0.325, top1=0.448, top2=0.454, top3=0.454 -> using top3 +[10:24:35][ 389s][WEIGHTS] round 2: teacher Dice on 56 labeled patches + 0 blank (636921 fg px): c-elegans=0.359, cerebellum=0.446, flywire=0.212, berghia=0.318, microns=0.448, oct-glia=0.171, oct-sfl=0.264, mouse-brain=0.305 | weights: c-elegans=0.004, cerebellum=0.106, flywire=0.000, berghia=0.000, microns=0.890, oct-glia=0.000, oct-sfl=0.000, mouse-brain=0.000 +[10:24:40][ 394s][CONS] round 2: on the 28 new patches student Dice=0.608 vs ensemble 0.318 -> pseudo-labels from the student | consistency x0.36 +[10:24:41][ 395s][FT] round 2/6 epoch 1/5 | patches=56 | loss=0.8341 +[10:24:43][ 396s][FT] round 2/6 epoch 2/5 | patches=56 | loss=0.7748 +[10:24:44][ 398s][FT] round 2/6 epoch 3/5 | patches=56 | loss=0.7368 +[10:24:45][ 399s][FT] round 2/6 epoch 4/5 | patches=56 | loss=0.7240 +[10:24:47][ 400s][FT] round 2/6 epoch 5/5 | patches=56 | loss=0.7130 +[10:24:48][ 402s][AL] round 2/6 | labeled=56 | selection_seconds=6.5000 round_seconds=20.8000 pseudo_student_dice=0.6085 pseudo_ensemble_dice=0.3182 consistency_scale=0.3620 val_dice=0.7421 val_vi=0.5560 val_ece=0.0160 +[10:24:54][ 408s][AL] round 3: 1/3374 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:25:43][ 456s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.397, cerebellum=0.488, flywire=0.243, berghia=0.387, microns=0.680, oct-glia=0.197, oct-sfl=0.278, mouse-brain=0.314 | blended with labeled Dice at 0.50 +[10:26:28][ 502s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.561, uniform=0.376, top1=0.562, top2=0.563, top3=0.563 -> using mixture +[10:26:28][ 502s][WEIGHTS] round 3: teacher Dice on 84 labeled patches + 0 blank (943818 fg px): c-elegans=0.377, cerebellum=0.452, flywire=0.221, berghia=0.344, microns=0.562, oct-glia=0.178, oct-sfl=0.251, mouse-brain=0.306 | weights: c-elegans=0.002, cerebellum=0.037, flywire=0.025, berghia=0.006, microns=0.925, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[10:26:33][ 507s][CONS] round 3: on the 28 new patches student Dice=0.645 vs ensemble 0.433 -> pseudo-labels from the student | consistency x0.48 +[10:26:35][ 509s][FT] round 3/6 epoch 1/5 | patches=84 | loss=0.7785 +[10:26:37][ 510s][FT] round 3/6 epoch 2/5 | patches=84 | loss=0.7679 +[10:26:39][ 512s][FT] round 3/6 epoch 3/5 | patches=84 | loss=0.7383 +[10:26:40][ 514s][FT] round 3/6 epoch 4/5 | patches=84 | loss=0.7332 +[10:26:42][ 516s][FT] round 3/6 epoch 5/5 | patches=84 | loss=0.7084 +[10:26:44][ 517s][AL] round 3/6 | labeled=84 | selection_seconds=6.4000 round_seconds=114.0000 pseudo_student_dice=0.6451 pseudo_ensemble_dice=0.4329 consistency_scale=0.4840 val_dice=0.7503 val_vi=0.5510 val_ece=0.0170 +[10:27:39][ 573s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.432, cerebellum=0.506, flywire=0.236, berghia=0.386, microns=0.658, oct-glia=0.193, oct-sfl=0.267, mouse-brain=0.343 | blended with labeled Dice at 0.50 +[10:28:25][ 618s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.558, uniform=0.362, top1=0.559, top2=0.560, top3=0.560 -> using mixture +[10:28:25][ 618s][WEIGHTS] round 4: teacher Dice on 112 labeled patches + 0 blank (1182224 fg px): c-elegans=0.390, cerebellum=0.449, flywire=0.215, berghia=0.337, microns=0.559, oct-glia=0.170, oct-sfl=0.239, mouse-brain=0.314 | weights: c-elegans=0.002, cerebellum=0.037, flywire=0.024, berghia=0.006, microns=0.926, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[10:28:30][ 623s][CONS] round 4: on the 28 new patches student Dice=0.644 vs ensemble 0.513 -> pseudo-labels from the student | consistency x0.48 +[10:28:32][ 626s][FT] round 4/6 epoch 1/5 | patches=112 | loss=0.7208 +[10:28:34][ 628s][FT] round 4/6 epoch 2/5 | patches=112 | loss=0.7066 +[10:28:36][ 630s][FT] round 4/6 epoch 3/5 | patches=112 | loss=0.6940 +[10:28:39][ 632s][FT] round 4/6 epoch 4/5 | patches=112 | loss=0.6903 +[10:28:41][ 634s][FT] round 4/6 epoch 5/5 | patches=112 | loss=0.6861 +[10:28:42][ 636s][AL] round 4/6 | labeled=112 | selection_seconds=6.3000 round_seconds=117.2000 pseudo_student_dice=0.6443 pseudo_ensemble_dice=0.5135 consistency_scale=0.4810 val_dice=0.7523 val_vi=0.5246 val_ece=0.0165 +[10:29:38][ 692s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.373, cerebellum=0.461, flywire=0.241, berghia=0.372, microns=0.648, oct-glia=0.189, oct-sfl=0.276, mouse-brain=0.293 | blended with labeled Dice at 0.50 +[10:30:25][ 738s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.557, uniform=0.367, top1=0.558, top2=0.559, top3=0.559 -> using mixture +[10:30:25][ 738s][WEIGHTS] round 5: teacher Dice on 140 labeled patches + 0 blank (1488314 fg px): c-elegans=0.360, cerebellum=0.427, flywire=0.212, berghia=0.333, microns=0.558, oct-glia=0.165, oct-sfl=0.250, mouse-brain=0.290 | weights: c-elegans=0.001, cerebellum=0.030, flywire=0.024, berghia=0.006, microns=0.934, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[10:30:29][ 743s][CONS] round 5: on the 28 new patches student Dice=0.673 vs ensemble 0.496 -> pseudo-labels from the student | consistency x0.58 +[10:30:32][ 746s][FT] round 5/6 epoch 1/5 | patches=140 | loss=0.7061 +[10:30:35][ 749s][FT] round 5/6 epoch 2/5 | patches=140 | loss=0.6878 +[10:30:38][ 751s][FT] round 5/6 epoch 3/5 | patches=140 | loss=0.6845 +[10:30:41][ 754s][FT] round 5/6 epoch 4/5 | patches=140 | loss=0.6812 +[10:30:43][ 757s][FT] round 5/6 epoch 5/5 | patches=140 | loss=0.6732 +[10:30:45][ 758s][AL] round 5/6 | labeled=140 | selection_seconds=6.4000 round_seconds=121.0000 pseudo_student_dice=0.6735 pseudo_ensemble_dice=0.4959 consistency_scale=0.5780 val_dice=0.7568 val_vi=0.5302 val_ece=0.0188 +[10:31:41][ 815s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.395, cerebellum=0.484, flywire=0.241, berghia=0.384, microns=0.645, oct-glia=0.199, oct-sfl=0.281, mouse-brain=0.315 | blended with labeled Dice at 0.50 +[10:32:28][ 862s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.560, uniform=0.374, top1=0.561, top2=0.562, top3=0.562 -> using mixture +[10:32:28][ 862s][WEIGHTS] round 6: teacher Dice on 168 labeled patches + 0 blank (1780338 fg px): c-elegans=0.374, cerebellum=0.437, flywire=0.213, berghia=0.339, microns=0.561, oct-glia=0.170, oct-sfl=0.254, mouse-brain=0.301 | weights: c-elegans=0.002, cerebellum=0.033, flywire=0.024, berghia=0.006, microns=0.932, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[10:32:33][ 867s][CONS] round 6: on the 28 new patches student Dice=0.664 vs ensemble 0.513 -> pseudo-labels from the student | consistency x0.55 +[10:32:36][ 870s][FT] round 6/6 epoch 1/5 | patches=168 | loss=0.6981 +[10:32:39][ 873s][FT] round 6/6 epoch 2/5 | patches=168 | loss=0.6750 +[10:32:43][ 876s][FT] round 6/6 epoch 3/5 | patches=168 | loss=0.6793 +[10:32:46][ 879s][FT] round 6/6 epoch 4/5 | patches=168 | loss=0.6753 +[10:32:49][ 883s][FT] round 6/6 epoch 5/5 | patches=168 | loss=0.6607 +[10:32:50][ 884s][AL] round 6/6 | labeled=168 | selection_seconds=6.3000 round_seconds=124.2000 pseudo_student_dice=0.6640 pseudo_ensemble_dice=0.5131 consistency_scale=0.5470 val_dice=0.7577 val_vi=0.5182 val_ece=0.0166 +[10:32:54][ 887s][CONS] final consolidation: pseudo-labels refreshed from the student +[10:32:57][ 890s][FT] final 1/10 epoch 1/1 | patches=168 | loss=0.6569 +[10:33:00][ 894s][FT] final 2/10 epoch 1/1 | patches=168 | loss=0.6460 +[10:33:03][ 897s][FT] final 3/10 epoch 1/1 | patches=168 | loss=0.6440 +[10:33:06][ 900s][FT] final 4/10 epoch 1/1 | patches=168 | loss=0.6361 +[10:33:10][ 903s][FT] final 5/10 epoch 1/1 | patches=168 | loss=0.6337 +[10:33:13][ 906s][FT] final 6/10 epoch 1/1 | patches=168 | loss=0.6272 +[10:33:16][ 910s][FT] final 7/10 epoch 1/1 | patches=168 | loss=0.6199 +[10:33:19][ 913s][FT] final 8/10 epoch 1/1 | patches=168 | loss=0.6179 +[10:33:22][ 916s][FT] final 9/10 epoch 1/1 | patches=168 | loss=0.6094 +[10:33:26][ 919s][FT] final 10/10 epoch 1/1 | patches=168 | loss=0.6205 +[10:33:27][ 921s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=35.2000 val_dice=0.7596 val_vi=0.5157 val_ece=0.0215 +[10:33:27][ 921s][WEIGHTS] final: c-elegans=0.002, cerebellum=0.033, flywire=0.024, berghia=0.006, microns=0.932, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003 +[10:33:37][ 931s][TEST] raw student (no EMA, no TTA): patch_dice=0.7530 patch_vi=0.5232 +[10:33:39][ 933s][TEST] final model without TTA: patch_dice=0.7530 patch_vi=0.5232 +[10:33:39][ 933s][RESOURCES] {"phases": {"setup": {"seconds": 1.16, "peak_cpu_rss_mb": 1027.9, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 25.23, "peak_cpu_rss_mb": 1809.6, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 192.42, "peak_cpu_rss_mb": 2896.9, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 700.54, "peak_cpu_rss_mb": 2896.9, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4646.0, "rounds": 6}, "inference": {"seconds": 11.74, "peak_cpu_rss_mb": 2896.9, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4646.0, "patches": 735, "images": 15, "patches_per_second": 62.6, "seconds_per_image": 0.783}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 2896.9, "total_seconds": 931.1} +[10:33:39][ 933s][TEST] patch_dice=0.7613 patch_iou=0.6242 patch_precision=0.7383 patch_recall=0.7886 patch_vi=0.5082 image_dice=0.8089 image_iou=0.6796 image_precision=0.7885 image_recall=0.8309 image_vi=0.5161 cal_ece=0.0111 cal_nll=0.1784 cal_brier=0.0565 cal_mean_confidence=0.9294 cal_accuracy=0.9183 cal_predicted_prevalence=0.2180 cal_true_prevalence=0.2059 cal_mean_probability=0.2176 +[10:33:39][ 933s][DONE] snemi3d macs-v11-norestart-acq-mcdropout b5 s3 | test patch Dice=0.7613 VI=0.5082 | image Dice=0.8089 VI=0.5161 | 933s diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/result.json b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..f4f17f5dceb1da5ed3444e7feaf310cf5b130f9b --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/result.json @@ -0,0 +1,785 @@ +{ + "target": "snemi3d", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 3, + "config": { + "task": "membrane", + "target": "snemi3d", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "mcdropout", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "snemi3d", + "--variants", + "macs-v11-norestart-acq-mcdropout", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "whole-mouse-brain" + ], + "alphas": [ + 0.00526521078604608, + 0.05775508166244069, + 0.39371182687582884, + 0.027420285915093805, + 0.47701492758994885, + 0.006835626005792793, + 0.011200163178997314, + 0.020796877985851614 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "whole-mouse-brain" + ], + "mixture": [ + 0.00526521078604608, + 0.05775508166244069, + 0.39371182687582884, + 0.027420285915093805, + 0.47701492758994885, + 0.006835626005792793, + 0.011200163178997314, + 0.020796877985851614 + ], + "per_source": [ + 0.0030099345500300633, + 0.30826329572645517, + 0.030874948568488214, + 0.005310476630139808, + 0.6243402297379833, + 0.021312995363178842, + 0.004803663858606227, + 0.0020844555651184054 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.04836304661062167, + "mmd2_mixture": 0.06012254011242163, + "mmd2_uniform": 0.15285652884935502, + "mmd2_single_source": [ + 0.4522663178230211, + 0.14701205466558376, + 0.12241636968711567, + 0.27265250499014093, + 0.08995147555294303, + 0.3744252998076626, + 0.3395783286723916, + 0.23687871798038196 + ], + "median_squared_distance": 0.18663079059841653, + "gammas": [ + 1.3395431654037109, + 2.6790863308074218, + 5.3581726616148435, + 10.716345323229687, + 21.432690646459374 + ], + "max_weight": 0.47701492758994885, + "entropy": 1.1759493501799958, + "effective_sources": 3.241218535100111, + "min_weight": 0.00526521078604608 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.46244015980269426, + 0.12211161125756897, + 0.2912827113901817, + 0.4206979810988831, + 0.07022524085432302, + 0.3185314378412357, + 0.4280722723409719, + 0.48945233449571823 + ], + "standardized": [ + 0.9323157340081956, + -1.3822022358410426, + -0.2316976633920459, + 0.6484341051420859, + -1.7350728286900672, + -0.0463835976280083, + 0.6985854358714068, + 1.1160210505294779 + ], + "tau": 0.5, + "max_weight": 0.6243402297379833, + "entropy": 0.9300685916208571, + "effective_sources": 2.534683029672754, + "min_weight": 0.0020844555651184054 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "microns_basil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.4124719184272143, + 0.4732143695257148, + 0.1963993253870582, + 0.31013775029960944, + 0.5894614335833763, + 0.13744118666208724, + 0.30544280939432356, + 0.35012843982905756 + ], + "ensemble_val_dice": { + "mixture": 0.4205090273501111, + "mixture@raw": 0.46291647296981747, + "mixture@logit": 0.3665771544587855, + "per_source": 0.5892233046139178, + "per_source@raw": 0.5905333874987907, + "per_source@logit": 0.5781025615679164, + "uniform": 0.41121937389276464, + "uniform@raw": 0.46695544697394986, + "uniform@logit": 0.22779528784246278 + }, + "val_patches": 735 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 2.6421806812286377, + 4.075904369354248 + ], + "top_source": "microns_basil", + "weight_stats": { + "max_weight": 0.47701492758994885, + "entropy": 1.1759493501799958, + "effective_sources": 3.241218535100111, + "min_weight": 0.00526521078604608 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.3571428571428572, + "spearman_weight_taxonomy": 0.50709255283711, + "top_source": "microns_basil", + "top_source_is_best_teacher": true + }, + "init_teacher": "microns_basil", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 192.4, + "round0_val": { + "val_dice": 0.3503957018118952, + "val_vi": 0.631135568604762, + "val_ece": 0.0757153987157513 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 28, + "selection_seconds": 29.4, + "round_seconds": 157.9, + "teacher_dice_labeled": [ + 0.36056026860693446, + 0.4508413766529139, + 0.21750134794876225, + 0.28356610561779927, + 0.560875393102026, + 0.18601963732494034, + 0.3724686404052985, + 0.28968887034603186 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 288982, + "evidence_nonempty_patches": 28, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.391711079681163, + "restart_teacher_dice": 0.560875393102026, + "weights": [ + 0.0013803357494966194, + 0.03734608199477827, + 0.024685983611284195, + 0.0033285842306526635, + 0.9269546251229578, + 0.000312843356437377, + 0.003307582804027515, + 0.00268396313036561 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5602980725744923, + "uniform": 0.40469001951724776, + "top1": 0.560875393102026, + "top2": 0.5626323977653401, + "top3": 0.5627847905743437 + }, + "val_dice": 0.7323560437845593, + "val_vi": 0.5694021627238198, + "val_ece": 0.011040160826576361 + }, + { + "round": 2, + "n_labeled": 56, + "selection_seconds": 6.5, + "round_seconds": 20.8, + "teacher_dice_labeled": [ + 0.3585644400140594, + 0.44569513378054804, + 0.21164106279410505, + 0.31840836229424285, + 0.44781682892290103, + 0.17125172597809413, + 0.2640817814474379, + 0.30486247967348346 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 636921, + "evidence_nonempty_patches": 56, + "first_correction": false, + "restart": "", + "weights": [ + 0.004025912650355965, + 0.10554619536395982, + 0.0, + 0.0, + 0.8904278919856843, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.4444176386674499, + "uniform": 0.32484691123850307, + "top1": 0.44781682892290103, + "top2": 0.45371291795123353, + "top3": 0.45396144205152433 + }, + "pseudo_student_dice": 0.6084848613600118, + "pseudo_ensemble_dice": 0.31824112897528706, + "pseudo_source": "student", + "consistency_scale": 0.362, + "val_dice": 0.7421486412029648, + "val_vi": 0.556016678738807, + "val_ece": 0.015960624579132065 + }, + { + "round": 3, + "n_labeled": 84, + "selection_seconds": 6.4, + "round_seconds": 114.0, + "teacher_dice_labeled": [ + 0.3772306767830903, + 0.45163488231981797, + 0.2206921454618092, + 0.34409742181789726, + 0.5617200882449684, + 0.17804109167411933, + 0.2511611799133086, + 0.3060860498014537 + ], + "teacher_dice_pool": [ + 0.39700488397740485, + 0.4883830214540865, + 0.24298449902950017, + 0.38680546140851324, + 0.6801822253345543, + 0.19701607358506548, + 0.2778733913497543, + 0.31410917782810194 + ], + "evidence_fg_pixels": 943818, + "evidence_nonempty_patches": 84, + "first_correction": false, + "restart": "", + "weights": [ + 0.0016144130645445057, + 0.03726683086288565, + 0.02523127275765228, + 0.006036357907624515, + 0.9254611767492045, + 0.00028596166549658167, + 0.000973442678300187, + 0.003130544314291847 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5608050244269053, + "uniform": 0.3759099719289151, + "top1": 0.5617200882449684, + "top2": 0.563139067283982, + "top3": 0.5632320431152088 + }, + "pseudo_student_dice": 0.6450631392491858, + "pseudo_ensemble_dice": 0.4328560747160571, + "pseudo_source": "student", + "consistency_scale": 0.484, + "val_dice": 0.7503492358738133, + "val_vi": 0.5510445083549702, + "val_ece": 0.017044766730611508 + }, + { + "round": 4, + "n_labeled": 112, + "selection_seconds": 6.3, + "round_seconds": 117.2, + "teacher_dice_labeled": [ + 0.39048865645881586, + 0.4491344212222539, + 0.21475358314891385, + 0.3369891850657348, + 0.5589963369618478, + 0.1701515660711788, + 0.2394433322730992, + 0.31370029514801345 + ], + "teacher_dice_pool": [ + 0.4315052627319388, + 0.506103214858209, + 0.236099460618685, + 0.3860519184665265, + 0.6576926480768255, + 0.19304244666447792, + 0.26686329163719663, + 0.3427654718654806 + ], + "evidence_fg_pixels": 1182224, + "evidence_nonempty_patches": 112, + "first_correction": false, + "restart": "", + "weights": [ + 0.0018950403875757182, + 0.037366999859284565, + 0.024444065719803742, + 0.005780021245019906, + 0.9258790041630323, + 0.0002716871988332553, + 0.0008901138320830429, + 0.0034730675943676205 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.557933292560871, + "uniform": 0.3623408682867729, + "top1": 0.5589963369618478, + "top2": 0.5602149432711729, + "top3": 0.5603361268165685 + }, + "pseudo_student_dice": 0.6443473735400261, + "pseudo_ensemble_dice": 0.5134618975328987, + "pseudo_source": "student", + "consistency_scale": 0.481, + "val_dice": 0.7522883045571649, + "val_vi": 0.524641625228006, + "val_ece": 0.016537528590195295 + }, + { + "round": 5, + "n_labeled": 140, + "selection_seconds": 6.4, + "round_seconds": 121.0, + "teacher_dice_labeled": [ + 0.36001979030446934, + 0.4266018157765912, + 0.21166665891913206, + 0.3326090449700532, + 0.5582016752054481, + 0.16480849083774873, + 0.25026909107295403, + 0.28984831642964815 + ], + "teacher_dice_pool": [ + 0.37284183823210754, + 0.4612999481718947, + 0.24096947941714422, + 0.37192028122390125, + 0.6479905837255117, + 0.1889133895654166, + 0.27553571279748673, + 0.2933843567028957 + ], + "evidence_fg_pixels": 1488314, + "evidence_nonempty_patches": 140, + "first_correction": false, + "restart": "", + "weights": [ + 0.0014214857178205984, + 0.03034443133386405, + 0.024111028469058503, + 0.005628016509243339, + 0.9344405975254152, + 0.0002620071420808458, + 0.0010090434270227865, + 0.0027833898754947807 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5570908624941996, + "uniform": 0.3666543830200925, + "top1": 0.5582016752054481, + "top2": 0.5591297232748379, + "top3": 0.5591888606233547 + }, + "pseudo_student_dice": 0.6734744719434842, + "pseudo_ensemble_dice": 0.4958697336902917, + "pseudo_source": "student", + "consistency_scale": 0.578, + "val_dice": 0.7568385587791026, + "val_vi": 0.5301980889611158, + "val_ece": 0.018759114485960196 + }, + { + "round": 6, + "n_labeled": 168, + "selection_seconds": 6.3, + "round_seconds": 124.2, + "teacher_dice_labeled": [ + 0.37408712100817315, + 0.4371792712153743, + 0.21264084317704324, + 0.33868259877237294, + 0.5607169060423229, + 0.16997856242071668, + 0.2538120015052483, + 0.3009675653092333 + ], + "teacher_dice_pool": [ + 0.39522086924433447, + 0.48409444015218395, + 0.24088776217932312, + 0.3843698480549815, + 0.6450175213456133, + 0.19881484436170063, + 0.28143892579597607, + 0.3146631055005121 + ], + "evidence_fg_pixels": 1780338, + "evidence_nonempty_patches": 168, + "first_correction": false, + "restart": "", + "weights": [ + 0.0015910641009763598, + 0.03279954685573042, + 0.023675442220052585, + 0.005815459119122665, + 0.931808660430251, + 0.00026829834944663096, + 0.0010165952065912859, + 0.003024933717829064 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.55973701472009, + "uniform": 0.37394979224484814, + "top1": 0.5607169060423229, + "top2": 0.5617850814822273, + "top3": 0.5618683450323583 + }, + "pseudo_student_dice": 0.6640267626039819, + "pseudo_ensemble_dice": 0.5130756838160614, + "pseudo_source": "student", + "consistency_scale": 0.547, + "val_dice": 0.7577150031400584, + "val_vi": 0.5181974799383757, + "val_ece": 0.01662966655423441 + }, + { + "round": 7, + "n_labeled": 168, + "selection_seconds": 0.0, + "round_seconds": 35.2, + "final_consolidation": true, + "val_dice": 0.759555873492989, + "val_vi": 0.515686209770237, + "val_ece": 0.02146679764354789 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.4795522004497552, + "auroc_top20_error": 0.8258676231842175 + }, + "al_seconds": 700.5, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.0015910641009763598, + 0.03279954685573042, + 0.023675442220052585, + 0.005815459119122665, + 0.931808660430251, + 0.00026829834944663096, + 0.0010165952065912859, + 0.003024933717829064 + ], + "weight_stats_final": { + "max_weight": 0.931808660430251, + "entropy": 0.333468035026498, + "effective_sources": 1.395800429104696, + "min_weight": 0.00026829834944663096 + }, + "top_source_final": "microns_basil", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7530073262298932, + "patch_iou": 0.6136276658297826, + "patch_precision": 0.7270609913951437, + "patch_recall": 0.7839162274714548, + "patch_vi": 0.5231999082442164, + "n_patches": 735, + "cal_ece": 0.019476051786440497, + "cal_nll": 0.19016047656597235, + "cal_brier": 0.05974153790932348, + "cal_mean_confidence": 0.9343915492354201, + "cal_accuracy": 0.9149154974489796, + "cal_predicted_prevalence": 0.22031914328231292, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.21871572358961278 + }, + "test_notta": { + "patch_dice": 0.7530073262298932, + "patch_iou": 0.6136276658297826, + "patch_precision": 0.7270609913951437, + "patch_recall": 0.7839162274714548, + "patch_vi": 0.5231999082442164, + "n_patches": 735, + "cal_ece": 0.019476051786440497, + "cal_nll": 0.19016047656597235, + "cal_brier": 0.05974153790932348, + "cal_mean_confidence": 0.9343915492354201, + "cal_accuracy": 0.9149154974489796, + "cal_predicted_prevalence": 0.22031914328231292, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.21871572358961278 + }, + "test": { + "patch_dice": 0.7612862326469151, + "patch_iou": 0.6241756457763458, + "patch_precision": 0.7382643458817646, + "patch_recall": 0.7886357133088496, + "patch_vi": 0.5082244541709398, + "n_patches": 735, + "image_dice": 0.8089188850134597, + "image_iou": 0.6795995339893497, + "image_precision": 0.7884908624044854, + "image_recall": 0.8308934273184174, + "image_vi": 0.5161480607947462, + "n_images": 15, + "cal_ece": 0.011077641126509934, + "cal_nll": 0.17836683789955543, + "cal_brier": 0.056488108008727, + "cal_mean_confidence": 0.9293559680822921, + "cal_accuracy": 0.9182783269557823, + "cal_predicted_prevalence": 0.2179793792517007, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.21756974344637195 + }, + "eval_seconds": 11.7, + "total_seconds": 932.7, + "resources": { + "phases": { + "setup": { + "seconds": 1.16, + "peak_cpu_rss_mb": 1027.9, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 25.23, + "peak_cpu_rss_mb": 1809.6, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 192.42, + "peak_cpu_rss_mb": 2896.9, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4638.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 700.54, + "peak_cpu_rss_mb": 2896.9, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4646.0, + "rounds": 6 + }, + "inference": { + "seconds": 11.74, + "peak_cpu_rss_mb": 2896.9, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4646.0, + "patches": 735, + "images": 15, + "patches_per_second": 62.6, + "seconds_per_image": 0.783 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 2896.9, + "total_seconds": 931.1 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/snemi3d/macs-v11-norestart-acq-mcdropout_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..081fe8a86e49f22599c8094da2980b36e33d359d --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,28,29.4,157.9,"[0.36056026860693446, 0.4508413766529139, 0.21750134794876225, 0.28356610561779927, 0.560875393102026, 0.18601963732494034, 0.3724686404052985, 0.28968887034603186]",,288982,28,True,kept,0.391711079681163,0.560875393102026,"[0.0013803357494966194, 0.03734608199477827, 0.024685983611284195, 0.0033285842306526635, 0.9269546251229578, 0.000312843356437377, 0.003307582804027515, 0.00268396313036561]",mixture,"{'mixture': 0.5602980725744923, 'uniform': 0.40469001951724776, 'top1': 0.560875393102026, 'top2': 0.5626323977653401, 'top3': 0.5627847905743437}",0.7323560437845593,0.5694021627238198,0.011040160826576361,,,,, +2,56,6.5,20.8,"[0.3585644400140594, 0.44569513378054804, 0.21164106279410505, 0.31840836229424285, 0.44781682892290103, 0.17125172597809413, 0.2640817814474379, 0.30486247967348346]",,636921,56,False,,,,"[0.004025912650355965, 0.10554619536395982, 0.0, 0.0, 0.8904278919856843, 0.0, 0.0, 0.0]",top3,"{'mixture': 0.4444176386674499, 'uniform': 0.32484691123850307, 'top1': 0.44781682892290103, 'top2': 0.45371291795123353, 'top3': 0.45396144205152433}",0.7421486412029648,0.556016678738807,0.015960624579132065,0.6084848613600118,0.31824112897528706,student,0.362, +3,84,6.4,114.0,"[0.3772306767830903, 0.45163488231981797, 0.2206921454618092, 0.34409742181789726, 0.5617200882449684, 0.17804109167411933, 0.2511611799133086, 0.3060860498014537]","[0.39700488397740485, 0.4883830214540865, 0.24298449902950017, 0.38680546140851324, 0.6801822253345543, 0.19701607358506548, 0.2778733913497543, 0.31410917782810194]",943818,84,False,,,,"[0.0016144130645445057, 0.03726683086288565, 0.02523127275765228, 0.006036357907624515, 0.9254611767492045, 0.00028596166549658167, 0.000973442678300187, 0.003130544314291847]",mixture,"{'mixture': 0.5608050244269053, 'uniform': 0.3759099719289151, 'top1': 0.5617200882449684, 'top2': 0.563139067283982, 'top3': 0.5632320431152088}",0.7503492358738133,0.5510445083549702,0.017044766730611508,0.6450631392491858,0.4328560747160571,student,0.484, +4,112,6.3,117.2,"[0.39048865645881586, 0.4491344212222539, 0.21475358314891385, 0.3369891850657348, 0.5589963369618478, 0.1701515660711788, 0.2394433322730992, 0.31370029514801345]","[0.4315052627319388, 0.506103214858209, 0.236099460618685, 0.3860519184665265, 0.6576926480768255, 0.19304244666447792, 0.26686329163719663, 0.3427654718654806]",1182224,112,False,,,,"[0.0018950403875757182, 0.037366999859284565, 0.024444065719803742, 0.005780021245019906, 0.9258790041630323, 0.0002716871988332553, 0.0008901138320830429, 0.0034730675943676205]",mixture,"{'mixture': 0.557933292560871, 'uniform': 0.3623408682867729, 'top1': 0.5589963369618478, 'top2': 0.5602149432711729, 'top3': 0.5603361268165685}",0.7522883045571649,0.524641625228006,0.016537528590195295,0.6443473735400261,0.5134618975328987,student,0.481, +5,140,6.4,121.0,"[0.36001979030446934, 0.4266018157765912, 0.21166665891913206, 0.3326090449700532, 0.5582016752054481, 0.16480849083774873, 0.25026909107295403, 0.28984831642964815]","[0.37284183823210754, 0.4612999481718947, 0.24096947941714422, 0.37192028122390125, 0.6479905837255117, 0.1889133895654166, 0.27553571279748673, 0.2933843567028957]",1488314,140,False,,,,"[0.0014214857178205984, 0.03034443133386405, 0.024111028469058503, 0.005628016509243339, 0.9344405975254152, 0.0002620071420808458, 0.0010090434270227865, 0.0027833898754947807]",mixture,"{'mixture': 0.5570908624941996, 'uniform': 0.3666543830200925, 'top1': 0.5582016752054481, 'top2': 0.5591297232748379, 'top3': 0.5591888606233547}",0.7568385587791026,0.5301980889611158,0.018759114485960196,0.6734744719434842,0.4958697336902917,student,0.578, +6,168,6.3,124.2,"[0.37408712100817315, 0.4371792712153743, 0.21264084317704324, 0.33868259877237294, 0.5607169060423229, 0.16997856242071668, 0.2538120015052483, 0.3009675653092333]","[0.39522086924433447, 0.48409444015218395, 0.24088776217932312, 0.3843698480549815, 0.6450175213456133, 0.19881484436170063, 0.28143892579597607, 0.3146631055005121]",1780338,168,False,,,,"[0.0015910641009763598, 0.03279954685573042, 0.023675442220052585, 0.005815459119122665, 0.931808660430251, 0.00026829834944663096, 0.0010165952065912859, 0.003024933717829064]",mixture,"{'mixture': 0.55973701472009, 'uniform': 0.37394979224484814, 'top1': 0.5607169060423229, 'top2': 0.5617850814822273, 'top3': 0.5618683450323583}",0.7577150031400584,0.5181974799383757,0.01662966655423441,0.6640267626039819,0.5130756838160614,student,0.547, +7,168,0.0,35.2,,,,,,,,,,,,0.759555873492989,0.515686209770237,0.02146679764354789,,,,,True diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/log.txt b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..fae79178f071bf2dcba7aaffb9f2a45fcae1d871 --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/log.txt @@ -0,0 +1,103 @@ +[10:33:41][ 0s][START] target=snemi3d variant=macs-v11-norestart-acq-random budget=5.0% seed=3 | {"task": "membrane", "target": "snemi3d", "variant": "macs-v11-norestart-acq-random", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "random", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 0, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[10:33:41][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[10:33:42][ 0s][DATA] snemi3d: train=3430 val=735 test=735 patches +[10:33:43][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, oct-sfl=2.64, mouse-brain=4.08 +[10:33:43][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[10:33:49][ 7s][WEIGHTS] mixture: c-elegans=0.005, cerebellum=0.058, flywire=0.394, berghia=0.027, microns=0.477, oct-glia=0.007, oct-sfl=0.011, mouse-brain=0.021 | eff.sources=3.24 | MMD2 mix=0.0601 uniform=0.1529 +[10:33:49][ 7s][WEIGHTS] per-source: c-elegans=0.003, cerebellum=0.308, flywire=0.031, berghia=0.005, microns=0.624, oct-glia=0.021, oct-sfl=0.005, mouse-brain=0.002 | min-MMD source (NeuroADDA rule) = microns +[10:34:07][ 25s][TEACHERS] target-val Dice: c-elegans=0.412, cerebellum=0.473, flywire=0.196, berghia=0.310, microns=0.589, oct-glia=0.137, oct-sfl=0.305, mouse-brain=0.350 +[10:34:07][ 25s][TEACHERS] ensemble target-val Dice: mixture=0.421, mixture@raw=0.463, mixture@logit=0.367, per_source=0.589, per_source@raw=0.591, per_source@logit=0.578, uniform=0.411, uniform@raw=0.467, uniform@logit=0.228 +[10:34:07][ 25s][WEIGHTS] using mixture: c-elegans=0.005, cerebellum=0.058, flywire=0.394, berghia=0.027, microns=0.477, oct-glia=0.007, oct-sfl=0.011, mouse-brain=0.021 +[10:34:07][ 25s][INIT] student initialised from teacher microns (weight 0.477) +[10:34:08][ 26s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[10:34:08][ 26s][KD] distilling the mixture ensemble (prob space) into the student +[10:34:47][ 65s][KD] soft labels for 3430 patches from 8 teachers in 39s +[10:35:18][ 97s][KD] epoch 1/5 | patches=3430 | loss=0.4446 | 31s +[10:35:49][ 127s][KD] epoch 2/5 | patches=3430 | loss=0.3724 | 31s +[10:36:20][ 158s][KD] epoch 3/5 | patches=3430 | loss=0.3568 | 31s +[10:36:50][ 189s][KD] epoch 4/5 | patches=3430 | loss=0.3541 | 31s +[10:37:20][ 219s][KD] epoch 5/5 | patches=3430 | loss=0.3478 | 30s +[10:37:22][ 220s][ROUND0] before any target label: val_dice=0.3580 val_vi=0.6345 +[10:37:22][ 220s][AL] patches=3430 (pool 3430 after dropping blank; free blank negatives: True) | budget=171 (5.0%) | rounds=6 | per_round=28 | acquisition=random | first_round=stratified +[10:37:27][ 226s][AL] round-1 uncertainty quality on 3430 patches: {'spearman_uncertainty_error': 0.005164451042246618, 'auroc_top20_error': 0.5020564348188254} +[10:37:50][ 249s][AL] round 1: stratified cold start: 14 patches from the membrane-rich half (ensemble prevalence >= 0.156) + 14 from the membrane-poor half of 2000 candidates +[10:37:52][ 251s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.560, uniform=0.405, top1=0.561, top2=0.563, top3=0.563 -> using mixture +[10:37:53][ 251s][WEIGHTS] round 1: student kept (labeled Dice 0.402 vs best teacher microns 0.561); distilling the corrected ensemble into it +[10:37:53][ 251s][WEIGHTS] round 1: teacher Dice on 28 labeled patches + 0 blank (288982 fg px): c-elegans=0.361, cerebellum=0.451, flywire=0.218, berghia=0.284, microns=0.561, oct-glia=0.186, oct-sfl=0.372, mouse-brain=0.290 | weights: c-elegans=0.001, cerebellum=0.037, flywire=0.025, berghia=0.003, microns=0.927, oct-glia=0.000, oct-sfl=0.003, mouse-brain=0.003 +[10:38:31][ 290s][KD] soft labels for 3402 patches from 8 teachers in 39s +[10:39:01][ 320s][KD] epoch 1/2 | patches=3402 | loss=0.2481 | 30s +[10:39:32][ 350s][KD] epoch 2/2 | patches=3402 | loss=0.2479 | 30s +[10:39:54][ 373s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[10:39:55][ 374s][FT] round 1/6 epoch 1/5 | patches=28 | loss=1.0342 +[10:39:56][ 375s][FT] round 1/6 epoch 2/5 | patches=28 | loss=0.9446 +[10:39:57][ 375s][FT] round 1/6 epoch 3/5 | patches=28 | loss=0.8956 +[10:39:58][ 376s][FT] round 1/6 epoch 4/5 | patches=28 | loss=0.8636 +[10:39:59][ 377s][FT] round 1/6 epoch 5/5 | patches=28 | loss=0.8517 +[10:40:00][ 379s][AL] round 1/6 | labeled=28 | selection_seconds=28.4000 round_seconds=156.9000 restart_student_dice=0.4024 restart_teacher_dice=0.5609 val_dice=0.7275 val_vi=0.5750 val_ece=0.0118 +[10:40:03][ 382s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.557, uniform=0.374, top1=0.559, top2=0.559, top3=0.559 -> using mixture +[10:40:03][ 382s][WEIGHTS] round 2: teacher Dice on 56 labeled patches + 0 blank (565126 fg px): c-elegans=0.348, cerebellum=0.419, flywire=0.206, berghia=0.301, microns=0.559, oct-glia=0.176, oct-sfl=0.325, mouse-brain=0.283 | weights: c-elegans=0.001, cerebellum=0.028, flywire=0.023, berghia=0.004, microns=0.939, oct-glia=0.000, oct-sfl=0.002, mouse-brain=0.003 +[10:40:08][ 387s][CONS] round 2: on the 28 new patches student Dice=0.685 vs ensemble 0.554 -> pseudo-labels from the student | consistency x0.62 +[10:40:10][ 388s][FT] round 2/6 epoch 1/5 | patches=56 | loss=0.7638 +[10:40:11][ 389s][FT] round 2/6 epoch 2/5 | patches=56 | loss=0.7486 +[10:40:12][ 391s][FT] round 2/6 epoch 3/5 | patches=56 | loss=0.7512 +[10:40:14][ 392s][FT] round 2/6 epoch 4/5 | patches=56 | loss=0.7207 +[10:40:15][ 393s][FT] round 2/6 epoch 5/5 | patches=56 | loss=0.7258 +[10:40:16][ 395s][AL] round 2/6 | labeled=56 | selection_seconds=0.2000 round_seconds=14.6000 pseudo_student_dice=0.6854 pseudo_ensemble_dice=0.5543 consistency_scale=0.6180 val_dice=0.7399 val_vi=0.5367 val_ece=0.0128 +[10:41:05][ 443s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.362, cerebellum=0.462, flywire=0.235, berghia=0.363, microns=0.664, oct-glia=0.194, oct-sfl=0.281, mouse-brain=0.290 | blended with labeled Dice at 0.50 +[10:41:50][ 488s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.615, uniform=0.410, top1=0.616, top2=0.616, top3=0.616 -> using mixture +[10:41:50][ 488s][WEIGHTS] round 3: teacher Dice on 84 labeled patches + 0 blank (846975 fg px): c-elegans=0.356, cerebellum=0.441, flywire=0.220, berghia=0.336, microns=0.616, oct-glia=0.181, oct-sfl=0.291, mouse-brain=0.284 | weights: c-elegans=0.001, cerebellum=0.020, flywire=0.015, berghia=0.003, microns=0.958, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.002 +[10:41:55][ 493s][CONS] round 3: on the 28 new patches student Dice=0.717 vs ensemble 0.582 -> pseudo-labels from the student | consistency x0.72 +[10:41:57][ 495s][FT] round 3/6 epoch 1/5 | patches=84 | loss=0.7123 +[10:41:58][ 497s][FT] round 3/6 epoch 2/5 | patches=84 | loss=0.6938 +[10:42:00][ 499s][FT] round 3/6 epoch 3/5 | patches=84 | loss=0.6929 +[10:42:02][ 501s][FT] round 3/6 epoch 4/5 | patches=84 | loss=0.6823 +[10:42:04][ 503s][FT] round 3/6 epoch 5/5 | patches=84 | loss=0.6725 +[10:42:06][ 504s][AL] round 3/6 | labeled=84 | selection_seconds=0.2000 round_seconds=107.8000 pseudo_student_dice=0.7165 pseudo_ensemble_dice=0.5822 consistency_scale=0.7220 val_dice=0.7494 val_vi=0.5370 val_ece=0.0221 +[10:42:54][ 553s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.388, cerebellum=0.473, flywire=0.236, berghia=0.386, microns=0.654, oct-glia=0.192, oct-sfl=0.269, mouse-brain=0.306 | blended with labeled Dice at 0.50 +[10:43:40][ 599s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.613, uniform=0.409, top1=0.613, top2=0.614, top3=0.614 -> using mixture +[10:43:40][ 599s][WEIGHTS] round 4: teacher Dice on 112 labeled patches + 0 blank (1109045 fg px): c-elegans=0.368, cerebellum=0.443, flywire=0.222, berghia=0.351, microns=0.613, oct-glia=0.183, oct-sfl=0.282, mouse-brain=0.291 | weights: c-elegans=0.001, cerebellum=0.021, flywire=0.016, berghia=0.004, microns=0.955, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.002 +[10:43:45][ 604s][CONS] round 4: on the 28 new patches student Dice=0.723 vs ensemble 0.588 -> pseudo-labels from the student | consistency x0.74 +[10:43:48][ 606s][FT] round 4/6 epoch 1/5 | patches=112 | loss=0.6614 +[10:43:50][ 608s][FT] round 4/6 epoch 2/5 | patches=112 | loss=0.6542 +[10:43:52][ 611s][FT] round 4/6 epoch 3/5 | patches=112 | loss=0.6548 +[10:43:54][ 613s][FT] round 4/6 epoch 4/5 | patches=112 | loss=0.6452 +[10:43:57][ 615s][FT] round 4/6 epoch 5/5 | patches=112 | loss=0.6399 +[10:43:58][ 617s][AL] round 4/6 | labeled=112 | selection_seconds=0.2000 round_seconds=111.1000 pseudo_student_dice=0.7229 pseudo_ensemble_dice=0.5877 consistency_scale=0.7430 val_dice=0.7508 val_vi=0.5354 val_ece=0.0242 +[10:44:48][ 666s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.384, cerebellum=0.470, flywire=0.242, berghia=0.380, microns=0.652, oct-glia=0.192, oct-sfl=0.281, mouse-brain=0.304 | blended with labeled Dice at 0.50 +[10:45:34][ 713s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.610, uniform=0.408, top1=0.611, top2=0.611, top3=0.611 -> using mixture +[10:45:34][ 713s][WEIGHTS] round 5: teacher Dice on 140 labeled patches + 0 blank (1372470 fg px): c-elegans=0.365, cerebellum=0.440, flywire=0.225, berghia=0.348, microns=0.611, oct-glia=0.181, oct-sfl=0.277, mouse-brain=0.288 | weights: c-elegans=0.001, cerebellum=0.021, flywire=0.017, berghia=0.004, microns=0.955, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.002 +[10:45:39][ 718s][CONS] round 5: on the 28 new patches student Dice=0.708 vs ensemble 0.558 -> pseudo-labels from the student | consistency x0.69 +[10:45:42][ 721s][FT] round 5/6 epoch 1/5 | patches=140 | loss=0.6426 +[10:45:45][ 723s][FT] round 5/6 epoch 2/5 | patches=140 | loss=0.6381 +[10:45:48][ 726s][FT] round 5/6 epoch 3/5 | patches=140 | loss=0.6368 +[10:45:50][ 729s][FT] round 5/6 epoch 4/5 | patches=140 | loss=0.6257 +[10:45:53][ 732s][FT] round 5/6 epoch 5/5 | patches=140 | loss=0.6263 +[10:45:55][ 733s][AL] round 5/6 | labeled=140 | selection_seconds=0.2000 round_seconds=115.0000 pseudo_student_dice=0.7077 pseudo_ensemble_dice=0.5583 consistency_scale=0.6920 val_dice=0.7525 val_vi=0.5317 val_ece=0.0236 +[10:46:45][ 784s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.377, cerebellum=0.466, flywire=0.243, berghia=0.380, microns=0.646, oct-glia=0.198, oct-sfl=0.283, mouse-brain=0.299 | blended with labeled Dice at 0.50 +[10:47:32][ 831s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.607, uniform=0.407, top1=0.608, top2=0.608, top3=0.608 -> using mixture +[10:47:32][ 831s][WEIGHTS] round 6: teacher Dice on 168 labeled patches + 0 blank (1676005 fg px): c-elegans=0.363, cerebellum=0.440, flywire=0.224, berghia=0.346, microns=0.608, oct-glia=0.184, oct-sfl=0.278, mouse-brain=0.287 | weights: c-elegans=0.001, cerebellum=0.022, flywire=0.017, berghia=0.004, microns=0.954, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.002 +[10:47:37][ 835s][CONS] round 6: on the 28 new patches student Dice=0.757 vs ensemble 0.564 -> pseudo-labels from the student | consistency x0.86 +[10:47:40][ 839s][FT] round 6/6 epoch 1/5 | patches=168 | loss=0.6402 +[10:47:43][ 842s][FT] round 6/6 epoch 2/5 | patches=168 | loss=0.6324 +[10:47:47][ 845s][FT] round 6/6 epoch 3/5 | patches=168 | loss=0.6399 +[10:47:50][ 848s][FT] round 6/6 epoch 4/5 | patches=168 | loss=0.6291 +[10:47:53][ 851s][FT] round 6/6 epoch 5/5 | patches=168 | loss=0.6404 +[10:47:54][ 853s][AL] round 6/6 | labeled=168 | selection_seconds=0.2000 round_seconds=118.4000 pseudo_student_dice=0.7570 pseudo_ensemble_dice=0.5641 consistency_scale=0.8570 val_dice=0.7530 val_vi=0.5240 val_ece=0.0258 +[10:47:58][ 856s][CONS] final consolidation: pseudo-labels refreshed from the student +[10:48:01][ 859s][FT] final 1/10 epoch 1/1 | patches=168 | loss=0.6136 +[10:48:04][ 863s][FT] final 2/10 epoch 1/1 | patches=168 | loss=0.6131 +[10:48:07][ 866s][FT] final 3/10 epoch 1/1 | patches=168 | loss=0.6124 +[10:48:10][ 869s][FT] final 4/10 epoch 1/1 | patches=168 | loss=0.6099 +[10:48:14][ 872s][FT] final 5/10 epoch 1/1 | patches=168 | loss=0.5997 +[10:48:17][ 875s][FT] final 6/10 epoch 1/1 | patches=168 | loss=0.5997 +[10:48:20][ 879s][FT] final 7/10 epoch 1/1 | patches=168 | loss=0.5953 +[10:48:23][ 882s][FT] final 8/10 epoch 1/1 | patches=168 | loss=0.5949 +[10:48:26][ 885s][FT] final 9/10 epoch 1/1 | patches=168 | loss=0.5923 +[10:48:30][ 888s][FT] final 10/10 epoch 1/1 | patches=168 | loss=0.5925 +[10:48:31][ 890s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=35.2000 val_dice=0.7561 val_vi=0.5234 val_ece=0.0284 +[10:48:31][ 890s][WEIGHTS] final: c-elegans=0.001, cerebellum=0.022, flywire=0.017, berghia=0.004, microns=0.954, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.002 +[10:48:41][ 900s][TEST] raw student (no EMA, no TTA): patch_dice=0.7512 patch_vi=0.5222 +[10:48:43][ 901s][TEST] final model without TTA: patch_dice=0.7512 patch_vi=0.5222 +[10:48:43][ 901s][RESOURCES] {"phases": {"setup": {"seconds": 1.14, "peak_cpu_rss_mb": 1027.6, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 24.99, "peak_cpu_rss_mb": 1810.2, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 192.59, "peak_cpu_rss_mb": 2896.7, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 669.28, "peak_cpu_rss_mb": 2896.7, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4646.0, "rounds": 6}, "inference": {"seconds": 11.6, "peak_cpu_rss_mb": 2896.7, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4646.0, "patches": 735, "images": 15, "patches_per_second": 63.4, "seconds_per_image": 0.773}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 2896.7, "total_seconds": 899.6} +[10:48:43][ 901s][TEST] patch_dice=0.7595 patch_iou=0.6221 patch_precision=0.7419 patch_recall=0.7813 patch_vi=0.5086 image_dice=0.8065 image_iou=0.6762 image_precision=0.7927 image_recall=0.8212 image_vi=0.5177 cal_ece=0.0171 cal_nll=0.1809 cal_brier=0.0571 cal_mean_confidence=0.9353 cal_accuracy=0.9183 cal_predicted_prevalence=0.2149 cal_true_prevalence=0.2059 cal_mean_probability=0.2135 +[10:48:43][ 901s][DONE] snemi3d macs-v11-norestart-acq-random b5 s3 | test patch Dice=0.7595 VI=0.5086 | image Dice=0.8065 VI=0.5177 | 901s diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/result.json b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..a2de2a4e18af88837ddb23e669a83cd717c9ba02 --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/result.json @@ -0,0 +1,785 @@ +{ + "target": "snemi3d", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 3, + "config": { + "task": "membrane", + "target": "snemi3d", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "random", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 0, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "snemi3d", + "--variants", + "macs-v11-norestart-acq-random", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "whole-mouse-brain" + ], + "alphas": [ + 0.00526521078604608, + 0.05775508166244069, + 0.39371182687582884, + 0.027420285915093805, + 0.47701492758994885, + 0.006835626005792793, + 0.011200163178997314, + 0.020796877985851614 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "whole-mouse-brain" + ], + "mixture": [ + 0.00526521078604608, + 0.05775508166244069, + 0.39371182687582884, + 0.027420285915093805, + 0.47701492758994885, + 0.006835626005792793, + 0.011200163178997314, + 0.020796877985851614 + ], + "per_source": [ + 0.0030099345500300633, + 0.30826329572645517, + 0.030874948568488214, + 0.005310476630139808, + 0.6243402297379833, + 0.021312995363178842, + 0.004803663858606227, + 0.0020844555651184054 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.04836304661062167, + "mmd2_mixture": 0.06012254011242163, + "mmd2_uniform": 0.15285652884935502, + "mmd2_single_source": [ + 0.4522663178230211, + 0.14701205466558376, + 0.12241636968711567, + 0.27265250499014093, + 0.08995147555294303, + 0.3744252998076626, + 0.3395783286723916, + 0.23687871798038196 + ], + "median_squared_distance": 0.18663079059841653, + "gammas": [ + 1.3395431654037109, + 2.6790863308074218, + 5.3581726616148435, + 10.716345323229687, + 21.432690646459374 + ], + "max_weight": 0.47701492758994885, + "entropy": 1.1759493501799958, + "effective_sources": 3.241218535100111, + "min_weight": 0.00526521078604608 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.46244015980269426, + 0.12211161125756897, + 0.2912827113901817, + 0.4206979810988831, + 0.07022524085432302, + 0.3185314378412357, + 0.4280722723409719, + 0.48945233449571823 + ], + "standardized": [ + 0.9323157340081956, + -1.3822022358410426, + -0.2316976633920459, + 0.6484341051420859, + -1.7350728286900672, + -0.0463835976280083, + 0.6985854358714068, + 1.1160210505294779 + ], + "tau": 0.5, + "max_weight": 0.6243402297379833, + "entropy": 0.9300685916208571, + "effective_sources": 2.534683029672754, + "min_weight": 0.0020844555651184054 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "microns_basil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.4124719184272143, + 0.4732143695257148, + 0.1963993253870582, + 0.31013775029960944, + 0.5894614335833763, + 0.13744118666208724, + 0.30544280939432356, + 0.35012843982905756 + ], + "ensemble_val_dice": { + "mixture": 0.4205090273501111, + "mixture@raw": 0.46291647296981747, + "mixture@logit": 0.3665771544587855, + "per_source": 0.5892233046139178, + "per_source@raw": 0.5905333874987907, + "per_source@logit": 0.5781025615679164, + "uniform": 0.41121937389276464, + "uniform@raw": 0.46695544697394986, + "uniform@logit": 0.22779528784246278 + }, + "val_patches": 735 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 2.6421806812286377, + 4.075904369354248 + ], + "top_source": "microns_basil", + "weight_stats": { + "max_weight": 0.47701492758994885, + "entropy": 1.1759493501799958, + "effective_sources": 3.241218535100111, + "min_weight": 0.00526521078604608 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.3571428571428572, + "spearman_weight_taxonomy": 0.50709255283711, + "top_source": "microns_basil", + "top_source_is_best_teacher": true + }, + "init_teacher": "microns_basil", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 192.6, + "round0_val": { + "val_dice": 0.35801025103979234, + "val_vi": 0.6344848740811435, + "val_ece": 0.07478441083095859 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 28, + "selection_seconds": 28.4, + "round_seconds": 156.9, + "teacher_dice_labeled": [ + 0.36056026860693446, + 0.4508413766529139, + 0.21750134794876225, + 0.28356610561779927, + 0.560875393102026, + 0.18601963732494034, + 0.3724686404052985, + 0.28968887034603186 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 288982, + "evidence_nonempty_patches": 28, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.4024189407361833, + "restart_teacher_dice": 0.560875393102026, + "weights": [ + 0.0013803357494966194, + 0.03734608199477827, + 0.024685983611284195, + 0.0033285842306526635, + 0.9269546251229578, + 0.000312843356437377, + 0.003307582804027515, + 0.00268396313036561 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5602980725744923, + "uniform": 0.40469001951724776, + "top1": 0.560875393102026, + "top2": 0.5626323977653401, + "top3": 0.5627847905743437 + }, + "val_dice": 0.727546285788197, + "val_vi": 0.5749785964732628, + "val_ece": 0.011788481275023454 + }, + { + "round": 2, + "n_labeled": 56, + "selection_seconds": 0.2, + "round_seconds": 14.6, + "teacher_dice_labeled": [ + 0.3479823236552919, + 0.4186981967226685, + 0.20557587495698246, + 0.3007128919242538, + 0.5589578670606897, + 0.17607239437484073, + 0.3247645305942548, + 0.28265004696019047 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 565126, + "evidence_nonempty_patches": 56, + "first_correction": false, + "restart": "", + "weights": [ + 0.0012569812895489763, + 0.02796522949451945, + 0.022627061811221816, + 0.004080341346961101, + 0.9390747514634746, + 0.00029247988784289957, + 0.0021198435345643842, + 0.0025833111718668077 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5571876015358332, + "uniform": 0.374202964280938, + "top1": 0.5589578670606897, + "top2": 0.5592919600765673, + "top3": 0.5593125851518601 + }, + "pseudo_student_dice": 0.6853953687609664, + "pseudo_ensemble_dice": 0.5542848049226852, + "pseudo_source": "student", + "consistency_scale": 0.618, + "val_dice": 0.7399397755468825, + "val_vi": 0.5367473626096051, + "val_ece": 0.012806996872993453 + }, + { + "round": 3, + "n_labeled": 84, + "selection_seconds": 0.2, + "round_seconds": 107.8, + "teacher_dice_labeled": [ + 0.3557500469410345, + 0.4409271340874671, + 0.21965401520887193, + 0.3356542715086659, + 0.6156078106682936, + 0.1806652407278319, + 0.29110385747882406, + 0.2844647798935356 + ], + "teacher_dice_pool": [ + 0.36221583553452585, + 0.46154148860749067, + 0.2351214263672385, + 0.36298119639097676, + 0.6642216725599263, + 0.19426701924887152, + 0.2809982631984566, + 0.2902214086069471 + ], + "evidence_fg_pixels": 846975, + "evidence_nonempty_patches": 84, + "first_correction": false, + "restart": "", + "weights": [ + 0.0007864934568305689, + 0.020220362781744897, + 0.015079962329175009, + 0.0033502443752500594, + 0.9579861938327272, + 0.0001772858169266955, + 0.0008764938397664921, + 0.0015229635675790894 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6149834430342458, + "uniform": 0.41044819487495626, + "top1": 0.6156078106682936, + "top2": 0.6161620454596852, + "top3": 0.6161824348043965 + }, + "pseudo_student_dice": 0.7165017188224702, + "pseudo_ensemble_dice": 0.5822070619568362, + "pseudo_source": "student", + "consistency_scale": 0.722, + "val_dice": 0.7493534762151666, + "val_vi": 0.537037416881874, + "val_ece": 0.022052875547815737 + }, + { + "round": 4, + "n_labeled": 112, + "selection_seconds": 0.2, + "round_seconds": 111.1, + "teacher_dice_labeled": [ + 0.36799533483144575, + 0.4434738198437197, + 0.22237860981915997, + 0.35119162408855786, + 0.6132689554994883, + 0.18307297983849213, + 0.281595677466981, + 0.2906295936185792 + ], + "teacher_dice_pool": [ + 0.38838436897907, + 0.47321562487468605, + 0.23612988589273762, + 0.3855286998335696, + 0.6543285906911006, + 0.19182863371607134, + 0.26864523780903776, + 0.30642074866250274 + ], + "evidence_fg_pixels": 1109045, + "evidence_nonempty_patches": 112, + "first_correction": false, + "restart": "", + "weights": [ + 0.0009075705663560531, + 0.021176478328176308, + 0.01582113731605874, + 0.003995388643561608, + 0.9554465806535456, + 0.00018541093767134145, + 0.0008136920001201178, + 0.0016537415545100844 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6125056280699729, + "uniform": 0.4087578286493045, + "top1": 0.6132689554994883, + "top2": 0.6137450526434824, + "top3": 0.6137686488987392 + }, + "pseudo_student_dice": 0.7228817582417583, + "pseudo_ensemble_dice": 0.5876644214959471, + "pseudo_source": "student", + "consistency_scale": 0.743, + "val_dice": 0.7508233511614852, + "val_vi": 0.5353514012696726, + "val_ece": 0.024242308398439487 + }, + { + "round": 5, + "n_labeled": 140, + "selection_seconds": 0.2, + "round_seconds": 115.0, + "teacher_dice_labeled": [ + 0.36545827887620963, + 0.44001314710549466, + 0.2250317553851608, + 0.3482704748138672, + 0.6108450152408786, + 0.18098147746558904, + 0.2772557088193389, + 0.28833909699736204 + ], + "teacher_dice_pool": [ + 0.38441743915813703, + 0.4699472512144608, + 0.242043795701895, + 0.38036188631835627, + 0.6520519630893523, + 0.19162936980644238, + 0.28072756803528426, + 0.30415916811101773 + ], + "evidence_fg_pixels": 1372470, + "evidence_nonempty_patches": 140, + "first_correction": false, + "restart": "", + "weights": [ + 0.0009060261815732048, + 0.020946085232241562, + 0.016635612810557043, + 0.0039732992909113215, + 0.9549002581410955, + 0.00018592196889683152, + 0.0007977934031757694, + 0.0016550029715487065 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.61001214706305, + "uniform": 0.40779566499551556, + "top1": 0.6108450152408786, + "top2": 0.6112827862046616, + "top3": 0.6113057282316456 + }, + "pseudo_student_dice": 0.7077233122972116, + "pseudo_ensemble_dice": 0.5583120517817071, + "pseudo_source": "student", + "consistency_scale": 0.692, + "val_dice": 0.7525420432275504, + "val_vi": 0.5316956049707146, + "val_ece": 0.02357947966251423 + }, + { + "round": 6, + "n_labeled": 168, + "selection_seconds": 0.2, + "round_seconds": 118.4, + "teacher_dice_labeled": [ + 0.3634730961229432, + 0.4400121039280854, + 0.22399140525566236, + 0.3461002739895306, + 0.6077325689334545, + 0.18446004079543676, + 0.27816643604219415, + 0.2869076314362251 + ], + "teacher_dice_pool": [ + 0.37718160768208964, + 0.4658483575908974, + 0.24283830333599785, + 0.37970663291846574, + 0.6464730400378235, + 0.19800368751348005, + 0.2827052621021349, + 0.29906355224330206 + ], + "evidence_fg_pixels": 1676005, + "evidence_nonempty_patches": 168, + "first_correction": false, + "restart": "", + "weights": [ + 0.0009152615352997819, + 0.02158362259693189, + 0.01696471714785591, + 0.0040063807443571735, + 0.9538208919175786, + 0.00019836448588131552, + 0.0008296056124843809, + 0.0016811559596108942 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.6068742421539629, + "uniform": 0.4065183273644009, + "top1": 0.6077325689334545, + "top2": 0.6082575257609559, + "top3": 0.6082816859281865 + }, + "pseudo_student_dice": 0.7569865654765929, + "pseudo_ensemble_dice": 0.5640826712486684, + "pseudo_source": "student", + "consistency_scale": 0.857, + "val_dice": 0.7529501278915827, + "val_vi": 0.5239912101194711, + "val_ece": 0.02575743938369193 + }, + { + "round": 7, + "n_labeled": 168, + "selection_seconds": 0.0, + "round_seconds": 35.2, + "final_consolidation": true, + "val_dice": 0.7560981292366642, + "val_vi": 0.5234298513862702, + "val_ece": 0.028411418873243124 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.005164451042246618, + "auroc_top20_error": 0.5020564348188254 + }, + "al_seconds": 669.3, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.0009152615352997819, + 0.02158362259693189, + 0.01696471714785591, + 0.0040063807443571735, + 0.9538208919175786, + 0.00019836448588131552, + 0.0008296056124843809, + 0.0016811559596108942 + ], + "weight_stats_final": { + "max_weight": 0.9538208919175786, + "entropy": 0.24388029175697593, + "effective_sources": 1.2761915506968406, + "min_weight": 0.00019836448588131552 + }, + "top_source_final": "microns_basil", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7511768441625362, + "patch_iou": 0.6116278656599864, + "patch_precision": 0.7322982005032396, + "patch_recall": 0.7746409359216964, + "patch_vi": 0.5222147527748564, + "n_patches": 735, + "cal_ece": 0.02511359686460786, + "cal_nll": 0.19427339443574926, + "cal_brier": 0.060363387699489846, + "cal_mean_confidence": 0.9400941984802542, + "cal_accuracy": 0.9149806016156462, + "cal_predicted_prevalence": 0.21563031462585033, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.21331799387816586 + }, + "test_notta": { + "patch_dice": 0.7511768441625362, + "patch_iou": 0.6116278656599864, + "patch_precision": 0.7322982005032396, + "patch_recall": 0.7746409359216964, + "patch_vi": 0.5222147527748564, + "n_patches": 735, + "cal_ece": 0.02511359686460786, + "cal_nll": 0.19427339443574926, + "cal_brier": 0.060363387699489846, + "cal_mean_confidence": 0.9400941984802542, + "cal_accuracy": 0.9149806016156462, + "cal_predicted_prevalence": 0.21563031462585033, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.21331799387816586 + }, + "test": { + "patch_dice": 0.7595023671503549, + "patch_iou": 0.6220891369072583, + "patch_precision": 0.7419020438291963, + "patch_recall": 0.7812715250362608, + "patch_vi": 0.5085818716439209, + "n_patches": 735, + "image_dice": 0.8065017348049895, + "image_iou": 0.6762415070696743, + "image_precision": 0.7927211103610438, + "image_recall": 0.8211628232225955, + "image_vi": 0.5176808942475893, + "n_images": 15, + "cal_ece": 0.01706907210860156, + "cal_nll": 0.1809449537218227, + "cal_brier": 0.05709824919899649, + "cal_mean_confidence": 0.935303017479304, + "cal_accuracy": 0.918304900085034, + "cal_predicted_prevalence": 0.21492081207482994, + "cal_true_prevalence": 0.20588727678571428, + "cal_mean_probability": 0.21349476381479376 + }, + "eval_seconds": 11.6, + "total_seconds": 901.3, + "resources": { + "phases": { + "setup": { + "seconds": 1.14, + "peak_cpu_rss_mb": 1027.6, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 24.99, + "peak_cpu_rss_mb": 1810.2, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 192.59, + "peak_cpu_rss_mb": 2896.7, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4638.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 669.28, + "peak_cpu_rss_mb": 2896.7, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4646.0, + "rounds": 6 + }, + "inference": { + "seconds": 11.6, + "peak_cpu_rss_mb": 2896.7, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4646.0, + "patches": 735, + "images": 15, + "patches_per_second": 63.4, + "seconds_per_image": 0.773 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 2896.7, + "total_seconds": 899.6 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/snemi3d/macs-v11-norestart-acq-random_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..879279c2b0dff41eeb8db36c442425d00c38d9fc --- /dev/null +++ b/macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-random/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,28,28.4,156.9,"[0.36056026860693446, 0.4508413766529139, 0.21750134794876225, 0.28356610561779927, 0.560875393102026, 0.18601963732494034, 0.3724686404052985, 0.28968887034603186]",,288982,28,True,kept,0.4024189407361833,0.560875393102026,"[0.0013803357494966194, 0.03734608199477827, 0.024685983611284195, 0.0033285842306526635, 0.9269546251229578, 0.000312843356437377, 0.003307582804027515, 0.00268396313036561]",mixture,"{'mixture': 0.5602980725744923, 'uniform': 0.40469001951724776, 'top1': 0.560875393102026, 'top2': 0.5626323977653401, 'top3': 0.5627847905743437}",0.727546285788197,0.5749785964732628,0.011788481275023454,,,,, +2,56,0.2,14.6,"[0.3479823236552919, 0.4186981967226685, 0.20557587495698246, 0.3007128919242538, 0.5589578670606897, 0.17607239437484073, 0.3247645305942548, 0.28265004696019047]",,565126,56,False,,,,"[0.0012569812895489763, 0.02796522949451945, 0.022627061811221816, 0.004080341346961101, 0.9390747514634746, 0.00029247988784289957, 0.0021198435345643842, 0.0025833111718668077]",mixture,"{'mixture': 0.5571876015358332, 'uniform': 0.374202964280938, 'top1': 0.5589578670606897, 'top2': 0.5592919600765673, 'top3': 0.5593125851518601}",0.7399397755468825,0.5367473626096051,0.012806996872993453,0.6853953687609664,0.5542848049226852,student,0.618, +3,84,0.2,107.8,"[0.3557500469410345, 0.4409271340874671, 0.21965401520887193, 0.3356542715086659, 0.6156078106682936, 0.1806652407278319, 0.29110385747882406, 0.2844647798935356]","[0.36221583553452585, 0.46154148860749067, 0.2351214263672385, 0.36298119639097676, 0.6642216725599263, 0.19426701924887152, 0.2809982631984566, 0.2902214086069471]",846975,84,False,,,,"[0.0007864934568305689, 0.020220362781744897, 0.015079962329175009, 0.0033502443752500594, 0.9579861938327272, 0.0001772858169266955, 0.0008764938397664921, 0.0015229635675790894]",mixture,"{'mixture': 0.6149834430342458, 'uniform': 0.41044819487495626, 'top1': 0.6156078106682936, 'top2': 0.6161620454596852, 'top3': 0.6161824348043965}",0.7493534762151666,0.537037416881874,0.022052875547815737,0.7165017188224702,0.5822070619568362,student,0.722, +4,112,0.2,111.1,"[0.36799533483144575, 0.4434738198437197, 0.22237860981915997, 0.35119162408855786, 0.6132689554994883, 0.18307297983849213, 0.281595677466981, 0.2906295936185792]","[0.38838436897907, 0.47321562487468605, 0.23612988589273762, 0.3855286998335696, 0.6543285906911006, 0.19182863371607134, 0.26864523780903776, 0.30642074866250274]",1109045,112,False,,,,"[0.0009075705663560531, 0.021176478328176308, 0.01582113731605874, 0.003995388643561608, 0.9554465806535456, 0.00018541093767134145, 0.0008136920001201178, 0.0016537415545100844]",mixture,"{'mixture': 0.6125056280699729, 'uniform': 0.4087578286493045, 'top1': 0.6132689554994883, 'top2': 0.6137450526434824, 'top3': 0.6137686488987392}",0.7508233511614852,0.5353514012696726,0.024242308398439487,0.7228817582417583,0.5876644214959471,student,0.743, +5,140,0.2,115.0,"[0.36545827887620963, 0.44001314710549466, 0.2250317553851608, 0.3482704748138672, 0.6108450152408786, 0.18098147746558904, 0.2772557088193389, 0.28833909699736204]","[0.38441743915813703, 0.4699472512144608, 0.242043795701895, 0.38036188631835627, 0.6520519630893523, 0.19162936980644238, 0.28072756803528426, 0.30415916811101773]",1372470,140,False,,,,"[0.0009060261815732048, 0.020946085232241562, 0.016635612810557043, 0.0039732992909113215, 0.9549002581410955, 0.00018592196889683152, 0.0007977934031757694, 0.0016550029715487065]",mixture,"{'mixture': 0.61001214706305, 'uniform': 0.40779566499551556, 'top1': 0.6108450152408786, 'top2': 0.6112827862046616, 'top3': 0.6113057282316456}",0.7525420432275504,0.5316956049707146,0.02357947966251423,0.7077233122972116,0.5583120517817071,student,0.692, +6,168,0.2,118.4,"[0.3634730961229432, 0.4400121039280854, 0.22399140525566236, 0.3461002739895306, 0.6077325689334545, 0.18446004079543676, 0.27816643604219415, 0.2869076314362251]","[0.37718160768208964, 0.4658483575908974, 0.24283830333599785, 0.37970663291846574, 0.6464730400378235, 0.19800368751348005, 0.2827052621021349, 0.29906355224330206]",1676005,168,False,,,,"[0.0009152615352997819, 0.02158362259693189, 0.01696471714785591, 0.0040063807443571735, 0.9538208919175786, 0.00019836448588131552, 0.0008296056124843809, 0.0016811559596108942]",mixture,"{'mixture': 0.6068742421539629, 'uniform': 0.4065183273644009, 'top1': 0.6077325689334545, 'top2': 0.6082575257609559, 'top3': 0.6082816859281865}",0.7529501278915827,0.5239912101194711,0.02575743938369193,0.7569865654765929,0.5640826712486684,student,0.857, +7,168,0.0,35.2,,,,,,,,,,,,0.7560981292366642,0.5234298513862702,0.028411418873243124,,,,,True diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/log.txt b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..89375596ba9b6cf6c667796983b643141451999c --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/log.txt @@ -0,0 +1,106 @@ +[10:48:45][ 0s][START] target=whole-mouse-brain variant=macs-v11-norestart-acq-entropy budget=5.0% seed=3 | {"task": "membrane", "target": "whole-mouse-brain", "variant": "macs-v11-norestart-acq-entropy", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "entropy", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[10:48:45][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[10:48:45][ 0s][DATA] whole-mouse-brain: train=5925 val=1250 test=1300 patches +[10:48:46][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, oct-sfl=2.64, snemi=3.55 +[10:48:47][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[10:48:53][ 7s][WEIGHTS] mixture: c-elegans=0.008, cerebellum=0.256, flywire=0.039, berghia=0.075, microns=0.376, oct-glia=0.134, oct-sfl=0.096, snemi=0.016 | eff.sources=5.13 | MMD2 mix=0.1295 uniform=0.1553 +[10:48:53][ 7s][WEIGHTS] per-source: c-elegans=0.145, cerebellum=0.042, flywire=0.001, berghia=0.142, microns=0.020, oct-glia=0.402, oct-sfl=0.243, snemi=0.004 | min-MMD source (NeuroADDA rule) = oct-glia +[10:49:16][ 31s][TEACHERS] target-val Dice: c-elegans=0.386, cerebellum=0.316, flywire=0.110, berghia=0.259, microns=0.329, oct-glia=0.407, oct-sfl=0.440, snemi=0.305 +[10:49:16][ 31s][TEACHERS] ensemble target-val Dice: mixture=0.385, mixture@raw=0.400, mixture@logit=0.380, per_source=0.420, per_source@raw=0.422, per_source@logit=0.419, uniform=0.413, uniform@raw=0.388, uniform@logit=0.405 +[10:49:16][ 31s][WEIGHTS] using mixture: c-elegans=0.008, cerebellum=0.256, flywire=0.039, berghia=0.075, microns=0.376, oct-glia=0.134, oct-sfl=0.096, snemi=0.016 +[10:49:16][ 31s][INIT] student initialised from teacher microns (weight 0.376) +[10:49:18][ 33s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[10:49:18][ 33s][KD] distilling the mixture ensemble (prob space) into the student +[10:50:25][ 100s][KD] soft labels for 5925 patches from 8 teachers in 67s +[10:51:18][ 153s][KD] epoch 1/5 | patches=5925 | loss=0.6690 | 53s +[10:52:11][ 205s][KD] epoch 2/5 | patches=5925 | loss=0.5006 | 52s +[10:53:03][ 258s][KD] epoch 3/5 | patches=5925 | loss=0.4703 | 52s +[10:53:55][ 310s][KD] epoch 4/5 | patches=5925 | loss=0.4599 | 52s +[10:54:46][ 361s][KD] epoch 5/5 | patches=5925 | loss=0.4555 | 51s +[10:54:48][ 363s][ROUND0] before any target label: val_dice=0.4218 val_vi=0.5128 +[10:54:48][ 363s][AL] patches=5925 (pool 5925 after dropping blank; free blank negatives: True) | budget=296 (5.0%) | rounds=6 | per_round=49 | acquisition=entropy | first_round=stratified +[10:54:57][ 372s][AL] round-1 uncertainty quality on 5925 patches: {'spearman_uncertainty_error': -0.20489397455499517, 'auroc_top20_error': 0.39909434029446844} +[10:55:20][ 395s][AL] round 1: stratified cold start: 25 patches from the membrane-rich half (ensemble prevalence >= 0.215) + 24 from the membrane-poor half of 2000 candidates +[10:55:23][ 398s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.433, uniform=0.435, top1=0.471, top2=0.445, top3=0.446 -> using top1 +[10:55:23][ 398s][WEIGHTS] round 1: student kept (labeled Dice 0.396 vs best teacher oct-sfl 0.471); distilling the corrected ensemble into it +[10:55:23][ 398s][WEIGHTS] round 1: teacher Dice on 49 labeled patches + 0 blank (640544 fg px): c-elegans=0.367, cerebellum=0.350, flywire=0.141, berghia=0.235, microns=0.310, oct-glia=0.385, oct-sfl=0.471, snemi=0.254 | weights: c-elegans=0.000, cerebellum=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=1.000, snemi=0.000 +[10:55:32][ 407s][KD] soft labels for 5876 patches from 8 teachers in 9s +[10:56:23][ 458s][KD] epoch 1/2 | patches=5876 | loss=0.1745 | 51s +[10:57:14][ 508s][KD] epoch 2/2 | patches=5876 | loss=0.1632 | 51s +[10:57:17][ 512s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 3s +[10:57:18][ 513s][FT] round 1/6 epoch 1/5 | patches=49 | loss=1.3008 +[10:57:19][ 514s][FT] round 1/6 epoch 2/5 | patches=49 | loss=1.0758 +[10:57:21][ 516s][FT] round 1/6 epoch 3/5 | patches=49 | loss=1.3884 +[10:57:22][ 517s][FT] round 1/6 epoch 4/5 | patches=49 | loss=1.0120 +[10:57:23][ 518s][FT] round 1/6 epoch 5/5 | patches=49 | loss=1.0581 +[10:57:25][ 520s][AL] round 1/6 | labeled=49 | selection_seconds=32.2000 round_seconds=155.3000 restart_student_dice=0.3963 restart_teacher_dice=0.4707 val_dice=0.6465 val_vi=0.4665 val_ece=0.0076 +[10:57:34][ 529s][AL] round 2: 530/5876 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:57:39][ 534s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.491, uniform=0.279, top1=0.461, top2=0.461, top3=0.460 -> using mixture +[10:57:39][ 534s][WEIGHTS] round 2: teacher Dice on 98 labeled patches + 0 blank (1735724 fg px): c-elegans=0.396, cerebellum=0.461, flywire=0.101, berghia=0.203, microns=0.255, oct-glia=0.258, oct-sfl=0.332, snemi=0.393 | weights: c-elegans=0.011, cerebellum=0.699, flywire=0.003, berghia=0.015, microns=0.130, oct-glia=0.048, oct-sfl=0.071, snemi=0.022 +[10:57:41][ 536s][CONS] round 2: on the 49 new patches student Dice=0.548 vs ensemble 0.557 -> pseudo-labels from the ensemble | consistency x1.00 +[10:57:43][ 538s][FT] round 2/6 epoch 1/5 | patches=98 | loss=1.2651 +[10:57:45][ 540s][FT] round 2/6 epoch 2/5 | patches=98 | loss=1.3382 +[10:57:47][ 542s][FT] round 2/6 epoch 3/5 | patches=98 | loss=1.1291 +[10:57:49][ 544s][FT] round 2/6 epoch 4/5 | patches=98 | loss=1.0845 +[10:57:51][ 546s][FT] round 2/6 epoch 5/5 | patches=98 | loss=1.0443 +[10:57:53][ 548s][AL] round 2/6 | labeled=98 | selection_seconds=9.0000 round_seconds=26.0000 pseudo_student_dice=0.5477 pseudo_ensemble_dice=0.5568 consistency_scale=1.0000 val_dice=0.6380 val_vi=0.5154 val_ece=0.0264 +[10:58:02][ 557s][AL] round 3: 503/5827 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:58:09][ 564s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.460, uniform=0.204, top1=0.456, top2=0.456, top3=0.456 -> using mixture +[10:58:09][ 564s][WEIGHTS] round 3: teacher Dice on 147 labeled patches + 0 blank (2564465 fg px): c-elegans=0.356, cerebellum=0.456, flywire=0.073, berghia=0.161, microns=0.188, oct-glia=0.189, oct-sfl=0.246, snemi=0.399 | weights: c-elegans=0.009, cerebellum=0.802, flywire=0.003, berghia=0.012, microns=0.080, oct-glia=0.029, oct-sfl=0.037, snemi=0.028 +[10:58:11][ 566s][CONS] round 3: on the 49 new patches student Dice=0.221 vs ensemble 0.440 -> pseudo-labels from the ensemble | consistency x1.00 +[10:58:14][ 569s][FT] round 3/6 epoch 1/5 | patches=147 | loss=1.2710 +[10:58:17][ 572s][FT] round 3/6 epoch 2/5 | patches=147 | loss=1.3728 +[10:58:20][ 575s][FT] round 3/6 epoch 3/5 | patches=147 | loss=1.2233 +[10:58:23][ 578s][FT] round 3/6 epoch 4/5 | patches=147 | loss=1.1298 +[10:58:26][ 581s][FT] round 3/6 epoch 5/5 | patches=147 | loss=1.0388 +[10:58:28][ 583s][AL] round 3/6 | labeled=147 | selection_seconds=9.0000 round_seconds=32.4000 pseudo_student_dice=0.2214 pseudo_ensemble_dice=0.4398 consistency_scale=1.0000 val_dice=0.6743 val_vi=0.4691 val_ece=0.0284 +[10:58:37][ 591s][AL] round 4: 697/5778 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[10:58:46][ 601s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.389, uniform=0.206, top1=0.388, top2=0.388, top3=0.388 -> using mixture +[10:58:46][ 601s][WEIGHTS] round 4: teacher Dice on 196 labeled patches + 0 blank (3036531 fg px): c-elegans=0.330, cerebellum=0.388, flywire=0.072, berghia=0.168, microns=0.184, oct-glia=0.198, oct-sfl=0.243, snemi=0.357 | weights: c-elegans=0.011, cerebellum=0.685, flywire=0.004, berghia=0.022, microns=0.132, oct-glia=0.054, oct-sfl=0.060, snemi=0.031 +[10:58:51][ 606s][CONS] round 4: on the 49 new patches student Dice=0.459 vs ensemble 0.213 -> pseudo-labels from the student | consistency x0.00 +[10:58:53][ 608s][FT] round 4/6 epoch 1/5 | patches=196 | loss=1.0920 +[10:58:55][ 610s][FT] round 4/6 epoch 2/5 | patches=196 | loss=0.9350 +[10:58:57][ 612s][FT] round 4/6 epoch 3/5 | patches=196 | loss=0.9393 +[10:58:59][ 614s][FT] round 4/6 epoch 4/5 | patches=196 | loss=0.9013 +[10:59:01][ 616s][FT] round 4/6 epoch 5/5 | patches=196 | loss=0.9538 +[10:59:03][ 618s][AL] round 4/6 | labeled=196 | selection_seconds=9.0000 round_seconds=32.9000 pseudo_student_dice=0.4592 pseudo_ensemble_dice=0.2127 consistency_scale=0.0000 val_dice=0.7087 val_vi=0.4190 val_ece=0.0029 +[10:59:11][ 626s][AL] round 5: 567/5729 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:00:03][ 678s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.413, cerebellum=0.421, flywire=0.151, berghia=0.286, microns=0.320, oct-glia=0.441, oct-sfl=0.539, snemi=0.334 | blended with labeled Dice at 0.50 +[11:00:52][ 727s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.450, uniform=0.317, top1=0.412, top2=0.429, top3=0.429 -> using mixture +[11:00:52][ 727s][WEIGHTS] round 5: teacher Dice on 245 labeled patches + 0 blank (4134678 fg px): c-elegans=0.381, cerebellum=0.412, flywire=0.107, berghia=0.232, microns=0.233, oct-glia=0.319, oct-sfl=0.385, snemi=0.362 | weights: c-elegans=0.012, cerebellum=0.540, flywire=0.004, berghia=0.026, microns=0.132, oct-glia=0.112, oct-sfl=0.154, snemi=0.020 +[11:00:57][ 732s][CONS] round 5: on the 49 new patches student Dice=0.700 vs ensemble 0.319 -> pseudo-labels from the student | consistency x0.67 +[11:01:01][ 736s][FT] round 5/6 epoch 1/5 | patches=245 | loss=1.0706 +[11:01:06][ 741s][FT] round 5/6 epoch 2/5 | patches=245 | loss=1.2142 +[11:01:11][ 745s][FT] round 5/6 epoch 3/5 | patches=245 | loss=1.0638 +[11:01:15][ 750s][FT] round 5/6 epoch 4/5 | patches=245 | loss=1.0411 +[11:01:20][ 754s][FT] round 5/6 epoch 5/5 | patches=245 | loss=1.0409 +[11:01:22][ 757s][AL] round 5/6 | labeled=245 | selection_seconds=8.8000 round_seconds=136.9000 pseudo_student_dice=0.6995 pseudo_ensemble_dice=0.3187 consistency_scale=0.6650 val_dice=0.7157 val_vi=0.3772 val_ece=0.0090 +[11:01:30][ 765s][AL] round 6: 707/5680 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:02:23][ 818s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.392, cerebellum=0.395, flywire=0.157, berghia=0.265, microns=0.334, oct-glia=0.443, oct-sfl=0.542, snemi=0.297 | blended with labeled Dice at 0.50 +[11:03:13][ 868s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.419, uniform=0.323, top1=0.399, top2=0.419, top3=0.419 -> using mixture +[11:03:13][ 868s][WEIGHTS] round 6: teacher Dice on 294 labeled patches + 0 blank (5686541 fg px): c-elegans=0.385, cerebellum=0.399, flywire=0.111, berghia=0.251, microns=0.231, oct-glia=0.326, oct-sfl=0.397, snemi=0.368 | weights: c-elegans=0.013, cerebellum=0.490, flywire=0.004, berghia=0.033, microns=0.134, oct-glia=0.124, oct-sfl=0.179, snemi=0.022 +[11:03:18][ 873s][CONS] round 6: on the 49 new patches student Dice=0.482 vs ensemble 0.215 -> pseudo-labels from the student | consistency x0.00 +[11:03:21][ 876s][FT] round 6/6 epoch 1/5 | patches=294 | loss=1.0749 +[11:03:24][ 879s][FT] round 6/6 epoch 2/5 | patches=294 | loss=0.9716 +[11:03:27][ 882s][FT] round 6/6 epoch 3/5 | patches=294 | loss=0.9355 +[11:03:30][ 885s][FT] round 6/6 epoch 4/5 | patches=294 | loss=0.9278 +[11:03:33][ 887s][FT] round 6/6 epoch 5/5 | patches=294 | loss=0.8386 +[11:03:34][ 889s][AL] round 6/6 | labeled=294 | selection_seconds=8.8000 round_seconds=130.9000 pseudo_student_dice=0.4819 pseudo_ensemble_dice=0.2152 consistency_scale=0.0000 val_dice=0.7126 val_vi=0.3923 val_ece=0.0091 +[11:03:37][ 892s][CONS] final consolidation: pseudo-labels refreshed from the student +[11:03:40][ 895s][FT] final 1/10 epoch 1/1 | patches=294 | loss=0.8899 +[11:03:43][ 898s][FT] final 2/10 epoch 1/1 | patches=294 | loss=0.8436 +[11:03:46][ 901s][FT] final 3/10 epoch 1/1 | patches=294 | loss=0.7589 +[11:03:48][ 903s][FT] final 4/10 epoch 1/1 | patches=294 | loss=0.8406 +[11:03:51][ 906s][FT] final 5/10 epoch 1/1 | patches=294 | loss=0.8507 +[11:03:54][ 909s][FT] final 6/10 epoch 1/1 | patches=294 | loss=0.7447 +[11:03:57][ 912s][FT] final 7/10 epoch 1/1 | patches=294 | loss=0.7217 +[11:04:00][ 914s][FT] final 8/10 epoch 1/1 | patches=294 | loss=0.6748 +[11:04:02][ 917s][FT] final 9/10 epoch 1/1 | patches=294 | loss=0.7146 +[11:04:05][ 920s][FT] final 10/10 epoch 1/1 | patches=294 | loss=0.6861 +[11:04:07][ 922s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=30.5000 val_dice=0.7200 val_vi=0.3878 val_ece=0.0197 +[11:04:07][ 922s][WEIGHTS] final: c-elegans=0.013, cerebellum=0.490, flywire=0.004, berghia=0.033, microns=0.134, oct-glia=0.124, oct-sfl=0.179, snemi=0.022 +[11:04:25][ 939s][TEST] raw student (no EMA, no TTA): patch_dice=0.7044 patch_vi=0.3511 +[11:04:27][ 942s][TEST] final model without TTA: patch_dice=0.7044 patch_vi=0.3511 +[11:04:27][ 942s][RESOURCES] {"phases": {"setup": {"seconds": 1.12, "peak_cpu_rss_mb": 1027.4, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 31.35, "peak_cpu_rss_mb": 1809.2, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 328.23, "peak_cpu_rss_mb": 3676.9, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 558.98, "peak_cpu_rss_mb": 3676.9, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4646.0, "rounds": 6}, "inference": {"seconds": 20.15, "peak_cpu_rss_mb": 3676.9, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4646.0, "patches": 1300, "images": 52, "patches_per_second": 64.5, "seconds_per_image": 0.387}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 3676.9, "total_seconds": 939.8} +[11:04:27][ 942s][TEST] patch_dice=0.7253 patch_iou=0.6459 patch_precision=0.7409 patch_recall=0.8422 patch_vi=0.3269 image_dice=0.8342 image_iou=0.7308 image_precision=0.8019 image_recall=0.8938 image_vi=0.3823 cal_ece=0.0061 cal_nll=0.1796 cal_brier=0.0510 cal_mean_confidence=0.9385 cal_accuracy=0.9328 cal_predicted_prevalence=0.2326 cal_true_prevalence=0.2108 cal_mean_probability=0.2424 +[11:04:27][ 942s][DONE] whole-mouse-brain macs-v11-norestart-acq-entropy b5 s3 | test patch Dice=0.7253 VI=0.3269 | image Dice=0.8342 VI=0.3823 | 942s diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/result.json b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..ebbb58db4565c35d3995a830e8062e01f0c485d5 --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/result.json @@ -0,0 +1,767 @@ +{ + "target": "whole-mouse-brain", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 3, + "config": { + "task": "membrane", + "target": "whole-mouse-brain", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "entropy", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "whole-mouse-brain", + "--variants", + "macs-v11-norestart-acq-entropy", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d" + ], + "alphas": [ + 0.007617762850063339, + 0.2559312601209969, + 0.03865993944852264, + 0.07533345106002441, + 0.3764169834180784, + 0.13449516240019677, + 0.09581340232315401, + 0.0157320383789636 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d" + ], + "mixture": [ + 0.007617762850063339, + 0.2559312601209969, + 0.03865993944852264, + 0.07533345106002441, + 0.3764169834180784, + 0.13449516240019677, + 0.09581340232315401, + 0.0157320383789636 + ], + "per_source": [ + 0.1450746039368088, + 0.041950766450651496, + 0.0008814721099573598, + 0.1418325124471684, + 0.02035312744213396, + 0.4023730240336797, + 0.24338555075976573, + 0.004148942819834595 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.11311670518772092, + "mmd2_mixture": 0.12945823384265215, + "mmd2_uniform": 0.155288743679944, + "mmd2_single_source": [ + 0.38071639717042494, + 0.17697894982207135, + 0.3194061281637849, + 0.24037642211575033, + 0.16266554797152122, + 0.2734526560532348, + 0.2563986706812869, + 0.2660414632502267 + ], + "median_squared_distance": 0.18592532667757627, + "gammas": [ + 1.3446258477390725, + 2.689251695478145, + 5.37850339095629, + 10.75700678191258, + 21.51401356382516 + ], + "max_weight": 0.3764169834180784, + "entropy": 1.6341528654931232, + "effective_sources": 5.125114496180783, + "min_weight": 0.007617762850063339 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.33220085501273156, + 0.46442573060595516, + 0.8760630244687474, + 0.33460943356720163, + 0.5415026072699861, + 0.2234870921926152, + 0.2770625113402161, + 0.7109869456345189 + ], + "standardized": [ + -0.6467275403932012, + -0.02635183204683673, + 1.9049774780863755, + -0.6354269162093369, + 0.33527922824378936, + -1.1567932698101913, + -0.9054268876938927, + 1.130469739823295 + ], + "tau": 0.5, + "max_weight": 0.4023730240336797, + "entropy": 1.508581788687784, + "effective_sources": 4.520315483702711, + "min_weight": 0.0008814721099573598 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.38573376502042933, + 0.31643357804427197, + 0.10997258971952562, + 0.2588485480329741, + 0.3288178379711084, + 0.4072116167356654, + 0.4397876396557443, + 0.30535999081848075 + ], + "ensemble_val_dice": { + "mixture": 0.3845088921696238, + "mixture@raw": 0.3998907914691259, + "mixture@logit": 0.3802050170706561, + "per_source": 0.4199022574247113, + "per_source@raw": 0.42186088103254527, + "per_source@logit": 0.41929713845028616, + "uniform": 0.4128319943503538, + "uniform@raw": 0.3877140229453584, + "uniform@logit": 0.4054972948683435 + }, + "val_patches": 1024 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 2.6421806812286377, + 3.5461363792419434 + ], + "top_source": "microns_basil", + "weight_stats": { + "max_weight": 0.3764169834180784, + "entropy": 1.6341528654931232, + "effective_sources": 5.125114496180783, + "min_weight": 0.007617762850063339 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.261904761904762, + "spearman_weight_taxonomy": 0.39440531887330776, + "top_source": "microns_basil", + "top_source_is_best_teacher": false + }, + "init_teacher": "microns_basil", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 328.2, + "round0_val": { + "val_dice": 0.42175774709778946, + "val_vi": 0.5128062998414359, + "val_ece": 0.14443836415563802 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 49, + "selection_seconds": 32.2, + "round_seconds": 155.3, + "teacher_dice_labeled": [ + 0.36738741011177045, + 0.34973487377165097, + 0.14100291682142935, + 0.2353012898060949, + 0.30996524321190877, + 0.38452188964078277, + 0.47067622607709797, + 0.25385926187706875 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 640544, + "evidence_nonempty_patches": 39, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.3962524167755847, + "restart_teacher_dice": 0.47067622607709797, + "weights": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.43280781937383384, + "uniform": 0.43493152236966137, + "top1": 0.47067622607709797, + "top2": 0.44454578875106776, + "top3": 0.4462152205328611 + }, + "val_dice": 0.646540074905994, + "val_vi": 0.4665198719972955, + "val_ece": 0.0075807375838696745 + }, + { + "round": 2, + "n_labeled": 98, + "selection_seconds": 9.0, + "round_seconds": 26.0, + "teacher_dice_labeled": [ + 0.3964949872852674, + 0.46138364251508823, + 0.1010757881063798, + 0.20256512904065874, + 0.25460547393201843, + 0.2577079727691892, + 0.33156287339796325, + 0.39328567712650325 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1735724, + "evidence_nonempty_patches": 84, + "first_correction": false, + "restart": "", + "weights": [ + 0.010880776159647547, + 0.699461172663094, + 0.0028780896424400485, + 0.015473680225314972, + 0.1301018016412744, + 0.04795067726707895, + 0.07149276637716834, + 0.02176103602398173 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.4910491059378369, + "uniform": 0.2787527762668842, + "top1": 0.46138364251508823, + "top2": 0.4609649416796084, + "top3": 0.45997326697860197 + }, + "pseudo_student_dice": 0.5476616446618572, + "pseudo_ensemble_dice": 0.556802486489929, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.6379528251954637, + "val_vi": 0.5153946538049008, + "val_ece": 0.02644548907751808 + }, + { + "round": 3, + "n_labeled": 147, + "selection_seconds": 9.0, + "round_seconds": 32.4, + "teacher_dice_labeled": [ + 0.35633572198115654, + 0.4561676677016383, + 0.07341647254380124, + 0.16051329920740956, + 0.18766835678359275, + 0.18921733847825212, + 0.24562380051934984, + 0.39910204344838873 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 2564465, + "evidence_nonempty_patches": 113, + "first_correction": false, + "restart": "", + "weights": [ + 0.008798245853375434, + 0.8021518907392878, + 0.0026371032015051134, + 0.012277516649360495, + 0.08048682486773641, + 0.029207164112188905, + 0.036574562641144275, + 0.027866691935401616 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.460341006527611, + "uniform": 0.2037175342073936, + "top1": 0.4561676677016383, + "top2": 0.4559419693890312, + "top3": 0.45570403413587507 + }, + "pseudo_student_dice": 0.22137009624307094, + "pseudo_ensemble_dice": 0.4397540958983941, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.6743464690510582, + "val_vi": 0.4690869017455335, + "val_ece": 0.02842435821653777 + }, + { + "round": 4, + "n_labeled": 196, + "selection_seconds": 9.0, + "round_seconds": 32.9, + "teacher_dice_labeled": [ + 0.3299771096263301, + 0.38804163172125516, + 0.07160612127896214, + 0.16794057559069106, + 0.18443843255050638, + 0.19807899270143609, + 0.24278902382700712, + 0.3569940772081713 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 3036531, + "evidence_nonempty_patches": 153, + "first_correction": false, + "restart": "", + "weights": [ + 0.01141576602419868, + 0.6854439711596396, + 0.004373688359092133, + 0.022333137905060703, + 0.13160742045978957, + 0.05389618569239227, + 0.060041402346467036, + 0.030888428053359913 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.3890792156145218, + "uniform": 0.20577203391014331, + "top1": 0.38804163172125516, + "top2": 0.3879763730386121, + "top3": 0.38780666396278846 + }, + "pseudo_student_dice": 0.45920203220412686, + "pseudo_ensemble_dice": 0.21272979415981438, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.7087167370495862, + "val_vi": 0.4190034422858896, + "val_ece": 0.002885292617739875 + }, + { + "round": 5, + "n_labeled": 245, + "selection_seconds": 8.8, + "round_seconds": 136.9, + "teacher_dice_labeled": [ + 0.38112177883028797, + 0.4121840427291803, + 0.10655492598420999, + 0.23207856359270612, + 0.2329279883222823, + 0.3189693538938384, + 0.3850992144352713, + 0.362244453467251 + ], + "teacher_dice_pool": [ + 0.4131676741656513, + 0.42103200046233735, + 0.1509888135870009, + 0.2862589597725059, + 0.3200670355049752, + 0.4407886865128643, + 0.5394127796666625, + 0.3344276364677657 + ], + "evidence_fg_pixels": 4134678, + "evidence_nonempty_patches": 193, + "first_correction": false, + "restart": "", + "weights": [ + 0.011779082999612873, + 0.5398947030835544, + 0.003838101998014584, + 0.0262413054958334, + 0.13223785280483505, + 0.11170333649150943, + 0.15416439364109832, + 0.020141223485542006 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.45044626372960384, + "uniform": 0.3174082551184142, + "top1": 0.4121840427291803, + "top2": 0.4287747359626445, + "top3": 0.42858896234699534 + }, + "pseudo_student_dice": 0.6995360803402509, + "pseudo_ensemble_dice": 0.31871799632824066, + "pseudo_source": "student", + "consistency_scale": 0.665, + "val_dice": 0.7156779814900249, + "val_vi": 0.3771571907269592, + "val_ece": 0.009027793309739329 + }, + { + "round": 6, + "n_labeled": 294, + "selection_seconds": 8.8, + "round_seconds": 130.9, + "teacher_dice_labeled": [ + 0.3854258140408431, + 0.399042101167703, + 0.11107803619300319, + 0.250959391571319, + 0.23089472687251988, + 0.32607846011094965, + 0.3967225724361204, + 0.36848059917839887 + ], + "teacher_dice_pool": [ + 0.39226917381905135, + 0.3954502206402, + 0.1565966359168242, + 0.265017680659922, + 0.3343952632446626, + 0.4432951310911836, + 0.541844091574079, + 0.2973078419837905 + ], + "evidence_fg_pixels": 5686541, + "evidence_nonempty_patches": 237, + "first_correction": false, + "restart": "", + "weights": [ + 0.012736176365196153, + 0.4903088525913747, + 0.004159063249292493, + 0.03282609531881109, + 0.13420257271254782, + 0.12421555848074521, + 0.17934904829414433, + 0.02220263298788817 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.4187754258232841, + "uniform": 0.3225095176404339, + "top1": 0.399042101167703, + "top2": 0.4194592007402292, + "top3": 0.41927779087623196 + }, + "pseudo_student_dice": 0.48187702682941513, + "pseudo_ensemble_dice": 0.21516510877757866, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.7125637098332068, + "val_vi": 0.3922931931632452, + "val_ece": 0.009128083181487133 + }, + { + "round": 7, + "n_labeled": 294, + "selection_seconds": 0.0, + "round_seconds": 30.5, + "final_consolidation": true, + "val_dice": 0.7199579951007331, + "val_vi": 0.3878433725098941, + "val_ece": 0.01970009244399582 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.20489397455499517, + "auroc_top20_error": 0.39909434029446844 + }, + "al_seconds": 559.0, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.012736176365196153, + 0.4903088525913747, + 0.004159063249292493, + 0.03282609531881109, + 0.13420257271254782, + 0.12421555848074521, + 0.17934904829414433, + 0.02220263298788817 + ], + "weight_stats_final": { + "max_weight": 0.4903088525913747, + "entropy": 1.461326814289988, + "effective_sources": 4.3116765288398, + "min_weight": 0.004159063249292493 + }, + "top_source_final": "dhanyasi-P14-mouse-cerebellum", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7044237441791064, + "patch_iou": 0.6208068695363332, + "patch_precision": 0.7148284002150658, + "patch_recall": 0.8299271833672399, + "patch_vi": 0.351098248100767, + "n_patches": 1300, + "cal_ece": 0.019415714926569648, + "cal_nll": 0.19742326733203194, + "cal_brier": 0.05522876654738458, + "cal_mean_confidence": 0.9472860574746466, + "cal_accuracy": 0.9278703425480769, + "cal_predicted_prevalence": 0.23288611778846155, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.23892273644979425 + }, + "test_notta": { + "patch_dice": 0.7044237441791064, + "patch_iou": 0.6208068695363332, + "patch_precision": 0.7148284002150658, + "patch_recall": 0.8299271833672399, + "patch_vi": 0.351098248100767, + "n_patches": 1300, + "cal_ece": 0.019415714926569648, + "cal_nll": 0.19742326733203194, + "cal_brier": 0.05522876654738458, + "cal_mean_confidence": 0.9472860574746466, + "cal_accuracy": 0.9278703425480769, + "cal_predicted_prevalence": 0.23288611778846155, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.23892273644979425 + }, + "test": { + "patch_dice": 0.7253355817647213, + "patch_iou": 0.6458858156903308, + "patch_precision": 0.7408743616027792, + "patch_recall": 0.842185801710432, + "patch_vi": 0.3268501847439328, + "n_patches": 1300, + "image_dice": 0.8342029779726485, + "image_iou": 0.7307789106547949, + "image_precision": 0.8018712059262382, + "image_recall": 0.8937876571487801, + "image_vi": 0.38229628640130375, + "n_images": 52, + "cal_ece": 0.006070841455767178, + "cal_nll": 0.17962633679523854, + "cal_brier": 0.05098272736595117, + "cal_mean_confidence": 0.9384511861769915, + "cal_accuracy": 0.9328049879807693, + "cal_predicted_prevalence": 0.23264197716346155, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.24242487424387058 + }, + "eval_seconds": 20.1, + "total_seconds": 942.0, + "resources": { + "phases": { + "setup": { + "seconds": 1.12, + "peak_cpu_rss_mb": 1027.4, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 31.35, + "peak_cpu_rss_mb": 1809.2, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 328.23, + "peak_cpu_rss_mb": 3676.9, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4638.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 558.98, + "peak_cpu_rss_mb": 3676.9, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4646.0, + "rounds": 6 + }, + "inference": { + "seconds": 20.15, + "peak_cpu_rss_mb": 3676.9, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4646.0, + "patches": 1300, + "images": 52, + "patches_per_second": 64.5, + "seconds_per_image": 0.387 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 3676.9, + "total_seconds": 939.8 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/whole-mouse-brain/macs-v11-norestart-acq-entropy_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..16545d8d3927e9959580d1253bf50f4fc2e0b5c4 --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,49,32.2,155.3,"[0.36738741011177045, 0.34973487377165097, 0.14100291682142935, 0.2353012898060949, 0.30996524321190877, 0.38452188964078277, 0.47067622607709797, 0.25385926187706875]",,640544,39,True,kept,0.3962524167755847,0.47067622607709797,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]",top1,"{'mixture': 0.43280781937383384, 'uniform': 0.43493152236966137, 'top1': 0.47067622607709797, 'top2': 0.44454578875106776, 'top3': 0.4462152205328611}",0.646540074905994,0.4665198719972955,0.0075807375838696745,,,,, +2,98,9.0,26.0,"[0.3964949872852674, 0.46138364251508823, 0.1010757881063798, 0.20256512904065874, 0.25460547393201843, 0.2577079727691892, 0.33156287339796325, 0.39328567712650325]",,1735724,84,False,,,,"[0.010880776159647547, 0.699461172663094, 0.0028780896424400485, 0.015473680225314972, 0.1301018016412744, 0.04795067726707895, 0.07149276637716834, 0.02176103602398173]",mixture,"{'mixture': 0.4910491059378369, 'uniform': 0.2787527762668842, 'top1': 0.46138364251508823, 'top2': 0.4609649416796084, 'top3': 0.45997326697860197}",0.6379528251954637,0.5153946538049008,0.02644548907751808,0.5476616446618572,0.556802486489929,ensemble,1.0, +3,147,9.0,32.4,"[0.35633572198115654, 0.4561676677016383, 0.07341647254380124, 0.16051329920740956, 0.18766835678359275, 0.18921733847825212, 0.24562380051934984, 0.39910204344838873]",,2564465,113,False,,,,"[0.008798245853375434, 0.8021518907392878, 0.0026371032015051134, 0.012277516649360495, 0.08048682486773641, 0.029207164112188905, 0.036574562641144275, 0.027866691935401616]",mixture,"{'mixture': 0.460341006527611, 'uniform': 0.2037175342073936, 'top1': 0.4561676677016383, 'top2': 0.4559419693890312, 'top3': 0.45570403413587507}",0.6743464690510582,0.4690869017455335,0.02842435821653777,0.22137009624307094,0.4397540958983941,ensemble,1.0, +4,196,9.0,32.9,"[0.3299771096263301, 0.38804163172125516, 0.07160612127896214, 0.16794057559069106, 0.18443843255050638, 0.19807899270143609, 0.24278902382700712, 0.3569940772081713]",,3036531,153,False,,,,"[0.01141576602419868, 0.6854439711596396, 0.004373688359092133, 0.022333137905060703, 0.13160742045978957, 0.05389618569239227, 0.060041402346467036, 0.030888428053359913]",mixture,"{'mixture': 0.3890792156145218, 'uniform': 0.20577203391014331, 'top1': 0.38804163172125516, 'top2': 0.3879763730386121, 'top3': 0.38780666396278846}",0.7087167370495862,0.4190034422858896,0.002885292617739875,0.45920203220412686,0.21272979415981438,student,0.0, +5,245,8.8,136.9,"[0.38112177883028797, 0.4121840427291803, 0.10655492598420999, 0.23207856359270612, 0.2329279883222823, 0.3189693538938384, 0.3850992144352713, 0.362244453467251]","[0.4131676741656513, 0.42103200046233735, 0.1509888135870009, 0.2862589597725059, 0.3200670355049752, 0.4407886865128643, 0.5394127796666625, 0.3344276364677657]",4134678,193,False,,,,"[0.011779082999612873, 0.5398947030835544, 0.003838101998014584, 0.0262413054958334, 0.13223785280483505, 0.11170333649150943, 0.15416439364109832, 0.020141223485542006]",mixture,"{'mixture': 0.45044626372960384, 'uniform': 0.3174082551184142, 'top1': 0.4121840427291803, 'top2': 0.4287747359626445, 'top3': 0.42858896234699534}",0.7156779814900249,0.3771571907269592,0.009027793309739329,0.6995360803402509,0.31871799632824066,student,0.665, +6,294,8.8,130.9,"[0.3854258140408431, 0.399042101167703, 0.11107803619300319, 0.250959391571319, 0.23089472687251988, 0.32607846011094965, 0.3967225724361204, 0.36848059917839887]","[0.39226917381905135, 0.3954502206402, 0.1565966359168242, 0.265017680659922, 0.3343952632446626, 0.4432951310911836, 0.541844091574079, 0.2973078419837905]",5686541,237,False,,,,"[0.012736176365196153, 0.4903088525913747, 0.004159063249292493, 0.03282609531881109, 0.13420257271254782, 0.12421555848074521, 0.17934904829414433, 0.02220263298788817]",mixture,"{'mixture': 0.4187754258232841, 'uniform': 0.3225095176404339, 'top1': 0.399042101167703, 'top2': 0.4194592007402292, 'top3': 0.41927779087623196}",0.7125637098332068,0.3922931931632452,0.009128083181487133,0.48187702682941513,0.21516510877757866,student,0.0, +7,294,0.0,30.5,,,,,,,,,,,,0.7199579951007331,0.3878433725098941,0.01970009244399582,,,,,True diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/log.txt b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..ad17dae1ddaed2f968672e1bfd89bf18133030d9 --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/log.txt @@ -0,0 +1,106 @@ +[11:04:30][ 0s][START] target=whole-mouse-brain variant=macs-v11-norestart-acq-margin budget=5.0% seed=3 | {"task": "membrane", "target": "whole-mouse-brain", "variant": "macs-v11-norestart-acq-margin", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "margin", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[11:04:30][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[11:04:30][ 0s][DATA] whole-mouse-brain: train=5925 val=1250 test=1300 patches +[11:04:31][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, oct-sfl=2.64, snemi=3.55 +[11:04:31][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s +[11:04:37][ 7s][WEIGHTS] mixture: c-elegans=0.008, cerebellum=0.256, flywire=0.039, berghia=0.075, microns=0.376, oct-glia=0.134, oct-sfl=0.096, snemi=0.016 | eff.sources=5.13 | MMD2 mix=0.1295 uniform=0.1553 +[11:04:37][ 7s][WEIGHTS] per-source: c-elegans=0.145, cerebellum=0.042, flywire=0.001, berghia=0.142, microns=0.020, oct-glia=0.402, oct-sfl=0.243, snemi=0.004 | min-MMD source (NeuroADDA rule) = oct-glia +[11:05:01][ 31s][TEACHERS] target-val Dice: c-elegans=0.386, cerebellum=0.316, flywire=0.110, berghia=0.259, microns=0.329, oct-glia=0.407, oct-sfl=0.440, snemi=0.305 +[11:05:01][ 31s][TEACHERS] ensemble target-val Dice: mixture=0.385, mixture@raw=0.400, mixture@logit=0.380, per_source=0.420, per_source@raw=0.422, per_source@logit=0.419, uniform=0.413, uniform@raw=0.388, uniform@logit=0.405 +[11:05:01][ 31s][WEIGHTS] using mixture: c-elegans=0.008, cerebellum=0.256, flywire=0.039, berghia=0.075, microns=0.376, oct-glia=0.134, oct-sfl=0.096, snemi=0.016 +[11:05:01][ 31s][INIT] student initialised from teacher microns (weight 0.376) +[11:05:03][ 33s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[11:05:03][ 33s][KD] distilling the mixture ensemble (prob space) into the student +[11:06:10][ 100s][KD] soft labels for 5925 patches from 8 teachers in 67s +[11:07:03][ 153s][KD] epoch 1/5 | patches=5925 | loss=0.6683 | 53s +[11:07:55][ 205s][KD] epoch 2/5 | patches=5925 | loss=0.4975 | 52s +[11:08:47][ 258s][KD] epoch 3/5 | patches=5925 | loss=0.4701 | 52s +[11:09:39][ 309s][KD] epoch 4/5 | patches=5925 | loss=0.4595 | 52s +[11:10:31][ 362s][KD] epoch 5/5 | patches=5925 | loss=0.4557 | 52s +[11:10:33][ 364s][ROUND0] before any target label: val_dice=0.4161 val_vi=0.5124 +[11:10:33][ 364s][AL] patches=5925 (pool 5925 after dropping blank; free blank negatives: True) | budget=296 (5.0%) | rounds=6 | per_round=49 | acquisition=margin | first_round=stratified +[11:10:43][ 373s][AL] round-1 uncertainty quality on 5925 patches: {'spearman_uncertainty_error': -0.24881207404064967, 'auroc_top20_error': 0.3594799622567608} +[11:11:06][ 396s][AL] round 1: stratified cold start: 25 patches from the membrane-rich half (ensemble prevalence >= 0.215) + 24 from the membrane-poor half of 2000 candidates +[11:11:08][ 399s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.433, uniform=0.435, top1=0.471, top2=0.445, top3=0.446 -> using top1 +[11:11:09][ 399s][WEIGHTS] round 1: student kept (labeled Dice 0.384 vs best teacher oct-sfl 0.471); distilling the corrected ensemble into it +[11:11:09][ 399s][WEIGHTS] round 1: teacher Dice on 49 labeled patches + 0 blank (640544 fg px): c-elegans=0.367, cerebellum=0.350, flywire=0.141, berghia=0.235, microns=0.310, oct-glia=0.385, oct-sfl=0.471, snemi=0.254 | weights: c-elegans=0.000, cerebellum=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=1.000, snemi=0.000 +[11:11:18][ 408s][KD] soft labels for 5876 patches from 8 teachers in 9s +[11:12:10][ 460s][KD] epoch 1/2 | patches=5876 | loss=0.1740 | 52s +[11:13:01][ 512s][KD] epoch 2/2 | patches=5876 | loss=0.1640 | 52s +[11:13:05][ 515s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 3s +[11:13:06][ 516s][FT] round 1/6 epoch 1/5 | patches=49 | loss=1.3153 +[11:13:07][ 517s][FT] round 1/6 epoch 2/5 | patches=49 | loss=1.0823 +[11:13:08][ 519s][FT] round 1/6 epoch 3/5 | patches=49 | loss=1.2696 +[11:13:10][ 520s][FT] round 1/6 epoch 4/5 | patches=49 | loss=1.0443 +[11:13:11][ 521s][FT] round 1/6 epoch 5/5 | patches=49 | loss=1.0469 +[11:13:13][ 523s][AL] round 1/6 | labeled=49 | selection_seconds=32.3000 round_seconds=157.5000 restart_student_dice=0.3843 restart_teacher_dice=0.4707 val_dice=0.6424 val_vi=0.4584 val_ece=0.0183 +[11:13:22][ 532s][AL] round 2: 525/5876 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:13:27][ 537s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.448, uniform=0.275, top1=0.479, top2=0.449, top3=0.449 -> using top1 +[11:13:27][ 537s][WEIGHTS] round 2: teacher Dice on 98 labeled patches + 0 blank (1973085 fg px): c-elegans=0.479, cerebellum=0.447, flywire=0.164, berghia=0.213, microns=0.222, oct-glia=0.255, oct-sfl=0.330, snemi=0.432 | weights: c-elegans=1.000, cerebellum=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.000, snemi=0.000 +[11:13:29][ 539s][CONS] round 2: on the 49 new patches student Dice=0.571 vs ensemble 0.563 -> pseudo-labels from the ensemble | consistency x1.00 +[11:13:31][ 541s][FT] round 2/6 epoch 1/5 | patches=98 | loss=1.2983 +[11:13:33][ 543s][FT] round 2/6 epoch 2/5 | patches=98 | loss=1.2472 +[11:13:35][ 545s][FT] round 2/6 epoch 3/5 | patches=98 | loss=1.1352 +[11:13:37][ 547s][FT] round 2/6 epoch 4/5 | patches=98 | loss=1.1293 +[11:13:39][ 549s][FT] round 2/6 epoch 5/5 | patches=98 | loss=1.0466 +[11:13:41][ 551s][AL] round 2/6 | labeled=98 | selection_seconds=9.0000 round_seconds=26.2000 pseudo_student_dice=0.5708 pseudo_ensemble_dice=0.5629 consistency_scale=1.0000 val_dice=0.6499 val_vi=0.4739 val_ece=0.0387 +[11:13:50][ 560s][AL] round 3: 556/5827 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:13:57][ 567s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.518, uniform=0.193, top1=0.518, top2=0.519, top3=0.519 -> using mixture +[11:13:57][ 567s][WEIGHTS] round 3: teacher Dice on 147 labeled patches + 0 blank (3447972 fg px): c-elegans=0.468, cerebellum=0.518, flywire=0.109, berghia=0.178, microns=0.156, oct-glia=0.188, oct-sfl=0.243, snemi=0.502 | weights: c-elegans=0.015, cerebellum=0.858, flywire=0.002, berghia=0.008, microns=0.034, oct-glia=0.017, oct-sfl=0.020, snemi=0.045 +[11:13:59][ 569s][CONS] round 3: on the 49 new patches student Dice=0.369 vs ensemble 0.598 -> pseudo-labels from the ensemble | consistency x1.00 +[11:14:02][ 572s][FT] round 3/6 epoch 1/5 | patches=147 | loss=1.5031 +[11:14:05][ 575s][FT] round 3/6 epoch 2/5 | patches=147 | loss=1.4121 +[11:14:07][ 578s][FT] round 3/6 epoch 3/5 | patches=147 | loss=1.2386 +[11:14:10][ 581s][FT] round 3/6 epoch 4/5 | patches=147 | loss=1.1837 +[11:14:13][ 584s][FT] round 3/6 epoch 5/5 | patches=147 | loss=1.1168 +[11:14:15][ 586s][AL] round 3/6 | labeled=147 | selection_seconds=8.9000 round_seconds=32.2000 pseudo_student_dice=0.3691 pseudo_ensemble_dice=0.5979 consistency_scale=1.0000 val_dice=0.6700 val_vi=0.4160 val_ece=0.0603 +[11:14:24][ 594s][AL] round 4: 811/5778 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:14:33][ 603s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.507, uniform=0.168, top1=0.507, top2=0.508, top3=0.508 -> using mixture +[11:14:33][ 603s][WEIGHTS] round 4: teacher Dice on 196 labeled patches + 0 blank (4475353 fg px): c-elegans=0.452, cerebellum=0.507, flywire=0.093, berghia=0.175, microns=0.140, oct-glia=0.174, oct-sfl=0.225, snemi=0.490 | weights: c-elegans=0.015, cerebellum=0.861, flywire=0.002, berghia=0.009, microns=0.032, oct-glia=0.016, oct-sfl=0.019, snemi=0.045 +[11:14:38][ 608s][CONS] round 4: on the 49 new patches student Dice=0.571 vs ensemble 0.466 -> pseudo-labels from the student | consistency x0.24 +[11:14:42][ 612s][FT] round 4/6 epoch 1/5 | patches=196 | loss=1.1983 +[11:14:45][ 616s][FT] round 4/6 epoch 2/5 | patches=196 | loss=1.1145 +[11:14:49][ 620s][FT] round 4/6 epoch 3/5 | patches=196 | loss=1.0540 +[11:14:53][ 623s][FT] round 4/6 epoch 4/5 | patches=196 | loss=0.9975 +[11:14:57][ 627s][FT] round 4/6 epoch 5/5 | patches=196 | loss=1.0069 +[11:14:59][ 629s][AL] round 4/6 | labeled=196 | selection_seconds=8.9000 round_seconds=41.5000 pseudo_student_dice=0.5709 pseudo_ensemble_dice=0.4664 consistency_scale=0.2360 val_dice=0.6300 val_vi=0.5393 val_ece=0.0166 +[11:15:07][ 638s][AL] round 5: 374/5729 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:15:59][ 690s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.452, cerebellum=0.459, flywire=0.132, berghia=0.303, microns=0.285, oct-glia=0.406, oct-sfl=0.497, snemi=0.405 | blended with labeled Dice at 0.50 +[11:16:48][ 738s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.451, uniform=0.289, top1=0.446, top2=0.446, top3=0.447 -> using mixture +[11:16:48][ 738s][WEIGHTS] round 5: teacher Dice on 245 labeled patches + 0 blank (4632452 fg px): c-elegans=0.422, cerebellum=0.446, flywire=0.112, berghia=0.237, microns=0.214, oct-glia=0.292, oct-sfl=0.363, snemi=0.414 | weights: c-elegans=0.015, cerebellum=0.653, flywire=0.003, berghia=0.024, microns=0.094, oct-glia=0.074, oct-sfl=0.107, snemi=0.029 +[11:16:53][ 743s][CONS] round 5: on the 49 new patches student Dice=0.157 vs ensemble 0.069 -> pseudo-labels from the student | consistency x0.00 +[11:16:55][ 746s][FT] round 5/6 epoch 1/5 | patches=245 | loss=1.0619 +[11:16:58][ 748s][FT] round 5/6 epoch 2/5 | patches=245 | loss=1.0047 +[11:17:00][ 751s][FT] round 5/6 epoch 3/5 | patches=245 | loss=0.9691 +[11:17:03][ 753s][FT] round 5/6 epoch 4/5 | patches=245 | loss=0.9047 +[11:17:05][ 755s][FT] round 5/6 epoch 5/5 | patches=245 | loss=0.8771 +[11:17:07][ 757s][AL] round 5/6 | labeled=245 | selection_seconds=8.8000 round_seconds=126.3000 pseudo_student_dice=0.1569 pseudo_ensemble_dice=0.0693 consistency_scale=0.0000 val_dice=0.6217 val_vi=0.5293 val_ece=0.0085 +[11:17:15][ 766s][AL] round 6: 298/5680 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:18:08][ 819s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.459, cerebellum=0.467, flywire=0.130, berghia=0.303, microns=0.278, oct-glia=0.400, oct-sfl=0.490, snemi=0.427 | blended with labeled Dice at 0.50 +[11:18:58][ 869s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.443, uniform=0.283, top1=0.439, top2=0.439, top3=0.440 -> using mixture +[11:18:58][ 869s][WEIGHTS] round 6: teacher Dice on 294 labeled patches + 0 blank (5289236 fg px): c-elegans=0.420, cerebellum=0.439, flywire=0.109, berghia=0.254, microns=0.206, oct-glia=0.296, oct-sfl=0.361, snemi=0.417 | weights: c-elegans=0.016, cerebellum=0.639, flywire=0.004, berghia=0.030, microns=0.091, oct-glia=0.081, oct-sfl=0.109, snemi=0.032 +[11:19:03][ 874s][CONS] round 6: on the 49 new patches student Dice=0.439 vs ensemble 0.169 -> pseudo-labels from the student | consistency x0.00 +[11:19:06][ 876s][FT] round 6/6 epoch 1/5 | patches=294 | loss=1.1014 +[11:19:09][ 879s][FT] round 6/6 epoch 2/5 | patches=294 | loss=1.0341 +[11:19:12][ 882s][FT] round 6/6 epoch 3/5 | patches=294 | loss=0.9358 +[11:19:15][ 885s][FT] round 6/6 epoch 4/5 | patches=294 | loss=0.9625 +[11:19:17][ 888s][FT] round 6/6 epoch 5/5 | patches=294 | loss=0.8936 +[11:19:19][ 890s][AL] round 6/6 | labeled=294 | selection_seconds=8.8000 round_seconds=130.4000 pseudo_student_dice=0.4388 pseudo_ensemble_dice=0.1688 consistency_scale=0.0000 val_dice=0.7115 val_vi=0.3985 val_ece=0.0027 +[11:19:22][ 893s][CONS] final consolidation: pseudo-labels refreshed from the student +[11:19:25][ 896s][FT] final 1/10 epoch 1/1 | patches=294 | loss=0.9068 +[11:19:28][ 898s][FT] final 2/10 epoch 1/1 | patches=294 | loss=0.9436 +[11:19:31][ 901s][FT] final 3/10 epoch 1/1 | patches=294 | loss=0.8761 +[11:19:34][ 904s][FT] final 4/10 epoch 1/1 | patches=294 | loss=0.9009 +[11:19:36][ 907s][FT] final 5/10 epoch 1/1 | patches=294 | loss=0.7947 +[11:19:39][ 910s][FT] final 6/10 epoch 1/1 | patches=294 | loss=0.7844 +[11:19:42][ 913s][FT] final 7/10 epoch 1/1 | patches=294 | loss=0.7615 +[11:19:45][ 915s][FT] final 8/10 epoch 1/1 | patches=294 | loss=0.7173 +[11:19:48][ 918s][FT] final 9/10 epoch 1/1 | patches=294 | loss=0.7306 +[11:19:50][ 921s][FT] final 10/10 epoch 1/1 | patches=294 | loss=0.7398 +[11:19:52][ 923s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=31.1000 val_dice=0.7158 val_vi=0.3846 val_ece=0.0160 +[11:19:52][ 923s][WEIGHTS] final: c-elegans=0.016, cerebellum=0.639, flywire=0.004, berghia=0.030, microns=0.091, oct-glia=0.081, oct-sfl=0.109, snemi=0.032 +[11:20:10][ 941s][TEST] raw student (no EMA, no TTA): patch_dice=0.7032 patch_vi=0.3580 +[11:20:13][ 943s][TEST] final model without TTA: patch_dice=0.7032 patch_vi=0.3580 +[11:20:13][ 943s][RESOURCES] {"phases": {"setup": {"seconds": 1.13, "peak_cpu_rss_mb": 1027.7, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 31.55, "peak_cpu_rss_mb": 1809.5, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 328.93, "peak_cpu_rss_mb": 3677.4, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 559.03, "peak_cpu_rss_mb": 3677.4, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4646.0, "rounds": 6}, "inference": {"seconds": 20.19, "peak_cpu_rss_mb": 3677.4, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4646.0, "patches": 1300, "images": 52, "patches_per_second": 64.4, "seconds_per_image": 0.388}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 3677.4, "total_seconds": 940.8} +[11:20:13][ 943s][TEST] patch_dice=0.7254 patch_iou=0.6456 patch_precision=0.7365 patch_recall=0.8469 patch_vi=0.3307 image_dice=0.8376 image_iou=0.7339 image_precision=0.8064 image_recall=0.8949 image_vi=0.3832 cal_ece=0.0033 cal_nll=0.1739 cal_brier=0.0503 cal_mean_confidence=0.9339 cal_accuracy=0.9333 cal_predicted_prevalence=0.2312 cal_true_prevalence=0.2108 cal_mean_probability=0.2435 +[11:20:13][ 943s][DONE] whole-mouse-brain macs-v11-norestart-acq-margin b5 s3 | test patch Dice=0.7254 VI=0.3307 | image Dice=0.8376 VI=0.3832 | 943s diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/result.json b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..455036577996fbc5dab2dfb098b6b35d82124fb5 --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/result.json @@ -0,0 +1,767 @@ +{ + "target": "whole-mouse-brain", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 3, + "config": { + "task": "membrane", + "target": "whole-mouse-brain", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "margin", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "whole-mouse-brain", + "--variants", + "macs-v11-norestart-acq-margin", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d" + ], + "alphas": [ + 0.007617762850063339, + 0.2559312601209969, + 0.03865993944852264, + 0.07533345106002441, + 0.3764169834180784, + 0.13449516240019677, + 0.09581340232315401, + 0.0157320383789636 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d" + ], + "mixture": [ + 0.007617762850063339, + 0.2559312601209969, + 0.03865993944852264, + 0.07533345106002441, + 0.3764169834180784, + 0.13449516240019677, + 0.09581340232315401, + 0.0157320383789636 + ], + "per_source": [ + 0.1450746039368088, + 0.041950766450651496, + 0.0008814721099573598, + 0.1418325124471684, + 0.02035312744213396, + 0.4023730240336797, + 0.24338555075976573, + 0.004148942819834595 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.11311670518772092, + "mmd2_mixture": 0.12945823384265215, + "mmd2_uniform": 0.155288743679944, + "mmd2_single_source": [ + 0.38071639717042494, + 0.17697894982207135, + 0.3194061281637849, + 0.24037642211575033, + 0.16266554797152122, + 0.2734526560532348, + 0.2563986706812869, + 0.2660414632502267 + ], + "median_squared_distance": 0.18592532667757627, + "gammas": [ + 1.3446258477390725, + 2.689251695478145, + 5.37850339095629, + 10.75700678191258, + 21.51401356382516 + ], + "max_weight": 0.3764169834180784, + "entropy": 1.6341528654931232, + "effective_sources": 5.125114496180783, + "min_weight": 0.007617762850063339 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.33220085501273156, + 0.46442573060595516, + 0.8760630244687474, + 0.33460943356720163, + 0.5415026072699861, + 0.2234870921926152, + 0.2770625113402161, + 0.7109869456345189 + ], + "standardized": [ + -0.6467275403932012, + -0.02635183204683673, + 1.9049774780863755, + -0.6354269162093369, + 0.33527922824378936, + -1.1567932698101913, + -0.9054268876938927, + 1.130469739823295 + ], + "tau": 0.5, + "max_weight": 0.4023730240336797, + "entropy": 1.508581788687784, + "effective_sources": 4.520315483702711, + "min_weight": 0.0008814721099573598 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "seconds": 6.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.38573376502042933, + 0.31643357804427197, + 0.10997258971952562, + 0.2588485480329741, + 0.3288178379711084, + 0.4072116167356654, + 0.4397876396557443, + 0.30535999081848075 + ], + "ensemble_val_dice": { + "mixture": 0.3845088921696238, + "mixture@raw": 0.3998907914691259, + "mixture@logit": 0.3802050170706561, + "per_source": 0.4199022574247113, + "per_source@raw": 0.42186088103254527, + "per_source@logit": 0.41929713845028616, + "uniform": 0.4128319943503538, + "uniform@raw": 0.3877140229453584, + "uniform@logit": 0.4054972948683435 + }, + "val_patches": 1024 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 2.6421806812286377, + 3.5461363792419434 + ], + "top_source": "microns_basil", + "weight_stats": { + "max_weight": 0.3764169834180784, + "entropy": 1.6341528654931232, + "effective_sources": 5.125114496180783, + "min_weight": 0.007617762850063339 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.261904761904762, + "spearman_weight_taxonomy": 0.39440531887330776, + "top_source": "microns_basil", + "top_source_is_best_teacher": false + }, + "init_teacher": "microns_basil", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 328.9, + "round0_val": { + "val_dice": 0.41606177435746006, + "val_vi": 0.5123552235778059, + "val_ece": 0.14459192201571608 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 49, + "selection_seconds": 32.3, + "round_seconds": 157.5, + "teacher_dice_labeled": [ + 0.36738741011177045, + 0.34973487377165097, + 0.14100291682142935, + 0.2353012898060949, + 0.30996524321190877, + 0.38452188964078277, + 0.47067622607709797, + 0.25385926187706875 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 640544, + "evidence_nonempty_patches": 39, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.3843282859342692, + "restart_teacher_dice": 0.47067622607709797, + "weights": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.43280781937383384, + "uniform": 0.43493152236966137, + "top1": 0.47067622607709797, + "top2": 0.44454578875106776, + "top3": 0.4462152205328611 + }, + "val_dice": 0.6423608970665755, + "val_vi": 0.4583910108960369, + "val_ece": 0.018296731206981286 + }, + { + "round": 2, + "n_labeled": 98, + "selection_seconds": 9.0, + "round_seconds": 26.2, + "teacher_dice_labeled": [ + 0.4790010720265582, + 0.4471774808650101, + 0.1635805331835569, + 0.2126323076912074, + 0.22214865194975508, + 0.2548709404689054, + 0.3297295554084256, + 0.43233750559567696 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1973085, + "evidence_nonempty_patches": 84, + "first_correction": false, + "restart": "", + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.44770498278126886, + "uniform": 0.2748470197475036, + "top1": 0.4790010720265582, + "top2": 0.4485533237171078, + "top3": 0.4488417346971152 + }, + "pseudo_student_dice": 0.5707606242464341, + "pseudo_ensemble_dice": 0.5629199461807312, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.6498913510594475, + "val_vi": 0.4739385598038719, + "val_ece": 0.03870070742571032 + }, + { + "round": 3, + "n_labeled": 147, + "selection_seconds": 8.9, + "round_seconds": 32.2, + "teacher_dice_labeled": [ + 0.4675352679158566, + 0.51830134874921, + 0.10851799996363382, + 0.17803503365781362, + 0.1555896524596807, + 0.18820785764113224, + 0.24280354165733387, + 0.5021665894135083 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 3447972, + "evidence_nonempty_patches": 121, + "first_correction": false, + "restart": "", + "weights": [ + 0.015380486145659464, + 0.8584998767475599, + 0.0021538308841062364, + 0.008410986483131318, + 0.033577567429096726, + 0.016624391153563504, + 0.020444259760168546, + 0.04490860139671429 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5183846978967368, + "uniform": 0.1932272503274666, + "top1": 0.51830134874921, + "top2": 0.5188958884279018, + "top3": 0.5190050811393596 + }, + "pseudo_student_dice": 0.3690770983734586, + "pseudo_ensemble_dice": 0.5979417700316343, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.6700320030612286, + "val_vi": 0.41596008318263494, + "val_ece": 0.0603106358419856 + }, + { + "round": 4, + "n_labeled": 196, + "selection_seconds": 8.9, + "round_seconds": 41.5, + "teacher_dice_labeled": [ + 0.4521817725179459, + 0.5070976087675007, + 0.09348266121186623, + 0.17504878863175116, + 0.14020116045283942, + 0.17422245803812333, + 0.22538984408502558, + 0.490394045937209 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 4475353, + "evidence_nonempty_patches": 163, + "first_correction": false, + "restart": "", + "weights": [ + 0.014804203531187304, + 0.8613454825173498, + 0.0020797365351979775, + 0.009161618543295417, + 0.03230816007313526, + 0.01622192158502868, + 0.01927698149206271, + 0.04480189572274281 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.506573562715877, + "uniform": 0.16827833119491115, + "top1": 0.5070976087675007, + "top2": 0.5079794458246457, + "top3": 0.5080628048620779 + }, + "pseudo_student_dice": 0.5709141779489735, + "pseudo_ensemble_dice": 0.4664405385589357, + "pseudo_source": "student", + "consistency_scale": 0.236, + "val_dice": 0.6299671169810852, + "val_vi": 0.5393323978378546, + "val_ece": 0.01660194871051421 + }, + { + "round": 5, + "n_labeled": 245, + "selection_seconds": 8.8, + "round_seconds": 126.3, + "teacher_dice_labeled": [ + 0.4222786728452023, + 0.44601629440755297, + 0.11182095835834197, + 0.23742002382343114, + 0.2138810167448923, + 0.29239287155867766, + 0.36293542540080537, + 0.41378682372365055 + ], + "teacher_dice_pool": [ + 0.45211503388520596, + 0.4586268562508522, + 0.13206520212610942, + 0.3034451235507459, + 0.2853314220634787, + 0.4058655940805688, + 0.4968728262174501, + 0.40482826965176766 + ], + "evidence_fg_pixels": 4632452, + "evidence_nonempty_patches": 192, + "first_correction": false, + "restart": "", + "weights": [ + 0.015338916240729503, + 0.6534039312404931, + 0.003490821758896233, + 0.02388494131779684, + 0.09431416337331966, + 0.07389023085416159, + 0.10657836184751224, + 0.02909863336709079 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.450717108090974, + "uniform": 0.28857802117043396, + "top1": 0.44601629440755297, + "top2": 0.44629305482704884, + "top3": 0.44681775473389657 + }, + "pseudo_student_dice": 0.15691007580540497, + "pseudo_ensemble_dice": 0.06933827105338418, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.6217327713376608, + "val_vi": 0.5292694280761847, + "val_ece": 0.008471389509414293 + }, + { + "round": 6, + "n_labeled": 294, + "selection_seconds": 8.8, + "round_seconds": 130.4, + "teacher_dice_labeled": [ + 0.4204763259856441, + 0.4391522114337465, + 0.10897358309180741, + 0.2540118077271043, + 0.20582054260607813, + 0.296415072255611, + 0.36062790679412976, + 0.4173540295143803 + ], + "teacher_dice_pool": [ + 0.4590040784121988, + 0.4673811142986949, + 0.12976248045272956, + 0.3033754435936046, + 0.2781588715045815, + 0.40047257679930554, + 0.490289506131676, + 0.42686846374501874 + ], + "evidence_fg_pixels": 5289236, + "evidence_nonempty_patches": 229, + "first_correction": false, + "restart": "", + "weights": [ + 0.01577566582337092, + 0.6388397404019729, + 0.003552893550954598, + 0.02952581508890634, + 0.09111528485345537, + 0.08055188987539245, + 0.10906066204483804, + 0.03157804836110938 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.44275603813402087, + "uniform": 0.28250357953908506, + "top1": 0.4391522114337465, + "top2": 0.4394290706035419, + "top3": 0.4402034746697075 + }, + "pseudo_student_dice": 0.43881066214162834, + "pseudo_ensemble_dice": 0.16880114668182505, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.7114711579891998, + "val_vi": 0.3985485231582955, + "val_ece": 0.002656605495560469 + }, + { + "round": 7, + "n_labeled": 294, + "selection_seconds": 0.0, + "round_seconds": 31.1, + "final_consolidation": true, + "val_dice": 0.7158009490992879, + "val_vi": 0.38461948070584473, + "val_ece": 0.016036664127452485 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.24881207404064967, + "auroc_top20_error": 0.3594799622567608 + }, + "al_seconds": 559.0, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.01577566582337092, + 0.6388397404019729, + 0.003552893550954598, + 0.02952581508890634, + 0.09111528485345537, + 0.08055188987539245, + 0.10906066204483804, + 0.03157804836110938 + ], + "weight_stats_final": { + "max_weight": 0.6388397404019729, + "entropy": 1.24771609642705, + "effective_sources": 3.4823804469747293, + "min_weight": 0.003552893550954598 + }, + "top_source_final": "dhanyasi-P14-mouse-cerebellum", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7031837305499606, + "patch_iou": 0.6186873385247618, + "patch_precision": 0.7015658527093758, + "patch_recall": 0.8501698170797606, + "patch_vi": 0.35798487868286893, + "n_patches": 1300, + "cal_ece": 0.014296868795295074, + "cal_nll": 0.19317754581198462, + "cal_brier": 0.055458809854427006, + "cal_mean_confidence": 0.9406963579779875, + "cal_accuracy": 0.9263994891826923, + "cal_predicted_prevalence": 0.2381670673076923, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.2453708863664394 + }, + "test_notta": { + "patch_dice": 0.7031837305499606, + "patch_iou": 0.6186873385247618, + "patch_precision": 0.7015658527093758, + "patch_recall": 0.8501698170797606, + "patch_vi": 0.35798487868286893, + "n_patches": 1300, + "cal_ece": 0.014296868795295074, + "cal_nll": 0.19317754581198462, + "cal_brier": 0.055458809854427006, + "cal_mean_confidence": 0.9406963579779875, + "cal_accuracy": 0.9263994891826923, + "cal_predicted_prevalence": 0.2381670673076923, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.2453708863664394 + }, + "test": { + "patch_dice": 0.7253981718195581, + "patch_iou": 0.6455533937479556, + "patch_precision": 0.7364661124292978, + "patch_recall": 0.8468879653434368, + "patch_vi": 0.33066736927078855, + "n_patches": 1300, + "image_dice": 0.8376132017999095, + "image_iou": 0.7338580259277356, + "image_precision": 0.8064019432082945, + "image_recall": 0.8948911544055365, + "image_vi": 0.3832235482166312, + "n_images": 52, + "cal_ece": 0.0032693595258011564, + "cal_nll": 0.17391863248092237, + "cal_brier": 0.05027658671011288, + "cal_mean_confidence": 0.9338601700799212, + "cal_accuracy": 0.9332977764423077, + "cal_predicted_prevalence": 0.23124924879807693, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.2435347047852322 + }, + "eval_seconds": 20.2, + "total_seconds": 943.0, + "resources": { + "phases": { + "setup": { + "seconds": 1.13, + "peak_cpu_rss_mb": 1027.7, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 31.55, + "peak_cpu_rss_mb": 1809.5, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 328.93, + "peak_cpu_rss_mb": 3677.4, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4638.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 559.03, + "peak_cpu_rss_mb": 3677.4, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4646.0, + "rounds": 6 + }, + "inference": { + "seconds": 20.19, + "peak_cpu_rss_mb": 3677.4, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4646.0, + "patches": 1300, + "images": 52, + "patches_per_second": 64.4, + "seconds_per_image": 0.388 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 3677.4, + "total_seconds": 940.8 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/whole-mouse-brain/macs-v11-norestart-acq-margin_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..48b631140ee7d5a798d745be5fdf9df266032c5a --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-margin/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,49,32.3,157.5,"[0.36738741011177045, 0.34973487377165097, 0.14100291682142935, 0.2353012898060949, 0.30996524321190877, 0.38452188964078277, 0.47067622607709797, 0.25385926187706875]",,640544,39,True,kept,0.3843282859342692,0.47067622607709797,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]",top1,"{'mixture': 0.43280781937383384, 'uniform': 0.43493152236966137, 'top1': 0.47067622607709797, 'top2': 0.44454578875106776, 'top3': 0.4462152205328611}",0.6423608970665755,0.4583910108960369,0.018296731206981286,,,,, +2,98,9.0,26.2,"[0.4790010720265582, 0.4471774808650101, 0.1635805331835569, 0.2126323076912074, 0.22214865194975508, 0.2548709404689054, 0.3297295554084256, 0.43233750559567696]",,1973085,84,False,,,,"[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.44770498278126886, 'uniform': 0.2748470197475036, 'top1': 0.4790010720265582, 'top2': 0.4485533237171078, 'top3': 0.4488417346971152}",0.6498913510594475,0.4739385598038719,0.03870070742571032,0.5707606242464341,0.5629199461807312,ensemble,1.0, +3,147,8.9,32.2,"[0.4675352679158566, 0.51830134874921, 0.10851799996363382, 0.17803503365781362, 0.1555896524596807, 0.18820785764113224, 0.24280354165733387, 0.5021665894135083]",,3447972,121,False,,,,"[0.015380486145659464, 0.8584998767475599, 0.0021538308841062364, 0.008410986483131318, 0.033577567429096726, 0.016624391153563504, 0.020444259760168546, 0.04490860139671429]",mixture,"{'mixture': 0.5183846978967368, 'uniform': 0.1932272503274666, 'top1': 0.51830134874921, 'top2': 0.5188958884279018, 'top3': 0.5190050811393596}",0.6700320030612286,0.41596008318263494,0.0603106358419856,0.3690770983734586,0.5979417700316343,ensemble,1.0, +4,196,8.9,41.5,"[0.4521817725179459, 0.5070976087675007, 0.09348266121186623, 0.17504878863175116, 0.14020116045283942, 0.17422245803812333, 0.22538984408502558, 0.490394045937209]",,4475353,163,False,,,,"[0.014804203531187304, 0.8613454825173498, 0.0020797365351979775, 0.009161618543295417, 0.03230816007313526, 0.01622192158502868, 0.01927698149206271, 0.04480189572274281]",mixture,"{'mixture': 0.506573562715877, 'uniform': 0.16827833119491115, 'top1': 0.5070976087675007, 'top2': 0.5079794458246457, 'top3': 0.5080628048620779}",0.6299671169810852,0.5393323978378546,0.01660194871051421,0.5709141779489735,0.4664405385589357,student,0.236, +5,245,8.8,126.3,"[0.4222786728452023, 0.44601629440755297, 0.11182095835834197, 0.23742002382343114, 0.2138810167448923, 0.29239287155867766, 0.36293542540080537, 0.41378682372365055]","[0.45211503388520596, 0.4586268562508522, 0.13206520212610942, 0.3034451235507459, 0.2853314220634787, 0.4058655940805688, 0.4968728262174501, 0.40482826965176766]",4632452,192,False,,,,"[0.015338916240729503, 0.6534039312404931, 0.003490821758896233, 0.02388494131779684, 0.09431416337331966, 0.07389023085416159, 0.10657836184751224, 0.02909863336709079]",mixture,"{'mixture': 0.450717108090974, 'uniform': 0.28857802117043396, 'top1': 0.44601629440755297, 'top2': 0.44629305482704884, 'top3': 0.44681775473389657}",0.6217327713376608,0.5292694280761847,0.008471389509414293,0.15691007580540497,0.06933827105338418,student,0.0, +6,294,8.8,130.4,"[0.4204763259856441, 0.4391522114337465, 0.10897358309180741, 0.2540118077271043, 0.20582054260607813, 0.296415072255611, 0.36062790679412976, 0.4173540295143803]","[0.4590040784121988, 0.4673811142986949, 0.12976248045272956, 0.3033754435936046, 0.2781588715045815, 0.40047257679930554, 0.490289506131676, 0.42686846374501874]",5289236,229,False,,,,"[0.01577566582337092, 0.6388397404019729, 0.003552893550954598, 0.02952581508890634, 0.09111528485345537, 0.08055188987539245, 0.10906066204483804, 0.03157804836110938]",mixture,"{'mixture': 0.44275603813402087, 'uniform': 0.28250357953908506, 'top1': 0.4391522114337465, 'top2': 0.4394290706035419, 'top3': 0.4402034746697075}",0.7114711579891998,0.3985485231582955,0.002656605495560469,0.43881066214162834,0.16880114668182505,student,0.0, +7,294,0.0,31.1,,,,,,,,,,,,0.7158009490992879,0.38461948070584473,0.016036664127452485,,,,,True diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/log.txt b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..e369c6538225c1ce8da3d5e47324735605e0e6d1 --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/log.txt @@ -0,0 +1,108 @@ +[05:58:12][ 0s][START] target=whole-mouse-brain variant=macs-v11-norestart-acq-mcdropout budget=5.0% seed=2 | {"task": "membrane", "target": "whole-mouse-brain", "variant": "macs-v11-norestart-acq-mcdropout", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "mcdropout", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[05:58:12][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[05:58:12][ 0s][DATA] whole-mouse-brain: train=5925 val=1250 test=1300 patches +[05:58:14][ 1s][CALIB] source-val temperatures: c-elegans=2.58, cerebellum=4.56, flywire=1.99, berghia=8.00, microns=1.92, oct-glia=2.55, oct-sfl=2.64, snemi=3.55 +[05:58:14][ 2s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 1s +[05:58:20][ 8s][WEIGHTS] mixture: c-elegans=0.008, cerebellum=0.256, flywire=0.039, berghia=0.075, microns=0.376, oct-glia=0.134, oct-sfl=0.096, snemi=0.016 | eff.sources=5.13 | MMD2 mix=0.1295 uniform=0.1553 +[05:58:20][ 8s][WEIGHTS] per-source: c-elegans=0.145, cerebellum=0.042, flywire=0.001, berghia=0.142, microns=0.020, oct-glia=0.402, oct-sfl=0.243, snemi=0.004 | min-MMD source (NeuroADDA rule) = oct-glia +[05:58:44][ 32s][TEACHERS] target-val Dice: c-elegans=0.386, cerebellum=0.316, flywire=0.110, berghia=0.259, microns=0.329, oct-glia=0.407, oct-sfl=0.440, snemi=0.305 +[05:58:44][ 32s][TEACHERS] ensemble target-val Dice: mixture=0.385, mixture@raw=0.400, mixture@logit=0.380, per_source=0.420, per_source@raw=0.422, per_source@logit=0.419, uniform=0.413, uniform@raw=0.388, uniform@logit=0.405 +[05:58:44][ 32s][WEIGHTS] using mixture: c-elegans=0.008, cerebellum=0.256, flywire=0.039, berghia=0.075, microns=0.376, oct-glia=0.134, oct-sfl=0.096, snemi=0.016 +[05:58:44][ 32s][INIT] student initialised from teacher microns (weight 0.376) +[05:58:45][ 33s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[05:58:45][ 33s][KD] distilling the mixture ensemble (prob space) into the student +[05:59:53][ 100s][KD] soft labels for 5925 patches from 8 teachers in 67s +[06:00:45][ 153s][KD] epoch 1/5 | patches=5925 | loss=0.6663 | 52s +[06:01:38][ 205s][KD] epoch 2/5 | patches=5925 | loss=0.4936 | 52s +[06:02:30][ 258s][KD] epoch 3/5 | patches=5925 | loss=0.4654 | 52s +[06:03:21][ 309s][KD] epoch 4/5 | patches=5925 | loss=0.4610 | 51s +[06:04:13][ 360s][KD] epoch 5/5 | patches=5925 | loss=0.4566 | 52s +[06:04:15][ 362s][ROUND0] before any target label: val_dice=0.3948 val_vi=0.5092 +[06:04:15][ 362s][AL] patches=5925 (pool 5925 after dropping blank; free blank negatives: True) | budget=296 (5.0%) | rounds=6 | per_round=49 | acquisition=mcdropout | first_round=stratified +[06:04:26][ 373s][AL] round-1 uncertainty quality on 5925 patches: {'spearman_uncertainty_error': -0.20304753183898672, 'auroc_top20_error': 0.41592640068365117} +[06:04:49][ 396s][AL] round 1: stratified cold start: 25 patches from the membrane-rich half (ensemble prevalence >= 0.213) + 24 from the membrane-poor half of 2000 candidates +[06:04:51][ 399s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.417, uniform=0.396, top1=0.453, top2=0.456, top3=0.424 -> using top2 +[06:04:52][ 399s][WEIGHTS] round 1: student kept (labeled Dice 0.302 vs best teacher oct-sfl 0.453); distilling the corrected ensemble into it +[06:04:52][ 399s][WEIGHTS] round 1: teacher Dice on 49 labeled patches + 0 blank (745730 fg px): c-elegans=0.417, cerebellum=0.359, flywire=0.153, berghia=0.310, microns=0.248, oct-glia=0.404, oct-sfl=0.453, snemi=0.315 | weights: c-elegans=0.052, cerebellum=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.948, snemi=0.000 +[06:05:09][ 417s][KD] soft labels for 5876 patches from 8 teachers in 17s +[06:06:00][ 468s][KD] epoch 1/2 | patches=5876 | loss=0.2661 | 51s +[06:06:52][ 520s][KD] epoch 2/2 | patches=5876 | loss=0.2587 | 52s +[06:06:58][ 525s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 6s +[06:06:59][ 527s][FT] round 1/6 epoch 1/5 | patches=49 | loss=1.4576 +[06:07:00][ 528s][FT] round 1/6 epoch 2/5 | patches=49 | loss=1.2015 +[06:07:02][ 529s][FT] round 1/6 epoch 3/5 | patches=49 | loss=1.1189 +[06:07:03][ 531s][FT] round 1/6 epoch 4/5 | patches=49 | loss=0.9829 +[06:07:04][ 532s][FT] round 1/6 epoch 5/5 | patches=49 | loss=1.0922 +[06:07:06][ 534s][AL] round 1/6 | labeled=49 | selection_seconds=34.0000 round_seconds=169.4000 restart_student_dice=0.3018 restart_teacher_dice=0.4531 val_dice=0.6500 val_vi=0.4698 val_ece=0.0157 +[06:07:17][ 545s][AL] round 2: 477/5876 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[06:07:22][ 549s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.411, uniform=0.230, top1=0.414, top2=0.414, top3=0.414 -> using mixture +[06:07:22][ 549s][WEIGHTS] round 2: teacher Dice on 98 labeled patches + 0 blank (1552702 fg px): c-elegans=0.366, cerebellum=0.414, flywire=0.094, berghia=0.242, microns=0.154, oct-glia=0.249, oct-sfl=0.298, snemi=0.355 | weights: c-elegans=0.013, cerebellum=0.695, flywire=0.004, berghia=0.037, microns=0.076, oct-glia=0.070, oct-sfl=0.082, snemi=0.024 +[06:07:27][ 555s][CONS] round 2: on the 49 new patches student Dice=0.533 vs ensemble 0.450 -> pseudo-labels from the student | consistency x0.11 +[06:07:29][ 557s][FT] round 2/6 epoch 1/5 | patches=98 | loss=1.2128 +[06:07:31][ 559s][FT] round 2/6 epoch 2/5 | patches=98 | loss=1.1617 +[06:07:33][ 561s][FT] round 2/6 epoch 3/5 | patches=98 | loss=1.1708 +[06:07:35][ 563s][FT] round 2/6 epoch 4/5 | patches=98 | loss=1.1620 +[06:07:37][ 565s][FT] round 2/6 epoch 5/5 | patches=98 | loss=1.0921 +[06:07:39][ 567s][AL] round 2/6 | labeled=98 | selection_seconds=10.9000 round_seconds=31.3000 pseudo_student_dice=0.5334 pseudo_ensemble_dice=0.4504 consistency_scale=0.1110 val_dice=0.6392 val_vi=0.5097 val_ece=0.0480 +[06:07:50][ 578s][AL] round 3: 542/5827 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[06:08:40][ 628s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.425, cerebellum=0.432, flywire=0.134, berghia=0.279, microns=0.286, oct-glia=0.402, oct-sfl=0.503, snemi=0.374 | blended with labeled Dice at 0.50 +[06:09:26][ 674s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.406, uniform=0.302, top1=0.385, top2=0.401, top3=0.401 -> using mixture +[06:09:26][ 674s][WEIGHTS] round 3: teacher Dice on 147 labeled patches + 0 blank (2045013 fg px): c-elegans=0.368, cerebellum=0.385, flywire=0.104, berghia=0.247, microns=0.216, oct-glia=0.308, oct-sfl=0.377, snemi=0.350 | weights: c-elegans=0.012, cerebellum=0.497, flywire=0.005, berghia=0.037, microns=0.135, oct-glia=0.121, oct-sfl=0.171, snemi=0.021 +[06:09:31][ 679s][CONS] round 3: on the 49 new patches student Dice=0.321 vs ensemble 0.316 -> pseudo-labels from the student | consistency x0.00 +[06:09:33][ 681s][FT] round 3/6 epoch 1/5 | patches=147 | loss=1.1314 +[06:09:34][ 682s][FT] round 3/6 epoch 2/5 | patches=147 | loss=1.1234 +[06:09:36][ 684s][FT] round 3/6 epoch 3/5 | patches=147 | loss=1.0685 +[06:09:37][ 685s][FT] round 3/6 epoch 4/5 | patches=147 | loss=1.0192 +[06:09:39][ 687s][FT] round 3/6 epoch 5/5 | patches=147 | loss=1.0596 +[06:09:41][ 688s][AL] round 3/6 | labeled=147 | selection_seconds=10.8000 round_seconds=119.4000 pseudo_student_dice=0.3206 pseudo_ensemble_dice=0.3164 consistency_scale=0.0000 val_dice=0.6527 val_vi=0.4833 val_ece=0.0120 +[06:09:51][ 699s][AL] round 4: 486/5778 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[06:10:42][ 750s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.429, cerebellum=0.420, flywire=0.147, berghia=0.276, microns=0.315, oct-glia=0.427, oct-sfl=0.522, snemi=0.351 | blended with labeled Dice at 0.50 +[06:11:30][ 798s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.375, uniform=0.314, top1=0.387, top2=0.390, top3=0.394 -> using top3 +[06:11:30][ 798s][WEIGHTS] round 4: teacher Dice on 196 labeled patches + 0 blank (2546814 fg px): c-elegans=0.367, cerebellum=0.364, flywire=0.108, berghia=0.253, microns=0.226, oct-glia=0.320, oct-sfl=0.387, snemi=0.331 | weights: c-elegans=0.020, cerebellum=0.665, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.315, snemi=0.000 +[06:11:35][ 803s][CONS] round 4: on the 49 new patches student Dice=0.372 vs ensemble 0.243 -> pseudo-labels from the student | consistency x0.00 +[06:11:37][ 805s][FT] round 4/6 epoch 1/5 | patches=196 | loss=1.1221 +[06:11:39][ 806s][FT] round 4/6 epoch 2/5 | patches=196 | loss=1.1144 +[06:11:41][ 808s][FT] round 4/6 epoch 3/5 | patches=196 | loss=1.0672 +[06:11:43][ 810s][FT] round 4/6 epoch 4/5 | patches=196 | loss=1.0260 +[06:11:44][ 812s][FT] round 4/6 epoch 5/5 | patches=196 | loss=1.0100 +[06:11:46][ 814s][AL] round 4/6 | labeled=196 | selection_seconds=10.7000 round_seconds=123.8000 pseudo_student_dice=0.3715 pseudo_ensemble_dice=0.2432 consistency_scale=0.0000 val_dice=0.7141 val_vi=0.3944 val_ece=0.0050 +[06:11:57][ 824s][AL] round 5: 847/5729 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[06:12:49][ 877s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.375, cerebellum=0.383, flywire=0.160, berghia=0.253, microns=0.320, oct-glia=0.437, oct-sfl=0.518, snemi=0.281 | blended with labeled Dice at 0.50 +[06:13:38][ 925s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.360, uniform=0.321, top1=0.384, top2=0.386, top3=0.384 -> using top3 +[06:13:38][ 925s][WEIGHTS] round 5: teacher Dice on 245 labeled patches + 0 blank (3577893 fg px): c-elegans=0.350, cerebellum=0.348, flywire=0.116, berghia=0.262, microns=0.226, oct-glia=0.332, oct-sfl=0.384, snemi=0.313 | weights: c-elegans=0.019, cerebellum=0.639, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.341, snemi=0.000 +[06:13:43][ 930s][CONS] round 5: on the 49 new patches student Dice=0.588 vs ensemble 0.292 -> pseudo-labels from the student | consistency x0.29 +[06:13:47][ 935s][FT] round 5/6 epoch 1/5 | patches=245 | loss=1.0574 +[06:13:52][ 939s][FT] round 5/6 epoch 2/5 | patches=245 | loss=1.0164 +[06:13:56][ 944s][FT] round 5/6 epoch 3/5 | patches=245 | loss=0.9954 +[06:14:01][ 948s][FT] round 5/6 epoch 4/5 | patches=245 | loss=0.9892 +[06:14:05][ 953s][FT] round 5/6 epoch 5/5 | patches=245 | loss=0.9721 +[06:14:07][ 955s][AL] round 5/6 | labeled=245 | selection_seconds=10.6000 round_seconds=138.8000 pseudo_student_dice=0.5885 pseudo_ensemble_dice=0.2919 consistency_scale=0.2950 val_dice=0.7095 val_vi=0.3976 val_ece=0.0023 +[06:14:17][ 965s][AL] round 6: 650/5680 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[06:15:11][ 1018s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.375, cerebellum=0.365, flywire=0.158, berghia=0.262, microns=0.330, oct-glia=0.464, oct-sfl=0.533, snemi=0.281 | blended with labeled Dice at 0.50 +[06:16:00][ 1068s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.365, uniform=0.322, top1=0.385, top2=0.387, top3=0.392 -> using top3 +[06:16:00][ 1068s][WEIGHTS] round 6: teacher Dice on 294 labeled patches + 0 blank (4756463 fg px): c-elegans=0.365, cerebellum=0.359, flywire=0.110, berghia=0.258, microns=0.223, oct-glia=0.340, oct-sfl=0.385, snemi=0.331 | weights: c-elegans=0.021, cerebellum=0.659, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.320, snemi=0.000 +[06:16:06][ 1073s][CONS] round 6: on the 49 new patches student Dice=0.540 vs ensemble 0.578 -> pseudo-labels from the student | consistency x0.13 +[06:16:11][ 1079s][FT] round 6/6 epoch 1/5 | patches=294 | loss=1.0482 +[06:16:16][ 1084s][FT] round 6/6 epoch 2/5 | patches=294 | loss=1.0003 +[06:16:22][ 1089s][FT] round 6/6 epoch 3/5 | patches=294 | loss=0.9577 +[06:16:27][ 1095s][FT] round 6/6 epoch 4/5 | patches=294 | loss=0.9268 +[06:16:32][ 1100s][FT] round 6/6 epoch 5/5 | patches=294 | loss=0.8717 +[06:16:34][ 1102s][AL] round 6/6 | labeled=294 | selection_seconds=10.6000 round_seconds=145.1000 pseudo_student_dice=0.5404 pseudo_ensemble_dice=0.5780 consistency_scale=0.1350 val_dice=0.7143 val_vi=0.4083 val_ece=0.0072 +[06:16:37][ 1105s][CONS] final consolidation: pseudo-labels refreshed from the student +[06:16:43][ 1110s][FT] final 1/10 epoch 1/1 | patches=294 | loss=0.8734 +[06:16:48][ 1116s][FT] final 2/10 epoch 1/1 | patches=294 | loss=0.9310 +[06:16:53][ 1121s][FT] final 3/10 epoch 1/1 | patches=294 | loss=0.8816 +[06:16:59][ 1126s][FT] final 4/10 epoch 1/1 | patches=294 | loss=0.8120 +[06:17:04][ 1132s][FT] final 5/10 epoch 1/1 | patches=294 | loss=0.7529 +[06:17:09][ 1137s][FT] final 6/10 epoch 1/1 | patches=294 | loss=0.7757 +[06:17:15][ 1142s][FT] final 7/10 epoch 1/1 | patches=294 | loss=0.7468 +[06:17:20][ 1148s][FT] final 8/10 epoch 1/1 | patches=294 | loss=0.7143 +[06:17:25][ 1153s][FT] final 9/10 epoch 1/1 | patches=294 | loss=0.6793 +[06:17:31][ 1158s][FT] final 10/10 epoch 1/1 | patches=294 | loss=0.7078 +[06:17:33][ 1161s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=56.6000 val_dice=0.7309 val_vi=0.3714 val_ece=0.0158 +[06:17:33][ 1161s][WEIGHTS] final: c-elegans=0.021, cerebellum=0.659, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.320, snemi=0.000 +[06:17:51][ 1178s][TEST] raw student (no EMA, no TTA): patch_dice=0.7117 patch_vi=0.3410 +[06:17:53][ 1181s][TEST] final model without TTA: patch_dice=0.7117 patch_vi=0.3410 +[06:17:53][ 1181s][RESOURCES] {"phases": {"setup": {"seconds": 1.15, "peak_cpu_rss_mb": 1027.9, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 31.61, "peak_cpu_rss_mb": 1801.2, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 327.55, "peak_cpu_rss_mb": 3676.1, "peak_gpu_alloc_mb": 4303.7, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 798.09, "peak_cpu_rss_mb": 3676.1, "peak_gpu_alloc_mb": 4424.0, "peak_gpu_reserved_mb": 4648.0, "rounds": 6}, "inference": {"seconds": 20.38, "peak_cpu_rss_mb": 3676.1, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4648.0, "patches": 1300, "images": 52, "patches_per_second": 63.8, "seconds_per_image": 0.392}}, "peak_gpu_alloc_mb": 4424.0, "peak_cpu_rss_mb": 3676.1, "total_seconds": 1178.8} +[06:17:53][ 1181s][TEST] patch_dice=0.7329 patch_iou=0.6539 patch_precision=0.7795 patch_recall=0.8102 patch_vi=0.3151 image_dice=0.8398 image_iou=0.7371 image_precision=0.8421 image_recall=0.8598 image_vi=0.3651 cal_ece=0.0033 cal_nll=0.1590 cal_brier=0.0466 cal_mean_confidence=0.9397 cal_accuracy=0.9366 cal_predicted_prevalence=0.2156 cal_true_prevalence=0.2108 cal_mean_probability=0.2234 +[06:17:53][ 1181s][DONE] whole-mouse-brain macs-v11-norestart-acq-mcdropout b5 s2 | test patch Dice=0.7329 VI=0.3151 | image Dice=0.8398 VI=0.3651 | 1181s diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/result.json b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..05e721ca3eea63006701739ed66dbca396af17ba --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/result.json @@ -0,0 +1,785 @@ +{ + "target": "whole-mouse-brain", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 2, + "config": { + "task": "membrane", + "target": "whole-mouse-brain", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "mcdropout", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "whole-mouse-brain", + "--variants", + "macs-v11-norestart-acq-mcdropout", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d" + ], + "alphas": [ + 0.007617762850063339, + 0.2559312601209969, + 0.03865993944852264, + 0.07533345106002441, + 0.3764169834180784, + 0.13449516240019677, + 0.09581340232315401, + 0.0157320383789636 + ], + "weights_all": { + "sources": [ + "c-elegans-dauer-stage", + "dhanyasi-P14-mouse-cerebellum", + "flywire", + "katz-lab-berghia-connective", + "microns_basil", + "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "octopus-vulgaris-vertical-lobe-sfltract", + "snemi3d" + ], + "mixture": [ + 0.007617762850063339, + 0.2559312601209969, + 0.03865993944852264, + 0.07533345106002441, + 0.3764169834180784, + 0.13449516240019677, + 0.09581340232315401, + 0.0157320383789636 + ], + "per_source": [ + 0.1450746039368088, + 0.041950766450651496, + 0.0008814721099573598, + 0.1418325124471684, + 0.02035312744213396, + 0.4023730240336797, + 0.24338555075976573, + 0.004148942819834595 + ], + "uniform": [ + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125, + 0.125 + ], + "mixture_info": { + "objective": 0.11311670518772092, + "mmd2_mixture": 0.12945823384265215, + "mmd2_uniform": 0.155288743679944, + "mmd2_single_source": [ + 0.38071639717042494, + 0.17697894982207135, + 0.3194061281637849, + 0.24037642211575033, + 0.16266554797152122, + 0.2734526560532348, + 0.2563986706812869, + 0.2660414632502267 + ], + "median_squared_distance": 0.18592532667757627, + "gammas": [ + 1.3446258477390725, + 2.689251695478145, + 5.37850339095629, + 10.75700678191258, + 21.51401356382516 + ], + "max_weight": 0.3764169834180784, + "entropy": 1.6341528654931232, + "effective_sources": 5.125114496180783, + "min_weight": 0.007617762850063339 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.33220085501273156, + 0.46442573060595516, + 0.8760630244687474, + 0.33460943356720163, + 0.5415026072699861, + 0.2234870921926152, + 0.2770625113402161, + 0.7109869456345189 + ], + "standardized": [ + -0.6467275403932012, + -0.02635183204683673, + 1.9049774780863755, + -0.6354269162093369, + 0.33527922824378936, + -1.1567932698101913, + -0.9054268876938927, + 1.130469739823295 + ], + "tau": 0.5, + "max_weight": 0.4023730240336797, + "entropy": 1.508581788687784, + "effective_sources": 4.520315483702711, + "min_weight": 0.0008814721099573598 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil", + "seconds": 6.2 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.38573376502042933, + 0.31643357804427197, + 0.10997258971952562, + 0.2588485480329741, + 0.3288178379711084, + 0.4072116167356654, + 0.4397876396557443, + 0.30535999081848075 + ], + "ensemble_val_dice": { + "mixture": 0.3845088921696238, + "mixture@raw": 0.3998907914691259, + "mixture@logit": 0.3802050170706561, + "per_source": 0.4199022574247113, + "per_source@raw": 0.42186088103254527, + "per_source@logit": 0.41929713845028616, + "uniform": 0.4128319943503538, + "uniform@raw": 0.3877140229453584, + "uniform@logit": 0.4054972948683435 + }, + "val_patches": 1024 + }, + "temperatures": [ + 2.575807571411133, + 4.557675361633301, + 1.990083932876587, + 8.0, + 1.9150984287261963, + 2.5513463020324707, + 2.6421806812286377, + 3.5461363792419434 + ], + "top_source": "microns_basil", + "weight_stats": { + "max_weight": 0.3764169834180784, + "entropy": 1.6341528654931232, + "effective_sources": 5.125114496180783, + "min_weight": 0.007617762850063339 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.261904761904762, + "spearman_weight_taxonomy": 0.39440531887330776, + "top_source": "microns_basil", + "top_source_is_best_teacher": false + }, + "init_teacher": "microns_basil", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 327.6, + "round0_val": { + "val_dice": 0.3947513405618871, + "val_vi": 0.5091646853725434, + "val_ece": 0.14075125837595415 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 49, + "selection_seconds": 34.0, + "round_seconds": 169.4, + "teacher_dice_labeled": [ + 0.4166773458661978, + 0.35928987385280325, + 0.153101914721845, + 0.3102874105923009, + 0.24827571594727177, + 0.40443287728637833, + 0.45309617093263127, + 0.3151852254719067 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 745730, + "evidence_nonempty_patches": 43, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.30178618768780596, + "restart_teacher_dice": 0.45309617093263127, + "weights": [ + 0.0523462898796494, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.9476537101203506, + 0.0 + ], + "ensemble_choice": "top2", + "candidate_dice": { + "mixture": 0.41707851191221385, + "uniform": 0.3962786850936092, + "top1": 0.45309617093263127, + "top2": 0.4560007178672816, + "top3": 0.4242840493249194 + }, + "val_dice": 0.649954056799635, + "val_vi": 0.46975670281025395, + "val_ece": 0.015682493902244875 + }, + { + "round": 2, + "n_labeled": 98, + "selection_seconds": 10.9, + "round_seconds": 31.3, + "teacher_dice_labeled": [ + 0.36611084588621456, + 0.41390494215223367, + 0.09410596399488945, + 0.24220693615965957, + 0.15365108848199444, + 0.24885494485948062, + 0.29824941598594545, + 0.355243465280686 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1552702, + "evidence_nonempty_patches": 76, + "first_correction": false, + "restart": "", + "weights": [ + 0.012823011947318825, + 0.6947891955189256, + 0.004286685040611986, + 0.036731834491645354, + 0.07570608809152632, + 0.07008633566128063, + 0.08182206111278438, + 0.023754788135906927 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.41136455031192193, + "uniform": 0.2299235728798188, + "top1": 0.41390494215223367, + "top2": 0.4141551875692704, + "top3": 0.4140180268546756 + }, + "pseudo_student_dice": 0.5334437898129569, + "pseudo_ensemble_dice": 0.4503609960060156, + "pseudo_source": "student", + "consistency_scale": 0.111, + "val_dice": 0.6391891344279756, + "val_vi": 0.5097387246188724, + "val_ece": 0.04795928938682097 + }, + { + "round": 3, + "n_labeled": 147, + "selection_seconds": 10.8, + "round_seconds": 119.4, + "teacher_dice_labeled": [ + 0.36768339021727037, + 0.3852192186783384, + 0.10416450316146425, + 0.24673547320391392, + 0.2164043919966956, + 0.3083610023527579, + 0.37673879727901105, + 0.34984706011994526 + ], + "teacher_dice_pool": [ + 0.4249715484975724, + 0.43163808487226474, + 0.13402683120843648, + 0.27919123067816787, + 0.28581213424874735, + 0.4018351343096836, + 0.5029478154698421, + 0.3738482728359761 + ], + "evidence_fg_pixels": 2045013, + "evidence_nonempty_patches": 111, + "first_correction": false, + "restart": "", + "weights": [ + 0.012424062776149956, + 0.49741245892995734, + 0.004521154756190894, + 0.03665674701240556, + 0.13524121291846142, + 0.12120177924292354, + 0.17107619568913765, + 0.02146638867477356 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.40562761102178846, + "uniform": 0.3019242904620139, + "top1": 0.3852192186783384, + "top2": 0.40122242697931965, + "top3": 0.40084130714444877 + }, + "pseudo_student_dice": 0.3206224561686304, + "pseudo_ensemble_dice": 0.31636739169836314, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.6526551922105499, + "val_vi": 0.48330637836874146, + "val_ece": 0.01201782947328647 + }, + { + "round": 4, + "n_labeled": 196, + "selection_seconds": 10.7, + "round_seconds": 123.8, + "teacher_dice_labeled": [ + 0.3669733595372897, + 0.3639770152858429, + 0.10806844150474738, + 0.25307971683616926, + 0.22584242770873106, + 0.320093417456483, + 0.38747785383420064, + 0.3311470636749818 + ], + "teacher_dice_pool": [ + 0.42873382874298777, + 0.4201025611225556, + 0.1465960471130387, + 0.27571882770491135, + 0.3152241379813523, + 0.4274018046756167, + 0.5223327910218584, + 0.3509337191657171 + ], + "evidence_fg_pixels": 2546814, + "evidence_nonempty_patches": 151, + "first_correction": false, + "restart": "", + "weights": [ + 0.020389457321379638, + 0.6647963031539222, + 0.0, + 0.0, + 0.0, + 0.0, + 0.31481423952469817, + 0.0 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.3754659094523485, + "uniform": 0.31416604860326686, + "top1": 0.38747785383420064, + "top2": 0.3897138845510255, + "top3": 0.39374676780961954 + }, + "pseudo_student_dice": 0.3715079220082423, + "pseudo_ensemble_dice": 0.24320091987757675, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.71405290677071, + "val_vi": 0.3943821589053846, + "val_ece": 0.0050473695531796405 + }, + { + "round": 5, + "n_labeled": 245, + "selection_seconds": 10.6, + "round_seconds": 138.8, + "teacher_dice_labeled": [ + 0.3502083366531371, + 0.34845351115348644, + 0.11604729844495462, + 0.2621757924425775, + 0.22619772716440484, + 0.33203357450989535, + 0.3840164960993779, + 0.313400181441156 + ], + "teacher_dice_pool": [ + 0.37548772339205577, + 0.3833013808355072, + 0.1600921317997613, + 0.25321508996470743, + 0.31990754606797517, + 0.4370099709404248, + 0.5179453131615449, + 0.280774759856714 + ], + "evidence_fg_pixels": 3577893, + "evidence_nonempty_patches": 189, + "first_correction": false, + "restart": "", + "weights": [ + 0.01936135902517909, + 0.6391615107247445, + 0.0, + 0.0, + 0.0, + 0.0, + 0.3414771302500763, + 0.0 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.3599807604482505, + "uniform": 0.3211029686375517, + "top1": 0.3840164960993779, + "top2": 0.38579360695658055, + "top3": 0.3835683473682102 + }, + "pseudo_student_dice": 0.5884572297615537, + "pseudo_ensemble_dice": 0.2919429375125578, + "pseudo_source": "student", + "consistency_scale": 0.295, + "val_dice": 0.7095045778301132, + "val_vi": 0.3975579165871835, + "val_ece": 0.0022580185247888613 + }, + { + "round": 6, + "n_labeled": 294, + "selection_seconds": 10.6, + "round_seconds": 145.1, + "teacher_dice_labeled": [ + 0.3652716200083902, + 0.3585971398683318, + 0.10992564142649246, + 0.2578325740522286, + 0.22277450818970798, + 0.33973973894072296, + 0.3846495074209926, + 0.3305959501728769 + ], + "teacher_dice_pool": [ + 0.37545897608442663, + 0.36517019215223734, + 0.15792099420152225, + 0.2615905184780217, + 0.330487145457808, + 0.46403301342316666, + 0.532613356475908, + 0.28124866584265834 + ], + "evidence_fg_pixels": 4756463, + "evidence_nonempty_patches": 232, + "first_correction": false, + "restart": "", + "weights": [ + 0.020966771040059683, + 0.6589319110869489, + 0.0, + 0.0, + 0.0, + 0.0, + 0.3201013178729914, + 0.0 + ], + "ensemble_choice": "top3", + "candidate_dice": { + "mixture": 0.3652920718151623, + "uniform": 0.3223976533149485, + "top1": 0.3846495074209926, + "top2": 0.38678028399483, + "top3": 0.39248325722874766 + }, + "pseudo_student_dice": 0.5404097731059258, + "pseudo_ensemble_dice": 0.5780078399479068, + "pseudo_source": "student", + "consistency_scale": 0.135, + "val_dice": 0.7143403163792835, + "val_vi": 0.40830035269899356, + "val_ece": 0.007214722442033303 + }, + { + "round": 7, + "n_labeled": 294, + "selection_seconds": 0.0, + "round_seconds": 56.6, + "final_consolidation": true, + "val_dice": 0.7309256869658374, + "val_vi": 0.3713934767695636, + "val_ece": 0.015823260301908625 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.20304753183898672, + "auroc_top20_error": 0.41592640068365117 + }, + "al_seconds": 798.1, + "source_archs": [ + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.020966771040059683, + 0.6589319110869489, + 0.0, + 0.0, + 0.0, + 0.0, + 0.3201013178729914, + 0.0 + ], + "weight_stats_final": { + "max_weight": 0.6589319110869489, + "entropy": 0.7205294126502648, + "effective_sources": 2.055521141531504, + "min_weight": 0.0 + }, + "top_source_final": "dhanyasi-P14-mouse-cerebellum", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7117326324003791, + "patch_iou": 0.6289278430482521, + "patch_precision": 0.7384001631619221, + "patch_recall": 0.8160601597829159, + "patch_vi": 0.34099656215267415, + "n_patches": 1300, + "cal_ece": 0.014679185918989899, + "cal_nll": 0.17894412255736725, + "cal_brier": 0.051316040682139796, + "cal_mean_confidence": 0.9461751954929392, + "cal_accuracy": 0.9315347055288462, + "cal_predicted_prevalence": 0.22203125, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.22613713504306693 + }, + "test_notta": { + "patch_dice": 0.7117326324003791, + "patch_iou": 0.6289278430482521, + "patch_precision": 0.7384001631619221, + "patch_recall": 0.8160601597829159, + "patch_vi": 0.34099656215267415, + "n_patches": 1300, + "cal_ece": 0.014679185918989899, + "cal_nll": 0.17894412255736725, + "cal_brier": 0.051316040682139796, + "cal_mean_confidence": 0.9461751954929392, + "cal_accuracy": 0.9315347055288462, + "cal_predicted_prevalence": 0.22203125, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.22613713504306693 + }, + "test": { + "patch_dice": 0.7328983606873765, + "patch_iou": 0.6539298086166608, + "patch_precision": 0.7794708341773859, + "patch_recall": 0.8101709827186498, + "patch_vi": 0.31509139568672867, + "n_patches": 1300, + "image_dice": 0.8398286469075072, + "image_iou": 0.737109970022708, + "image_precision": 0.8420662330155432, + "image_recall": 0.8598087525051639, + "image_vi": 0.3650706363380156, + "n_images": 52, + "cal_ece": 0.003320093970549238, + "cal_nll": 0.15897419372119356, + "cal_brier": 0.04664112409434639, + "cal_mean_confidence": 0.9397174220413317, + "cal_accuracy": 0.9365895432692307, + "cal_predicted_prevalence": 0.21556715745192306, + "cal_true_prevalence": 0.21078049879807692, + "cal_mean_probability": 0.2233936774178567 + }, + "eval_seconds": 20.4, + "total_seconds": 1180.9, + "resources": { + "phases": { + "setup": { + "seconds": 1.15, + "peak_cpu_rss_mb": 1027.9, + "peak_gpu_alloc_mb": 1066.3, + "peak_gpu_reserved_mb": 1086.0 + }, + "features_and_weights": { + "seconds": 31.61, + "peak_cpu_rss_mb": 1801.2, + "peak_gpu_alloc_mb": 2820.1, + "peak_gpu_reserved_mb": 3212.0 + }, + "distillation": { + "seconds": 327.55, + "peak_cpu_rss_mb": 3676.1, + "peak_gpu_alloc_mb": 4303.7, + "peak_gpu_reserved_mb": 4638.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 798.09, + "peak_cpu_rss_mb": 3676.1, + "peak_gpu_alloc_mb": 4424.0, + "peak_gpu_reserved_mb": 4648.0, + "rounds": 6 + }, + "inference": { + "seconds": 20.38, + "peak_cpu_rss_mb": 3676.1, + "peak_gpu_alloc_mb": 2623.4, + "peak_gpu_reserved_mb": 4648.0, + "patches": 1300, + "images": 52, + "patches_per_second": 63.8, + "seconds_per_image": 0.392 + } + }, + "peak_gpu_alloc_mb": 4424.0, + "peak_cpu_rss_mb": 3676.1, + "total_seconds": 1178.8 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/whole-mouse-brain/macs-v11-norestart-acq-mcdropout_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..c0fcb47c5772c630806ff1052a5395c42f1d7706 --- /dev/null +++ b/macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,49,34.0,169.4,"[0.4166773458661978, 0.35928987385280325, 0.153101914721845, 0.3102874105923009, 0.24827571594727177, 0.40443287728637833, 0.45309617093263127, 0.3151852254719067]",,745730,43,True,kept,0.30178618768780596,0.45309617093263127,"[0.0523462898796494, 0.0, 0.0, 0.0, 0.0, 0.0, 0.9476537101203506, 0.0]",top2,"{'mixture': 0.41707851191221385, 'uniform': 0.3962786850936092, 'top1': 0.45309617093263127, 'top2': 0.4560007178672816, 'top3': 0.4242840493249194}",0.649954056799635,0.46975670281025395,0.015682493902244875,,,,, +2,98,10.9,31.3,"[0.36611084588621456, 0.41390494215223367, 0.09410596399488945, 0.24220693615965957, 0.15365108848199444, 0.24885494485948062, 0.29824941598594545, 0.355243465280686]",,1552702,76,False,,,,"[0.012823011947318825, 0.6947891955189256, 0.004286685040611986, 0.036731834491645354, 0.07570608809152632, 0.07008633566128063, 0.08182206111278438, 0.023754788135906927]",mixture,"{'mixture': 0.41136455031192193, 'uniform': 0.2299235728798188, 'top1': 0.41390494215223367, 'top2': 0.4141551875692704, 'top3': 0.4140180268546756}",0.6391891344279756,0.5097387246188724,0.04795928938682097,0.5334437898129569,0.4503609960060156,student,0.111, +3,147,10.8,119.4,"[0.36768339021727037, 0.3852192186783384, 0.10416450316146425, 0.24673547320391392, 0.2164043919966956, 0.3083610023527579, 0.37673879727901105, 0.34984706011994526]","[0.4249715484975724, 0.43163808487226474, 0.13402683120843648, 0.27919123067816787, 0.28581213424874735, 0.4018351343096836, 0.5029478154698421, 0.3738482728359761]",2045013,111,False,,,,"[0.012424062776149956, 0.49741245892995734, 0.004521154756190894, 0.03665674701240556, 0.13524121291846142, 0.12120177924292354, 0.17107619568913765, 0.02146638867477356]",mixture,"{'mixture': 0.40562761102178846, 'uniform': 0.3019242904620139, 'top1': 0.3852192186783384, 'top2': 0.40122242697931965, 'top3': 0.40084130714444877}",0.6526551922105499,0.48330637836874146,0.01201782947328647,0.3206224561686304,0.31636739169836314,student,0.0, +4,196,10.7,123.8,"[0.3669733595372897, 0.3639770152858429, 0.10806844150474738, 0.25307971683616926, 0.22584242770873106, 0.320093417456483, 0.38747785383420064, 0.3311470636749818]","[0.42873382874298777, 0.4201025611225556, 0.1465960471130387, 0.27571882770491135, 0.3152241379813523, 0.4274018046756167, 0.5223327910218584, 0.3509337191657171]",2546814,151,False,,,,"[0.020389457321379638, 0.6647963031539222, 0.0, 0.0, 0.0, 0.0, 0.31481423952469817, 0.0]",top3,"{'mixture': 0.3754659094523485, 'uniform': 0.31416604860326686, 'top1': 0.38747785383420064, 'top2': 0.3897138845510255, 'top3': 0.39374676780961954}",0.71405290677071,0.3943821589053846,0.0050473695531796405,0.3715079220082423,0.24320091987757675,student,0.0, +5,245,10.6,138.8,"[0.3502083366531371, 0.34845351115348644, 0.11604729844495462, 0.2621757924425775, 0.22619772716440484, 0.33203357450989535, 0.3840164960993779, 0.313400181441156]","[0.37548772339205577, 0.3833013808355072, 0.1600921317997613, 0.25321508996470743, 0.31990754606797517, 0.4370099709404248, 0.5179453131615449, 0.280774759856714]",3577893,189,False,,,,"[0.01936135902517909, 0.6391615107247445, 0.0, 0.0, 0.0, 0.0, 0.3414771302500763, 0.0]",top3,"{'mixture': 0.3599807604482505, 'uniform': 0.3211029686375517, 'top1': 0.3840164960993779, 'top2': 0.38579360695658055, 'top3': 0.3835683473682102}",0.7095045778301132,0.3975579165871835,0.0022580185247888613,0.5884572297615537,0.2919429375125578,student,0.295, +6,294,10.6,145.1,"[0.3652716200083902, 0.3585971398683318, 0.10992564142649246, 0.2578325740522286, 0.22277450818970798, 0.33973973894072296, 0.3846495074209926, 0.3305959501728769]","[0.37545897608442663, 0.36517019215223734, 0.15792099420152225, 0.2615905184780217, 0.330487145457808, 0.46403301342316666, 0.532613356475908, 0.28124866584265834]",4756463,232,False,,,,"[0.020966771040059683, 0.6589319110869489, 0.0, 0.0, 0.0, 0.0, 0.3201013178729914, 0.0]",top3,"{'mixture': 0.3652920718151623, 'uniform': 0.3223976533149485, 'top1': 0.3846495074209926, 'top2': 0.38678028399483, 'top3': 0.39248325722874766}",0.7143403163792835,0.40830035269899356,0.007214722442033303,0.5404097731059258,0.5780078399479068,student,0.135, +7,294,0.0,56.6,,,,,,,,,,,,0.7309256869658374,0.3713934767695636,0.015823260301908625,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..7c44232b64d567770d740809218ee1ff0fadb09c --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt @@ -0,0 +1,107 @@ +[11:20:15][ 0s][START] target=drosophila-vnc variant=macs-v11-norestart-acq-laplace budget=5.0% seed=3 | {"task": "mitochondria", "target": "drosophila-vnc", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "laplace", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[11:20:15][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[11:20:15][ 0s][DATA] drosophila-vnc: train=686 val=147 test=147 patches +[11:20:16][ 1s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, multiclass=2.60 +[11:20:16][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[11:20:17][ 2s][WEIGHTS] mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020 | eff.sources=1.77 | MMD2 mix=0.1975 uniform=0.3743 +[11:20:17][ 2s][WEIGHTS] per-source: fly-mito=0.039, kasthuri=0.943, lucchi=0.014, multiclass=0.004 | min-MMD source (NeuroADDA rule) = kasthuri +[11:20:21][ 5s][TEACHERS] target-val Dice: fly-mito=0.278, kasthuri=0.183, lucchi=0.058, multiclass=0.017 +[11:20:21][ 5s][TEACHERS] ensemble target-val Dice: mixture=0.187, mixture@raw=0.185, mixture@logit=0.193, per_source=0.184, per_source@raw=0.184, per_source@logit=0.189, uniform=0.070, uniform@raw=0.088, uniform@logit=0.065 +[11:20:21][ 5s][WEIGHTS] using mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020 +[11:20:21][ 5s][INIT] student initialised from teacher kasthuri (weight 0.838) +[11:20:22][ 6s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[11:20:22][ 6s][KD] distilling the mixture ensemble (prob space) into the student +[11:20:26][ 10s][KD] soft labels for 686 patches from 4 teachers in 4s +[11:20:32][ 17s][KD] epoch 1/5 | patches=686 | loss=0.1985 | 6s +[11:20:38][ 23s][KD] epoch 2/5 | patches=686 | loss=0.1315 | 6s +[11:20:44][ 29s][KD] epoch 3/5 | patches=686 | loss=0.1251 | 6s +[11:20:51][ 36s][KD] epoch 4/5 | patches=686 | loss=0.1224 | 6s +[11:20:57][ 42s][KD] epoch 5/5 | patches=686 | loss=0.1199 | 6s +[11:20:57][ 42s][ROUND0] before any target label: val_dice=0.2226 val_vi=0.2217 +[11:20:57][ 42s][AL] patches=686 (pool 686 after dropping blank; free blank negatives: True) | budget=34 (5.0%) | rounds=6 | per_round=5 | acquisition=laplace | first_round=stratified +[11:21:00][ 44s][AL] round-1 uncertainty quality on 686 patches: {'spearman_uncertainty_error': 0.10850904905588832, 'auroc_top20_error': 0.5691314926478366} +[11:21:04][ 48s][AL] round 1: stratified cold start: 3 patches from the membrane-rich half (ensemble prevalence >= 0.047) + 2 from the membrane-poor half of 686 candidates +[11:21:04][ 49s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.573, uniform=0.044, top1=0.572, top2=0.573, top3=0.573 -> using mixture +[11:21:04][ 49s][WEIGHTS] round 1: student kept (labeled Dice 0.659 vs best teacher kasthuri 0.572); distilling the corrected ensemble into it +[11:21:04][ 49s][WEIGHTS] round 1: teacher Dice on 5 labeled patches + 0 blank (17096 fg px): fly-mito=0.327, kasthuri=0.572, lucchi=0.000, multiclass=0.001 | weights: fly-mito=0.012, kasthuri=0.988, lucchi=0.000, multiclass=0.000 +[11:21:08][ 53s][KD] soft labels for 681 patches from 4 teachers in 4s +[11:21:15][ 59s][KD] epoch 1/2 | patches=681 | loss=0.0830 | 6s +[11:21:21][ 66s][KD] epoch 2/2 | patches=681 | loss=0.0810 | 6s +[11:21:25][ 70s][CONS] pseudo-labels for 681 unlabeled patches from the current ensemble in 4s +[11:21:25][ 70s][FT] round 1/6 epoch 1/5 | patches=5 | loss=0.8492 +[11:21:26][ 71s][FT] round 1/6 epoch 2/5 | patches=5 | loss=0.8436 +[11:21:26][ 71s][FT] round 1/6 epoch 3/5 | patches=5 | loss=0.8554 +[11:21:26][ 71s][FT] round 1/6 epoch 4/5 | patches=5 | loss=0.7927 +[11:21:27][ 72s][FT] round 1/6 epoch 5/5 | patches=5 | loss=0.8922 +[11:21:27][ 72s][AL] round 1/6 | labeled=5 | selection_seconds=6.4000 round_seconds=29.6000 restart_student_dice=0.6587 restart_teacher_dice=0.5724 val_dice=0.2651 val_vi=0.2537 val_ece=0.0339 +[11:21:30][ 74s][AL] round 2: 126/681 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:21:30][ 75s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.291, uniform=0.022, top1=0.344, top2=0.291, top3=0.291 -> using top1 +[11:21:30][ 75s][WEIGHTS] round 2: teacher Dice on 10 labeled patches + 0 blank (49513 fg px): fly-mito=0.344, kasthuri=0.288, lucchi=0.002, multiclass=0.003 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[11:21:31][ 76s][CONS] round 2: on the 5 new patches student Dice=0.049 vs ensemble 0.355 -> pseudo-labels from the ensemble | consistency x1.00 +[11:21:32][ 77s][FT] round 2/6 epoch 1/5 | patches=10 | loss=1.2737 +[11:21:32][ 77s][FT] round 2/6 epoch 2/5 | patches=10 | loss=1.1118 +[11:21:33][ 78s][FT] round 2/6 epoch 3/5 | patches=10 | loss=0.9654 +[11:21:34][ 78s][FT] round 2/6 epoch 4/5 | patches=10 | loss=0.9057 +[11:21:34][ 79s][FT] round 2/6 epoch 5/5 | patches=10 | loss=0.8844 +[11:21:35][ 79s][AL] round 2/6 | labeled=10 | selection_seconds=2.3000 round_seconds=6.8000 pseudo_student_dice=0.0485 pseudo_ensemble_dice=0.3552 consistency_scale=1.0000 val_dice=0.5304 val_vi=0.1854 val_ece=0.0069 +[11:21:37][ 82s][AL] round 3: 113/676 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:21:38][ 83s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.228, uniform=0.020, top1=0.313, top2=0.229, top3=0.229 -> using top1 +[11:21:38][ 83s][WEIGHTS] round 3: teacher Dice on 15 labeled patches + 0 blank (67007 fg px): fly-mito=0.313, kasthuri=0.224, lucchi=0.002, multiclass=0.005 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[11:21:40][ 85s][CONS] round 3: on the 5 new patches student Dice=0.371 vs ensemble 0.254 -> pseudo-labels from the student | consistency x0.00 +[11:21:40][ 85s][FT] round 3/6 epoch 1/5 | patches=15 | loss=0.7781 +[11:21:41][ 85s][FT] round 3/6 epoch 2/5 | patches=15 | loss=0.7260 +[11:21:41][ 86s][FT] round 3/6 epoch 3/5 | patches=15 | loss=0.6272 +[11:21:41][ 86s][FT] round 3/6 epoch 4/5 | patches=15 | loss=0.5273 +[11:21:42][ 87s][FT] round 3/6 epoch 5/5 | patches=15 | loss=0.4273 +[11:21:42][ 87s][AL] round 3/6 | labeled=15 | selection_seconds=2.4000 round_seconds=7.2000 pseudo_student_dice=0.3711 pseudo_ensemble_dice=0.2543 consistency_scale=0.0000 val_dice=0.5918 val_vi=0.1755 val_ece=0.0097 +[11:21:45][ 89s][AL] round 4: 47/671 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:21:54][ 99s][WEIGHTS] round 4: teacher agreement with the student on 666 unlabeled patches: fly-mito=0.343, kasthuri=0.444, lucchi=0.010, multiclass=0.002 | blended with labeled Dice at 0.50 +[11:22:01][ 106s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.341, uniform=0.035, top1=0.341, top2=0.341, top3=0.341 -> using mixture +[11:22:01][ 106s][WEIGHTS] round 4: teacher Dice on 20 labeled patches + 0 blank (74131 fg px): fly-mito=0.315, kasthuri=0.341, lucchi=0.006, multiclass=0.003 | weights: fly-mito=0.097, kasthuri=0.901, lucchi=0.001, multiclass=0.001 +[11:22:03][ 108s][CONS] round 4: on the 5 new patches student Dice=0.436 vs ensemble 0.357 -> pseudo-labels from the student | consistency x0.00 +[11:22:04][ 109s][FT] round 4/6 epoch 1/5 | patches=20 | loss=0.4633 +[11:22:04][ 109s][FT] round 4/6 epoch 2/5 | patches=20 | loss=0.4212 +[11:22:05][ 110s][FT] round 4/6 epoch 3/5 | patches=20 | loss=0.3607 +[11:22:05][ 110s][FT] round 4/6 epoch 4/5 | patches=20 | loss=0.3613 +[11:22:06][ 111s][FT] round 4/6 epoch 5/5 | patches=20 | loss=0.3089 +[11:22:06][ 111s][AL] round 4/6 | labeled=20 | selection_seconds=2.3000 round_seconds=23.4000 pseudo_student_dice=0.4365 pseudo_ensemble_dice=0.3567 consistency_scale=0.0000 val_dice=0.7202 val_vi=0.1155 val_ece=0.0083 +[11:22:08][ 113s][AL] round 5: 104/666 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:22:17][ 122s][WEIGHTS] round 5: teacher agreement with the student on 661 unlabeled patches: fly-mito=0.382, kasthuri=0.466, lucchi=0.009, multiclass=0.001 | blended with labeled Dice at 0.50 +[11:22:25][ 130s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.374, uniform=0.044, top1=0.373, top2=0.374, top3=0.374 -> using mixture +[11:22:25][ 130s][WEIGHTS] round 5: teacher Dice on 25 labeled patches + 0 blank (82397 fg px): fly-mito=0.327, kasthuri=0.373, lucchi=0.012, multiclass=0.003 | weights: fly-mito=0.081, kasthuri=0.918, lucchi=0.001, multiclass=0.001 +[11:22:27][ 132s][CONS] round 5: on the 5 new patches student Dice=0.860 vs ensemble 0.498 -> pseudo-labels from the student | consistency x1.00 +[11:22:28][ 133s][FT] round 5/6 epoch 1/5 | patches=25 | loss=0.3228 +[11:22:29][ 134s][FT] round 5/6 epoch 2/5 | patches=25 | loss=0.4003 +[11:22:29][ 134s][FT] round 5/6 epoch 3/5 | patches=25 | loss=0.3622 +[11:22:30][ 135s][FT] round 5/6 epoch 4/5 | patches=25 | loss=0.2707 +[11:22:31][ 136s][FT] round 5/6 epoch 5/5 | patches=25 | loss=0.2498 +[11:22:32][ 136s][AL] round 5/6 | labeled=25 | selection_seconds=2.3000 round_seconds=25.1000 pseudo_student_dice=0.8602 pseudo_ensemble_dice=0.4984 consistency_scale=1.0000 val_dice=0.7097 val_vi=0.1201 val_ece=0.0088 +[11:22:34][ 139s][AL] round 6: 111/661 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:22:43][ 148s][WEIGHTS] round 6: teacher agreement with the student on 656 unlabeled patches: fly-mito=0.391, kasthuri=0.449, lucchi=0.009, multiclass=0.001 | blended with labeled Dice at 0.50 +[11:22:51][ 156s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.394, uniform=0.048, top1=0.394, top2=0.394, top3=0.394 -> using mixture +[11:22:51][ 156s][WEIGHTS] round 6: teacher Dice on 30 labeled patches + 0 blank (91279 fg px): fly-mito=0.327, kasthuri=0.394, lucchi=0.011, multiclass=0.002 | weights: fly-mito=0.067, kasthuri=0.932, lucchi=0.001, multiclass=0.000 +[11:22:53][ 158s][CONS] round 6: on the 5 new patches student Dice=0.907 vs ensemble 0.639 -> pseudo-labels from the student | consistency x1.00 +[11:22:54][ 158s][FT] round 6/6 epoch 1/5 | patches=30 | loss=0.2780 +[11:22:54][ 159s][FT] round 6/6 epoch 2/5 | patches=30 | loss=0.2534 +[11:22:55][ 160s][FT] round 6/6 epoch 3/5 | patches=30 | loss=0.2623 +[11:22:56][ 161s][FT] round 6/6 epoch 4/5 | patches=30 | loss=0.2548 +[11:22:57][ 162s][FT] round 6/6 epoch 5/5 | patches=30 | loss=0.2277 +[11:22:58][ 162s][AL] round 6/6 | labeled=30 | selection_seconds=2.4000 round_seconds=25.5000 pseudo_student_dice=0.9074 pseudo_ensemble_dice=0.6390 consistency_scale=1.0000 val_dice=0.7320 val_vi=0.1066 val_ece=0.0094 +[11:22:59][ 163s][CONS] final consolidation: pseudo-labels refreshed from the student +[11:22:59][ 164s][FT] final 1/10 epoch 1/1 | patches=30 | loss=0.2279 +[11:23:00][ 165s][FT] final 2/10 epoch 1/1 | patches=30 | loss=0.1906 +[11:23:01][ 166s][FT] final 3/10 epoch 1/1 | patches=30 | loss=0.1898 +[11:23:02][ 167s][FT] final 4/10 epoch 1/1 | patches=30 | loss=0.1942 +[11:23:03][ 168s][FT] final 5/10 epoch 1/1 | patches=30 | loss=0.1911 +[11:23:04][ 169s][FT] final 6/10 epoch 1/1 | patches=30 | loss=0.1783 +[11:23:05][ 169s][FT] final 7/10 epoch 1/1 | patches=30 | loss=0.1626 +[11:23:05][ 170s][FT] final 8/10 epoch 1/1 | patches=30 | loss=0.2015 +[11:23:06][ 171s][FT] final 9/10 epoch 1/1 | patches=30 | loss=0.1667 +[11:23:07][ 172s][FT] final 10/10 epoch 1/1 | patches=30 | loss=0.1817 +[11:23:08][ 172s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=9.6000 val_dice=0.7444 val_vi=0.1016 val_ece=0.0095 +[11:23:08][ 173s][WEIGHTS] final: fly-mito=0.067, kasthuri=0.932, lucchi=0.001, multiclass=0.000 +[11:23:10][ 175s][TEST] raw student (no EMA, no TTA): patch_dice=0.7977 patch_vi=0.1139 +[11:23:11][ 175s][TEST] final model without TTA: patch_dice=0.7977 patch_vi=0.1139 +[11:23:11][ 175s][RESOURCES] {"phases": {"setup": {"seconds": 0.64, "peak_cpu_rss_mb": 1025.3, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 5.64, "peak_cpu_rss_mb": 1806.2, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 35.29, "peak_cpu_rss_mb": 2104.5, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4168.0, "kd_epochs": 5}, "active_learning": {"seconds": 130.37, "peak_cpu_rss_mb": 2410.7, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 2.88, "peak_cpu_rss_mb": 2410.7, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 147, "images": 3, "patches_per_second": 51.0, "seconds_per_image": 0.96}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 2410.7, "total_seconds": 174.8} +[11:23:11][ 175s][TEST] patch_dice=0.8348 patch_iou=0.7589 patch_precision=0.8195 patch_recall=0.9316 patch_vi=0.0987 image_dice=0.9122 image_iou=0.8396 image_precision=0.8615 image_recall=0.9710 image_vi=0.1005 cal_ece=0.0037 cal_nll=0.0333 cal_brier=0.0097 cal_mean_confidence=0.9907 cal_accuracy=0.9874 cal_predicted_prevalence=0.0674 cal_true_prevalence=0.0594 cal_mean_probability=0.0674 +[11:23:11][ 176s][DONE] drosophila-vnc macs-v11-norestart-acq-laplace b5 s3 | test patch Dice=0.8348 VI=0.0987 | image Dice=0.9122 VI=0.1005 | 175s diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..aae99b1220c1a36b437eae02c84a330acfd7127b --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json @@ -0,0 +1,662 @@ +{ + "target": "drosophila-vnc", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 3, + "config": { + "task": "mitochondria", + "target": "drosophila-vnc", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "laplace", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "drosophila-vnc", + "--variants", + "macs-v11-norestart-acq-laplace", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "multiclass" + ], + "alphas": [ + 0.11690531690536884, + 0.8375898290916663, + 0.02560239431814963, + 0.01990245968481518 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "multiclass" + ], + "mixture": [ + 0.11690531690536884, + 0.8375898290916663, + 0.02560239431814963, + 0.01990245968481518 + ], + "per_source": [ + 0.038599381604120485, + 0.943285983973492, + 0.013712262708195049, + 0.004402371714192633 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.1917814945701421, + "mmd2_mixture": 0.19749308036725388, + "mmd2_uniform": 0.3743480458439246, + "mmd2_single_source": [ + 0.7042945447726864, + 0.2070680114579429, + 0.6307945959829456, + 0.6534353423959632 + ], + "median_squared_distance": 0.1964580928585633, + "gammas": [ + 1.2725360221224549, + 2.5450720442449097, + 5.0901440884898195, + 10.180288176979639, + 20.360576353959278 + ], + "max_weight": 0.8375898290916663, + "entropy": 0.5711585797111782, + "effective_sources": 1.7703169170005038, + "min_weight": 0.01990245968481518 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6050756121533214, + 0.3110169227475098, + 0.7002952924749369, + 0.8048259551971948 + ], + "standardized": [ + -0.0012381647254944845, + -1.5993047903844642, + 0.5162347028009382, + 1.0843082523090217 + ], + "tau": 0.5, + "max_weight": 0.943285983973492, + "entropy": 0.2634007298678481, + "effective_sources": 1.3013481035311127, + "min_weight": 0.004402371714192633 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "kasthuri-pp", + "seconds": 1.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.2784035818310388, + 0.18341782772166676, + 0.057702367443012846, + 0.01681466600081078 + ], + "ensemble_val_dice": { + "mixture": 0.186522633793688, + "mixture@raw": 0.18548325188563394, + "mixture@logit": 0.19296315829990748, + "per_source": 0.18415887092754818, + "per_source@raw": 0.18350562678267812, + "per_source@logit": 0.1890579186582894, + "uniform": 0.06967222752121038, + "uniform@raw": 0.0884686387663988, + "uniform@logit": 0.06545324142703345 + }, + "val_patches": 147 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 3.7969236373901367, + 2.597255229949951 + ], + "top_source": "kasthuri-pp", + "weight_stats": { + "max_weight": 0.8375898290916663, + "entropy": 0.5711585797111782, + "effective_sources": 1.7703169170005038, + "min_weight": 0.01990245968481518 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.21081851067789195, + "top_source": "kasthuri-pp", + "top_source_is_best_teacher": false + }, + "init_teacher": "kasthuri-pp", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 35.3, + "round0_val": { + "val_dice": 0.22259710119393636, + "val_vi": 0.2217486787180137, + "val_ece": 0.019118359085155114 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 5, + "selection_seconds": 6.4, + "round_seconds": 29.6, + "teacher_dice_labeled": [ + 0.3265584832904884, + 0.5723588778124132, + 5.7760064691272456e-05, + 0.0008705240554813998 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 17096, + "evidence_nonempty_patches": 4, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.6586710683430498, + "restart_teacher_dice": 0.5723588778124132, + "weights": [ + 0.011805122497730317, + 0.9880187339743336, + 9.875115690298976e-05, + 7.739237103305626e-05 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.5725535002933638, + "uniform": 0.04434476493300023, + "top1": 0.5723588778124132, + "top2": 0.5725535002933638, + "top3": 0.5725535002933638 + }, + "val_dice": 0.2651183621363361, + "val_vi": 0.2536928580768912, + "val_ece": 0.03392489668719592 + }, + { + "round": 2, + "n_labeled": 10, + "selection_seconds": 2.3, + "round_seconds": 6.8, + "teacher_dice_labeled": [ + 0.34446258487008174, + 0.2877706560379188, + 0.0021410276932927806, + 0.002911120480234496 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 49513, + "evidence_nonempty_patches": 9, + "first_correction": false, + "restart": "", + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.2912965056802886, + "uniform": 0.02191753601589667, + "top1": 0.34446258487008174, + "top2": 0.2913448340824071, + "top3": 0.2913145287183076 + }, + "pseudo_student_dice": 0.048524089711711266, + "pseudo_ensemble_dice": 0.35520279127149357, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.5303885061058371, + "val_vi": 0.1853514846468518, + "val_ece": 0.006875929179194766 + }, + { + "round": 3, + "n_labeled": 15, + "selection_seconds": 2.4, + "round_seconds": 7.2, + "teacher_dice_labeled": [ + 0.3131746656176239, + 0.22434563791634468, + 0.002236142580005035, + 0.005071563756801514 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 67007, + "evidence_nonempty_patches": 14, + "first_correction": false, + "restart": "", + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.22849699359095008, + "uniform": 0.01984394708280778, + "top1": 0.3131746656176239, + "top2": 0.22876961744740645, + "top3": 0.2286186652938789 + }, + "pseudo_student_dice": 0.3711265455655625, + "pseudo_ensemble_dice": 0.25430492104338365, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.5918325005506857, + "val_vi": 0.17548227962258417, + "val_ece": 0.009676093882371277 + }, + { + "round": 4, + "n_labeled": 20, + "selection_seconds": 2.3, + "round_seconds": 23.4, + "teacher_dice_labeled": [ + 0.31547631096594764, + 0.34123593983097356, + 0.0059793031811133535, + 0.0034405810135199576 + ], + "teacher_dice_pool": [ + 0.3425381911105748, + 0.4438936034888215, + 0.009888876495357073, + 0.002299520650677445 + ], + "evidence_fg_pixels": 74131, + "evidence_nonempty_patches": 18, + "first_correction": false, + "restart": "", + "weights": [ + 0.09720801201042216, + 0.9010976140529733, + 0.0009638733077765422, + 0.0007305006288280354 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.3409487021488296, + "uniform": 0.03477595874011665, + "top1": 0.34123593983097356, + "top2": 0.34102760479558736, + "top3": 0.34098090067463205 + }, + "pseudo_student_dice": 0.43649569353387324, + "pseudo_ensemble_dice": 0.3566652909616178, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.7202478275643855, + "val_vi": 0.11551100599924623, + "val_ece": 0.00827971477000551 + }, + { + "round": 5, + "n_labeled": 25, + "selection_seconds": 2.3, + "round_seconds": 25.1, + "teacher_dice_labeled": [ + 0.32748527014276285, + 0.3732444408424384, + 0.01181775174042917, + 0.002690437438876882 + ], + "teacher_dice_pool": [ + 0.3817607993473308, + 0.4657078120427067, + 0.009169410998209292, + 0.001259368545293597 + ], + "evidence_fg_pixels": 82397, + "evidence_nonempty_patches": 23, + "first_correction": false, + "restart": "", + "weights": [ + 0.08105022278551019, + 0.917658090242581, + 0.0007555686292442745, + 0.0005361183426645543 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.3736519199765924, + "uniform": 0.04405944383878502, + "top1": 0.3732444408424384, + "top2": 0.3737204245680453, + "top3": 0.373694044966715 + }, + "pseudo_student_dice": 0.8601795540110049, + "pseudo_ensemble_dice": 0.4983764769549923, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.7097438014736671, + "val_vi": 0.12009500484663031, + "val_ece": 0.00879034593597929 + }, + { + "round": 6, + "n_labeled": 30, + "selection_seconds": 2.4, + "round_seconds": 25.5, + "teacher_dice_labeled": [ + 0.32742006938767654, + 0.39371978128758006, + 0.010862133549017123, + 0.002418636813878437 + ], + "teacher_dice_pool": [ + 0.39091620078222555, + 0.4491281687389267, + 0.008656109805333422, + 0.0011212111372366253 + ], + "evidence_fg_pixels": 91279, + "evidence_nonempty_patches": 28, + "first_correction": false, + "restart": "", + "weights": [ + 0.06702579943774015, + 0.9319124729363013, + 0.0006192908650021591, + 0.00044243676095642704 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.39385881615174234, + "uniform": 0.048451003797465905, + "top1": 0.39371978128758006, + "top2": 0.39387514465174567, + "top3": 0.39385988022438434 + }, + "pseudo_student_dice": 0.9074005794047932, + "pseudo_ensemble_dice": 0.638978505072492, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.7320486463671512, + "val_vi": 0.10657063095909225, + "val_ece": 0.009392135846348293 + }, + { + "round": 7, + "n_labeled": 30, + "selection_seconds": 0.0, + "round_seconds": 9.6, + "final_consolidation": true, + "val_dice": 0.7444199834311062, + "val_vi": 0.10161531775765006, + "val_ece": 0.009509370836124294 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.10850904905588832, + "auroc_top20_error": 0.5691314926478366 + }, + "al_seconds": 130.4, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.06702579943774015, + 0.9319124729363013, + 0.0006192908650021591, + 0.00044243676095642704 + ], + "weight_stats_final": { + "max_weight": 0.9319124729363013, + "entropy": 0.2548559221187666, + "effective_sources": 1.2902757072900808, + "min_weight": 0.00044243676095642704 + }, + "top_source_final": "kasthuri-pp", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7977066471645559, + "patch_iou": 0.712263446875584, + "patch_precision": 0.762104894454457, + "patch_recall": 0.9284656145665409, + "patch_vi": 0.11394406286768838, + "n_patches": 147, + "cal_ece": 0.010102019099312812, + "cal_nll": 0.04980973616501659, + "cal_brier": 0.012402466143411325, + "cal_mean_confidence": 0.9944937530522531, + "cal_accuracy": 0.9843949298469388, + "cal_predicted_prevalence": 0.06852545705782313, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.06730707713274209 + }, + "test_notta": { + "patch_dice": 0.7977066471645559, + "patch_iou": 0.712263446875584, + "patch_precision": 0.762104894454457, + "patch_recall": 0.9284656145665409, + "patch_vi": 0.11394406286768838, + "n_patches": 147, + "cal_ece": 0.010102019099312812, + "cal_nll": 0.04980973616501659, + "cal_brier": 0.012402466143411325, + "cal_mean_confidence": 0.9944937530522531, + "cal_accuracy": 0.9843949298469388, + "cal_predicted_prevalence": 0.06852545705782313, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.06730707713274209 + }, + "test": { + "patch_dice": 0.83483573452591, + "patch_iou": 0.7589463120306933, + "patch_precision": 0.8194718578465112, + "patch_recall": 0.931564815747692, + "patch_vi": 0.09867003215150039, + "n_patches": 147, + "image_dice": 0.9122375170485886, + "image_iou": 0.8396129565317924, + "image_precision": 0.8614710995782565, + "image_recall": 0.9710319246167654, + "image_vi": 0.10048469250582326, + "n_images": 3, + "cal_ece": 0.003700974915991743, + "cal_nll": 0.03327139559211665, + "cal_brier": 0.009700747389481646, + "cal_mean_confidence": 0.9907116323509012, + "cal_accuracy": 0.9874043367346939, + "cal_predicted_prevalence": 0.06740274234693877, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.06744815259594922 + }, + "eval_seconds": 2.9, + "total_seconds": 175.4, + "resources": { + "phases": { + "setup": { + "seconds": 0.64, + "peak_cpu_rss_mb": 1025.3, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 5.64, + "peak_cpu_rss_mb": 1806.2, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 35.29, + "peak_cpu_rss_mb": 2104.5, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4168.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 130.37, + "peak_cpu_rss_mb": 2410.7, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4176.0, + "rounds": 6 + }, + "inference": { + "seconds": 2.88, + "peak_cpu_rss_mb": 2410.7, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4176.0, + "patches": 147, + "images": 3, + "patches_per_second": 51.0, + "seconds_per_image": 0.96 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 2410.7, + "total_seconds": 174.8 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/drosophila-vnc/macs-v11-norestart-acq-laplace_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..137852b6f7df813becf4650c261ef232385b4290 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,5,6.4,29.6,"[0.3265584832904884, 0.5723588778124132, 5.7760064691272456e-05, 0.0008705240554813998]",,17096,4,True,kept,0.6586710683430498,0.5723588778124132,"[0.011805122497730317, 0.9880187339743336, 9.875115690298976e-05, 7.739237103305626e-05]",mixture,"{'mixture': 0.5725535002933638, 'uniform': 0.04434476493300023, 'top1': 0.5723588778124132, 'top2': 0.5725535002933638, 'top3': 0.5725535002933638}",0.2651183621363361,0.2536928580768912,0.03392489668719592,,,,, +2,10,2.3,6.8,"[0.34446258487008174, 0.2877706560379188, 0.0021410276932927806, 0.002911120480234496]",,49513,9,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.2912965056802886, 'uniform': 0.02191753601589667, 'top1': 0.34446258487008174, 'top2': 0.2913448340824071, 'top3': 0.2913145287183076}",0.5303885061058371,0.1853514846468518,0.006875929179194766,0.048524089711711266,0.35520279127149357,ensemble,1.0, +3,15,2.4,7.2,"[0.3131746656176239, 0.22434563791634468, 0.002236142580005035, 0.005071563756801514]",,67007,14,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.22849699359095008, 'uniform': 0.01984394708280778, 'top1': 0.3131746656176239, 'top2': 0.22876961744740645, 'top3': 0.2286186652938789}",0.5918325005506857,0.17548227962258417,0.009676093882371277,0.3711265455655625,0.25430492104338365,student,0.0, +4,20,2.3,23.4,"[0.31547631096594764, 0.34123593983097356, 0.0059793031811133535, 0.0034405810135199576]","[0.3425381911105748, 0.4438936034888215, 0.009888876495357073, 0.002299520650677445]",74131,18,False,,,,"[0.09720801201042216, 0.9010976140529733, 0.0009638733077765422, 0.0007305006288280354]",mixture,"{'mixture': 0.3409487021488296, 'uniform': 0.03477595874011665, 'top1': 0.34123593983097356, 'top2': 0.34102760479558736, 'top3': 0.34098090067463205}",0.7202478275643855,0.11551100599924623,0.00827971477000551,0.43649569353387324,0.3566652909616178,student,0.0, +5,25,2.3,25.1,"[0.32748527014276285, 0.3732444408424384, 0.01181775174042917, 0.002690437438876882]","[0.3817607993473308, 0.4657078120427067, 0.009169410998209292, 0.001259368545293597]",82397,23,False,,,,"[0.08105022278551019, 0.917658090242581, 0.0007555686292442745, 0.0005361183426645543]",mixture,"{'mixture': 0.3736519199765924, 'uniform': 0.04405944383878502, 'top1': 0.3732444408424384, 'top2': 0.3737204245680453, 'top3': 0.373694044966715}",0.7097438014736671,0.12009500484663031,0.00879034593597929,0.8601795540110049,0.4983764769549923,student,1.0, +6,30,2.4,25.5,"[0.32742006938767654, 0.39371978128758006, 0.010862133549017123, 0.002418636813878437]","[0.39091620078222555, 0.4491281687389267, 0.008656109805333422, 0.0011212111372366253]",91279,28,False,,,,"[0.06702579943774015, 0.9319124729363013, 0.0006192908650021591, 0.00044243676095642704]",mixture,"{'mixture': 0.39385881615174234, 'uniform': 0.048451003797465905, 'top1': 0.39371978128758006, 'top2': 0.39387514465174567, 'top3': 0.39385988022438434}",0.7320486463671512,0.10657063095909225,0.009392135846348293,0.9074005794047932,0.638978505072492,student,1.0, +7,30,0.0,9.6,,,,,,,,,,,,0.7444199834311062,0.10161531775765006,0.009509370836124294,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..9a65cbc167e3e52e12fe66444999e6e5ccddbdfc --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt @@ -0,0 +1,108 @@ +[04:57:02][ 0s][START] target=drosophila-vnc variant=macs-v11-norestart-acq-margin budget=5.0% seed=1 | {"task": "mitochondria", "target": "drosophila-vnc", "variant": "macs-v11-norestart-acq-margin", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "margin", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[04:57:02][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[04:57:02][ 0s][DATA] drosophila-vnc: train=686 val=147 test=147 patches +[04:57:03][ 1s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, multiclass=2.60 +[04:57:03][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[04:57:04][ 2s][WEIGHTS] mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020 | eff.sources=1.77 | MMD2 mix=0.1975 uniform=0.3743 +[04:57:04][ 2s][WEIGHTS] per-source: fly-mito=0.039, kasthuri=0.943, lucchi=0.014, multiclass=0.004 | min-MMD source (NeuroADDA rule) = kasthuri +[04:57:08][ 6s][TEACHERS] target-val Dice: fly-mito=0.278, kasthuri=0.183, lucchi=0.058, multiclass=0.017 +[04:57:08][ 6s][TEACHERS] ensemble target-val Dice: mixture=0.187, mixture@raw=0.185, mixture@logit=0.193, per_source=0.184, per_source@raw=0.184, per_source@logit=0.189, uniform=0.070, uniform@raw=0.088, uniform@logit=0.065 +[04:57:08][ 6s][WEIGHTS] using mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020 +[04:57:08][ 6s][INIT] student initialised from teacher kasthuri (weight 0.838) +[04:57:09][ 7s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[04:57:09][ 7s][KD] distilling the mixture ensemble (prob space) into the student +[04:57:13][ 11s][KD] soft labels for 686 patches from 4 teachers in 4s +[04:57:19][ 17s][KD] epoch 1/5 | patches=686 | loss=0.2005 | 7s +[04:57:26][ 24s][KD] epoch 2/5 | patches=686 | loss=0.1322 | 7s +[04:57:32][ 30s][KD] epoch 3/5 | patches=686 | loss=0.1258 | 6s +[04:57:39][ 37s][KD] epoch 4/5 | patches=686 | loss=0.1216 | 6s +[04:57:45][ 43s][KD] epoch 5/5 | patches=686 | loss=0.1204 | 6s +[04:57:46][ 44s][ROUND0] before any target label: val_dice=0.1917 val_vi=0.2212 +[04:57:46][ 44s][AL] patches=686 (pool 686 after dropping blank; free blank negatives: True) | budget=34 (5.0%) | rounds=6 | per_round=5 | acquisition=margin | first_round=stratified +[04:57:47][ 45s][AL] round-1 uncertainty quality on 686 patches: {'spearman_uncertainty_error': 0.08899991394046061, 'auroc_top20_error': 0.5872474346768222} +[04:57:51][ 49s][AL] round 1: stratified cold start: 3 patches from the membrane-rich half (ensemble prevalence >= 0.047) + 2 from the membrane-poor half of 686 candidates +[04:57:51][ 49s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.303, uniform=0.018, top1=0.312, top2=0.305, top3=0.304 -> using top1 +[04:57:52][ 50s][WEIGHTS] round 1: student kept (labeled Dice 0.122 vs best teacher fly-mito 0.312); distilling the corrected ensemble into it +[04:57:52][ 50s][WEIGHTS] round 1: teacher Dice on 5 labeled patches + 0 blank (15368 fg px): fly-mito=0.312, kasthuri=0.097, lucchi=0.000, multiclass=0.003 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[04:57:53][ 51s][KD] soft labels for 681 patches from 4 teachers in 1s +[04:57:59][ 57s][KD] epoch 1/2 | patches=681 | loss=0.2886 | 6s +[04:58:06][ 64s][KD] epoch 2/2 | patches=681 | loss=0.2383 | 6s +[04:58:07][ 65s][CONS] pseudo-labels for 681 unlabeled patches from the current ensemble in 1s +[04:58:07][ 65s][FT] round 1/6 epoch 1/5 | patches=5 | loss=1.2861 +[04:58:08][ 66s][FT] round 1/6 epoch 2/5 | patches=5 | loss=1.2453 +[04:58:08][ 66s][FT] round 1/6 epoch 3/5 | patches=5 | loss=1.1900 +[04:58:09][ 67s][FT] round 1/6 epoch 4/5 | patches=5 | loss=1.1678 +[04:58:09][ 67s][FT] round 1/6 epoch 5/5 | patches=5 | loss=1.1580 +[04:58:10][ 68s][AL] round 1/6 | labeled=5 | selection_seconds=5.2000 round_seconds=23.5000 restart_student_dice=0.1222 restart_teacher_dice=0.3119 val_dice=0.2709 val_vi=0.3049 val_ece=0.0198 +[04:58:11][ 69s][AL] round 2: 10/681 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:58:12][ 70s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.186, uniform=0.049, top1=0.197, top2=0.186, top3=0.186 -> using top1 +[04:58:12][ 70s][WEIGHTS] round 2: teacher Dice on 10 labeled patches + 0 blank (18558 fg px): fly-mito=0.197, kasthuri=0.181, lucchi=0.000, multiclass=0.002 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[04:58:14][ 72s][CONS] round 2: on the 5 new patches student Dice=0.108 vs ensemble 0.085 -> pseudo-labels from the student | consistency x0.00 +[04:58:14][ 72s][FT] round 2/6 epoch 1/5 | patches=10 | loss=1.0486 +[04:58:14][ 72s][FT] round 2/6 epoch 2/5 | patches=10 | loss=1.0370 +[04:58:15][ 73s][FT] round 2/6 epoch 3/5 | patches=10 | loss=1.0217 +[04:58:15][ 73s][FT] round 2/6 epoch 4/5 | patches=10 | loss=0.9820 +[04:58:16][ 74s][FT] round 2/6 epoch 5/5 | patches=10 | loss=0.9480 +[04:58:16][ 74s][AL] round 2/6 | labeled=10 | selection_seconds=1.3000 round_seconds=6.0000 pseudo_student_dice=0.1078 pseudo_ensemble_dice=0.0846 consistency_scale=0.0000 val_dice=0.3910 val_vi=0.2760 val_ece=0.0339 +[04:58:17][ 75s][AL] round 3: 72/676 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:58:26][ 84s][WEIGHTS] round 3: teacher agreement with the student on 671 unlabeled patches: fly-mito=0.542, kasthuri=0.514, lucchi=0.004, multiclass=0.000 | blended with labeled Dice at 0.50 +[04:58:34][ 92s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.340, uniform=0.047, top1=0.364, top2=0.340, top3=0.340 -> using top1 +[04:58:34][ 92s][WEIGHTS] round 3: teacher Dice on 15 labeled patches + 0 blank (24653 fg px): fly-mito=0.364, kasthuri=0.332, lucchi=0.002, multiclass=0.001 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[04:58:36][ 94s][CONS] round 3: on the 5 new patches student Dice=0.221 vs ensemble 0.166 -> pseudo-labels from the student | consistency x0.00 +[04:58:36][ 94s][FT] round 3/6 epoch 1/5 | patches=15 | loss=0.9594 +[04:58:37][ 95s][FT] round 3/6 epoch 2/5 | patches=15 | loss=0.9165 +[04:58:37][ 95s][FT] round 3/6 epoch 3/5 | patches=15 | loss=0.8829 +[04:58:38][ 96s][FT] round 3/6 epoch 4/5 | patches=15 | loss=0.8287 +[04:58:38][ 96s][FT] round 3/6 epoch 5/5 | patches=15 | loss=0.7922 +[04:58:38][ 96s][AL] round 3/6 | labeled=15 | selection_seconds=1.3000 round_seconds=22.0000 pseudo_student_dice=0.2206 pseudo_ensemble_dice=0.1655 consistency_scale=0.0000 val_dice=0.4794 val_vi=0.2113 val_ece=0.0096 +[04:58:40][ 98s][AL] round 4: 91/671 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:58:49][ 107s][WEIGHTS] round 4: teacher agreement with the student on 666 unlabeled patches: fly-mito=0.488, kasthuri=0.518, lucchi=0.005, multiclass=0.000 | blended with labeled Dice at 0.50 +[04:58:57][ 115s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.336, uniform=0.042, top1=0.339, top2=0.336, top3=0.336 -> using mixture +[04:58:57][ 115s][WEIGHTS] round 4: teacher Dice on 20 labeled patches + 0 blank (39174 fg px): fly-mito=0.339, kasthuri=0.332, lucchi=0.003, multiclass=0.001 | weights: fly-mito=0.130, kasthuri=0.868, lucchi=0.001, multiclass=0.001 +[04:58:59][ 117s][CONS] round 4: on the 5 new patches student Dice=0.266 vs ensemble 0.144 -> pseudo-labels from the student | consistency x0.00 +[04:58:59][ 117s][FT] round 4/6 epoch 1/5 | patches=20 | loss=0.7862 +[04:59:00][ 118s][FT] round 4/6 epoch 2/5 | patches=20 | loss=0.7593 +[04:59:00][ 118s][FT] round 4/6 epoch 3/5 | patches=20 | loss=0.6879 +[04:59:01][ 119s][FT] round 4/6 epoch 4/5 | patches=20 | loss=0.6546 +[04:59:01][ 119s][FT] round 4/6 epoch 5/5 | patches=20 | loss=0.6581 +[04:59:02][ 120s][AL] round 4/6 | labeled=20 | selection_seconds=1.3000 round_seconds=22.6000 pseudo_student_dice=0.2664 pseudo_ensemble_dice=0.1440 consistency_scale=0.0000 val_dice=0.5896 val_vi=0.1470 val_ece=0.0053 +[04:59:03][ 121s][AL] round 5: 110/666 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:59:12][ 130s][WEIGHTS] round 5: teacher agreement with the student on 661 unlabeled patches: fly-mito=0.377, kasthuri=0.468, lucchi=0.009, multiclass=0.000 | blended with labeled Dice at 0.50 +[04:59:19][ 137s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.298, uniform=0.040, top1=0.296, top2=0.298, top3=0.298 -> using mixture +[04:59:19][ 137s][WEIGHTS] round 5: teacher Dice on 25 labeled patches + 0 blank (52837 fg px): fly-mito=0.283, kasthuri=0.296, lucchi=0.005, multiclass=0.001 | weights: fly-mito=0.109, kasthuri=0.888, lucchi=0.001, multiclass=0.001 +[04:59:21][ 139s][CONS] round 5: on the 5 new patches student Dice=0.449 vs ensemble 0.043 -> pseudo-labels from the student | consistency x0.00 +[04:59:22][ 140s][FT] round 5/6 epoch 1/5 | patches=25 | loss=0.6253 +[04:59:22][ 140s][FT] round 5/6 epoch 2/5 | patches=25 | loss=0.5868 +[04:59:23][ 141s][FT] round 5/6 epoch 3/5 | patches=25 | loss=0.5319 +[04:59:23][ 141s][FT] round 5/6 epoch 4/5 | patches=25 | loss=0.5841 +[04:59:24][ 142s][FT] round 5/6 epoch 5/5 | patches=25 | loss=0.5223 +[04:59:24][ 142s][AL] round 5/6 | labeled=25 | selection_seconds=1.3000 round_seconds=22.2000 pseudo_student_dice=0.4489 pseudo_ensemble_dice=0.0427 consistency_scale=0.0000 val_dice=0.7094 val_vi=0.1120 val_ece=0.0088 +[04:59:25][ 143s][AL] round 6: 158/661 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:59:34][ 152s][WEIGHTS] round 6: teacher agreement with the student on 656 unlabeled patches: fly-mito=0.350, kasthuri=0.455, lucchi=0.010, multiclass=0.001 | blended with labeled Dice at 0.50 +[04:59:42][ 160s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.285, uniform=0.042, top1=0.283, top2=0.285, top3=0.285 -> using mixture +[04:59:42][ 160s][WEIGHTS] round 6: teacher Dice on 30 labeled patches + 0 blank (67074 fg px): fly-mito=0.272, kasthuri=0.283, lucchi=0.006, multiclass=0.001 | weights: fly-mito=0.111, kasthuri=0.886, lucchi=0.002, multiclass=0.001 +[04:59:44][ 162s][CONS] round 6: on the 5 new patches student Dice=0.399 vs ensemble 0.059 -> pseudo-labels from the student | consistency x0.00 +[04:59:45][ 163s][FT] round 6/6 epoch 1/5 | patches=30 | loss=0.5660 +[04:59:45][ 163s][FT] round 6/6 epoch 2/5 | patches=30 | loss=0.6058 +[04:59:46][ 164s][FT] round 6/6 epoch 3/5 | patches=30 | loss=0.5576 +[04:59:46][ 164s][FT] round 6/6 epoch 4/5 | patches=30 | loss=0.5193 +[04:59:47][ 165s][FT] round 6/6 epoch 5/5 | patches=30 | loss=0.5235 +[04:59:47][ 165s][AL] round 6/6 | labeled=30 | selection_seconds=1.2000 round_seconds=22.6000 pseudo_student_dice=0.3987 pseudo_ensemble_dice=0.0586 consistency_scale=0.0000 val_dice=0.6698 val_vi=0.1306 val_ece=0.0119 +[04:59:48][ 166s][CONS] final consolidation: pseudo-labels refreshed from the student +[04:59:49][ 167s][FT] final 1/10 epoch 1/1 | patches=30 | loss=0.5154 +[04:59:49][ 167s][FT] final 2/10 epoch 1/1 | patches=30 | loss=0.4676 +[04:59:50][ 168s][FT] final 3/10 epoch 1/1 | patches=30 | loss=0.4743 +[04:59:50][ 168s][FT] final 4/10 epoch 1/1 | patches=30 | loss=0.4537 +[04:59:51][ 169s][FT] final 5/10 epoch 1/1 | patches=30 | loss=0.4082 +[04:59:51][ 169s][FT] final 6/10 epoch 1/1 | patches=30 | loss=0.4052 +[04:59:52][ 170s][FT] final 7/10 epoch 1/1 | patches=30 | loss=0.4700 +[04:59:52][ 170s][FT] final 8/10 epoch 1/1 | patches=30 | loss=0.4553 +[04:59:53][ 171s][FT] final 9/10 epoch 1/1 | patches=30 | loss=0.3955 +[04:59:53][ 171s][FT] final 10/10 epoch 1/1 | patches=30 | loss=0.4500 +[04:59:54][ 172s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=6.2000 val_dice=0.7056 val_vi=0.1168 val_ece=0.0112 +[04:59:54][ 172s][WEIGHTS] final: fly-mito=0.111, kasthuri=0.886, lucchi=0.002, multiclass=0.001 +[04:59:56][ 174s][TEST] raw student (no EMA, no TTA): patch_dice=0.8127 patch_vi=0.1193 +[04:59:56][ 174s][TEST] final model without TTA: patch_dice=0.8127 patch_vi=0.1193 +[04:59:56][ 174s][RESOURCES] {"phases": {"setup": {"seconds": 0.61, "peak_cpu_rss_mb": 1026.1, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 5.97, "peak_cpu_rss_mb": 1806.7, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 36.39, "peak_cpu_rss_mb": 2105.8, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4168.0, "kd_epochs": 5}, "active_learning": {"seconds": 128.19, "peak_cpu_rss_mb": 2391.1, "peak_gpu_alloc_mb": 3948.5, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 2.58, "peak_cpu_rss_mb": 2391.1, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 147, "images": 3, "patches_per_second": 57.0, "seconds_per_image": 0.86}}, "peak_gpu_alloc_mb": 3948.5, "peak_cpu_rss_mb": 2391.1, "total_seconds": 173.7} +[04:59:56][ 174s][TEST] patch_dice=0.8160 patch_iou=0.7287 patch_precision=0.7998 patch_recall=0.9001 patch_vi=0.1139 image_dice=0.8812 image_iou=0.7880 image_precision=0.8454 image_recall=0.9222 image_vi=0.1310 cal_ece=0.0042 cal_nll=0.0412 cal_brier=0.0119 cal_mean_confidence=0.9879 cal_accuracy=0.9841 cal_predicted_prevalence=0.0646 cal_true_prevalence=0.0594 cal_mean_probability=0.0660 +[04:59:57][ 175s][DONE] drosophila-vnc macs-v11-norestart-acq-margin b5 s1 | test patch Dice=0.8160 VI=0.1139 | image Dice=0.8812 VI=0.1310 | 174s diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json new file mode 100644 index 0000000000000000000000000000000000000000..cd5ac387053a1991dd2e45e87abf0351993ca995 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json @@ -0,0 +1,667 @@ +{ + "target": "drosophila-vnc", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 1, + "config": { + "task": "mitochondria", + "target": "drosophila-vnc", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 1, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "margin", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "drosophila-vnc", + "--variants", + "macs-v11-norestart-acq-margin", + "--budgets", + "5", + "--seeds", + "1", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "multiclass" + ], + "alphas": [ + 0.11690531690536884, + 0.8375898290916663, + 0.02560239431814963, + 0.01990245968481518 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "multiclass" + ], + "mixture": [ + 0.11690531690536884, + 0.8375898290916663, + 0.02560239431814963, + 0.01990245968481518 + ], + "per_source": [ + 0.038599381604120485, + 0.943285983973492, + 0.013712262708195049, + 0.004402371714192633 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.1917814945701421, + "mmd2_mixture": 0.19749308036725388, + "mmd2_uniform": 0.3743480458439246, + "mmd2_single_source": [ + 0.7042945447726864, + 0.2070680114579429, + 0.6307945959829456, + 0.6534353423959632 + ], + "median_squared_distance": 0.1964580928585633, + "gammas": [ + 1.2725360221224549, + 2.5450720442449097, + 5.0901440884898195, + 10.180288176979639, + 20.360576353959278 + ], + "max_weight": 0.8375898290916663, + "entropy": 0.5711585797111782, + "effective_sources": 1.7703169170005038, + "min_weight": 0.01990245968481518 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6050756121533214, + 0.3110169227475098, + 0.7002952924749369, + 0.8048259551971948 + ], + "standardized": [ + -0.0012381647254944845, + -1.5993047903844642, + 0.5162347028009382, + 1.0843082523090217 + ], + "tau": 0.5, + "max_weight": 0.943285983973492, + "entropy": 0.2634007298678481, + "effective_sources": 1.3013481035311127, + "min_weight": 0.004402371714192633 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "kasthuri-pp", + "seconds": 1.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.2784035818310388, + 0.18341782772166676, + 0.057702367443012846, + 0.01681466600081078 + ], + "ensemble_val_dice": { + "mixture": 0.186522633793688, + "mixture@raw": 0.18548325188563394, + "mixture@logit": 0.19296315829990748, + "per_source": 0.18415887092754818, + "per_source@raw": 0.18350562678267812, + "per_source@logit": 0.1890579186582894, + "uniform": 0.06967222752121038, + "uniform@raw": 0.0884686387663988, + "uniform@logit": 0.06545324142703345 + }, + "val_patches": 147 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 3.7969236373901367, + 2.597255229949951 + ], + "top_source": "kasthuri-pp", + "weight_stats": { + "max_weight": 0.8375898290916663, + "entropy": 0.5711585797111782, + "effective_sources": 1.7703169170005038, + "min_weight": 0.01990245968481518 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.21081851067789195, + "top_source": "kasthuri-pp", + "top_source_is_best_teacher": false + }, + "init_teacher": "kasthuri-pp", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 36.4, + "round0_val": { + "val_dice": 0.19168290697357765, + "val_vi": 0.22123099590519313, + "val_ece": 0.02235104280276007 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 5, + "selection_seconds": 5.2, + "round_seconds": 23.5, + "teacher_dice_labeled": [ + 0.3118739095489208, + 0.0968435913812229, + 0.0004512053628980276, + 0.00288 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 15368, + "evidence_nonempty_patches": 5, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.12218059783032857, + "restart_teacher_dice": 0.3118739095489208, + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.3032701474337657, + "uniform": 0.018090264826557255, + "top1": 0.3118739095489208, + "top2": 0.3048324685603729, + "top3": 0.3042586054721977 + }, + "val_dice": 0.27092069976166006, + "val_vi": 0.3048773627261147, + "val_ece": 0.019770819577808685 + }, + { + "round": 2, + "n_labeled": 10, + "selection_seconds": 1.3, + "round_seconds": 6.0, + "teacher_dice_labeled": [ + 0.19717803802753986, + 0.1805899733130004, + 0.0003737319807794981, + 0.0023752969121140144 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 18558, + "evidence_nonempty_patches": 8, + "first_correction": false, + "restart": "", + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.18626228722128985, + "uniform": 0.049455026993990014, + "top1": 0.19717803802753986, + "top2": 0.18620293462910215, + "top3": 0.18620970445945298 + }, + "pseudo_student_dice": 0.10775674024954192, + "pseudo_ensemble_dice": 0.0846065037521647, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.39099553159399997, + "val_vi": 0.27599488257770194, + "val_ece": 0.03385775081323113 + }, + { + "round": 3, + "n_labeled": 15, + "selection_seconds": 1.3, + "round_seconds": 22.0, + "teacher_dice_labeled": [ + 0.3640246067431969, + 0.3316745838639298, + 0.0019141719919512118, + 0.0010269183279976834 + ], + "teacher_dice_pool": [ + 0.5423140882497323, + 0.5143849788839585, + 0.003546472356295111, + 0.0002619360467491595 + ], + "evidence_fg_pixels": 24653, + "evidence_nonempty_patches": 11, + "first_correction": false, + "restart": "", + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.3395188668558075, + "uniform": 0.04689973786647725, + "top1": 0.3640246067431969, + "top2": 0.3395821856889235, + "top3": 0.33954572307904063 + }, + "pseudo_student_dice": 0.22056063804832277, + "pseudo_ensemble_dice": 0.16550525645658046, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.47937634338993435, + "val_vi": 0.21128977939475435, + "val_ece": 0.009649798385442703 + }, + { + "round": 4, + "n_labeled": 20, + "selection_seconds": 1.3, + "round_seconds": 22.6, + "teacher_dice_labeled": [ + 0.33894606300963487, + 0.33203760786110254, + 0.003406787294381386, + 0.0008236433469535518 + ], + "teacher_dice_pool": [ + 0.4882980688802892, + 0.5177185039317563, + 0.005067358057912947, + 0.0002118076972313708 + ], + "evidence_fg_pixels": 39174, + "evidence_nonempty_patches": 16, + "first_correction": false, + "restart": "", + "weights": [ + 0.12987252082567358, + 0.8683830988706824, + 0.0009925103168831313, + 0.0007518699867609326 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.33624660092618736, + "uniform": 0.04154563567677867, + "top1": 0.33894606300963487, + "top2": 0.33631352136850234, + "top3": 0.33625433868249766 + }, + "pseudo_student_dice": 0.2663955078828018, + "pseudo_ensemble_dice": 0.14395372526193248, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.5895830169053273, + "val_vi": 0.14701486319551668, + "val_ece": 0.005259925208484618 + }, + { + "round": 5, + "n_labeled": 25, + "selection_seconds": 1.3, + "round_seconds": 22.2, + "teacher_dice_labeled": [ + 0.28347866835979546, + 0.29629355806286156, + 0.005265936456685989, + 0.0011409426433959182 + ], + "teacher_dice_pool": [ + 0.37725700020977354, + 0.46814559181560483, + 0.009089817328683986, + 0.0002059936819967136 + ], + "evidence_fg_pixels": 52837, + "evidence_nonempty_patches": 20, + "first_correction": false, + "restart": "", + "weights": [ + 0.10907589581677156, + 0.8883421701062245, + 0.0014788106325029009, + 0.0011031234445009884 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.2981801868826781, + "uniform": 0.040404132361419064, + "top1": 0.29629355806286156, + "top2": 0.29825831529887437, + "top3": 0.2982222264110502 + }, + "pseudo_student_dice": 0.4488768227000044, + "pseudo_ensemble_dice": 0.042720602432902106, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.7094354984789913, + "val_vi": 0.11197468489454974, + "val_ece": 0.008760546333009399 + }, + { + "round": 6, + "n_labeled": 30, + "selection_seconds": 1.2, + "round_seconds": 22.6, + "teacher_dice_labeled": [ + 0.2724591555035009, + 0.2829514427188626, + 0.005805749140485103, + 0.0011238169277639535 + ], + "teacher_dice_pool": [ + 0.35042379468295565, + 0.45451658602563566, + 0.01047669215596468, + 0.0006093541967623729 + ], + "evidence_fg_pixels": 67074, + "evidence_nonempty_patches": 23, + "first_correction": false, + "restart": "", + "weights": [ + 0.11131163266174687, + 0.8857376563778061, + 0.0016940485074520217, + 0.0012566624529949536 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.284854314782343, + "uniform": 0.041772513841434294, + "top1": 0.2829514427188626, + "top2": 0.284905995057746, + "top3": 0.28487357388139867 + }, + "pseudo_student_dice": 0.39869975876027314, + "pseudo_ensemble_dice": 0.05862639635912288, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.6698150851720401, + "val_vi": 0.1305862633741253, + "val_ece": 0.011866765588902227 + }, + { + "round": 7, + "n_labeled": 30, + "selection_seconds": 0.0, + "round_seconds": 6.2, + "final_consolidation": true, + "val_dice": 0.7055667561203255, + "val_vi": 0.11683635188920495, + "val_ece": 0.011193001351058354 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.08899991394046061, + "auroc_top20_error": 0.5872474346768222 + }, + "al_seconds": 128.2, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.11131163266174687, + 0.8857376563778061, + 0.0016940485074520217, + 0.0012566624529949536 + ], + "weight_stats_final": { + "max_weight": 0.8857376563778061, + "entropy": 0.37104918671331083, + "effective_sources": 1.449254355692067, + "min_weight": 0.0012566624529949536 + }, + "top_source_final": "kasthuri-pp", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.812722205200734, + "patch_iou": 0.725063429876011, + "patch_precision": 0.7900888338957477, + "patch_recall": 0.905797142842391, + "patch_vi": 0.11934194083093992, + "n_patches": 147, + "cal_ece": 0.01095954651793581, + "cal_nll": 0.05754046776538521, + "cal_brier": 0.013845200752527742, + "cal_mean_confidence": 0.994078189838488, + "cal_accuracy": 0.9832456420068028, + "cal_predicted_prevalence": 0.06584821428571429, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.06584790728883756 + }, + "test_notta": { + "patch_dice": 0.812722205200734, + "patch_iou": 0.725063429876011, + "patch_precision": 0.7900888338957477, + "patch_recall": 0.905797142842391, + "patch_vi": 0.11934194083093992, + "n_patches": 147, + "cal_ece": 0.01095954651793581, + "cal_nll": 0.05754046776538521, + "cal_brier": 0.013845200752527742, + "cal_mean_confidence": 0.994078189838488, + "cal_accuracy": 0.9832456420068028, + "cal_predicted_prevalence": 0.06584821428571429, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.06584790728883756 + }, + "test": { + "patch_dice": 0.8160149021553218, + "patch_iou": 0.728680459250354, + "patch_precision": 0.7997959914735588, + "patch_recall": 0.9001494344816307, + "patch_vi": 0.11392365994158125, + "n_patches": 147, + "image_dice": 0.8812000396184795, + "image_iou": 0.7880471247693754, + "image_precision": 0.8454001509224097, + "image_recall": 0.9222251231022218, + "image_vi": 0.1309689113182238, + "n_images": 3, + "cal_ece": 0.004152246751486293, + "cal_nll": 0.04119385390428756, + "cal_brier": 0.011927609300364279, + "cal_mean_confidence": 0.987949771162949, + "cal_accuracy": 0.9841225552721088, + "cal_predicted_prevalence": 0.06462585034013606, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.06603734781981793 + }, + "eval_seconds": 2.6, + "total_seconds": 174.4, + "resources": { + "phases": { + "setup": { + "seconds": 0.61, + "peak_cpu_rss_mb": 1026.1, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 5.97, + "peak_cpu_rss_mb": 1806.7, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 36.39, + "peak_cpu_rss_mb": 2105.8, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4168.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 128.19, + "peak_cpu_rss_mb": 2391.1, + "peak_gpu_alloc_mb": 3948.5, + "peak_gpu_reserved_mb": 4176.0, + "rounds": 6 + }, + "inference": { + "seconds": 2.58, + "peak_cpu_rss_mb": 2391.1, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4176.0, + "patches": 147, + "images": 3, + "patches_per_second": 57.0, + "seconds_per_image": 0.86 + } + }, + "peak_gpu_alloc_mb": 3948.5, + "peak_cpu_rss_mb": 2391.1, + "total_seconds": 173.7 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/drosophila-vnc/macs-v11-norestart-acq-margin_b5_s1.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..8e675a8c82e56fa5a02640948d95aa88d9c451f9 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,5,5.2,23.5,"[0.3118739095489208, 0.0968435913812229, 0.0004512053628980276, 0.00288]",,15368,5,True,kept,0.12218059783032857,0.3118739095489208,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.3032701474337657, 'uniform': 0.018090264826557255, 'top1': 0.3118739095489208, 'top2': 0.3048324685603729, 'top3': 0.3042586054721977}",0.27092069976166006,0.3048773627261147,0.019770819577808685,,,,, +2,10,1.3,6.0,"[0.19717803802753986, 0.1805899733130004, 0.0003737319807794981, 0.0023752969121140144]",,18558,8,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.18626228722128985, 'uniform': 0.049455026993990014, 'top1': 0.19717803802753986, 'top2': 0.18620293462910215, 'top3': 0.18620970445945298}",0.39099553159399997,0.27599488257770194,0.03385775081323113,0.10775674024954192,0.0846065037521647,student,0.0, +3,15,1.3,22.0,"[0.3640246067431969, 0.3316745838639298, 0.0019141719919512118, 0.0010269183279976834]","[0.5423140882497323, 0.5143849788839585, 0.003546472356295111, 0.0002619360467491595]",24653,11,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.3395188668558075, 'uniform': 0.04689973786647725, 'top1': 0.3640246067431969, 'top2': 0.3395821856889235, 'top3': 0.33954572307904063}",0.47937634338993435,0.21128977939475435,0.009649798385442703,0.22056063804832277,0.16550525645658046,student,0.0, +4,20,1.3,22.6,"[0.33894606300963487, 0.33203760786110254, 0.003406787294381386, 0.0008236433469535518]","[0.4882980688802892, 0.5177185039317563, 0.005067358057912947, 0.0002118076972313708]",39174,16,False,,,,"[0.12987252082567358, 0.8683830988706824, 0.0009925103168831313, 0.0007518699867609326]",mixture,"{'mixture': 0.33624660092618736, 'uniform': 0.04154563567677867, 'top1': 0.33894606300963487, 'top2': 0.33631352136850234, 'top3': 0.33625433868249766}",0.5895830169053273,0.14701486319551668,0.005259925208484618,0.2663955078828018,0.14395372526193248,student,0.0, +5,25,1.3,22.2,"[0.28347866835979546, 0.29629355806286156, 0.005265936456685989, 0.0011409426433959182]","[0.37725700020977354, 0.46814559181560483, 0.009089817328683986, 0.0002059936819967136]",52837,20,False,,,,"[0.10907589581677156, 0.8883421701062245, 0.0014788106325029009, 0.0011031234445009884]",mixture,"{'mixture': 0.2981801868826781, 'uniform': 0.040404132361419064, 'top1': 0.29629355806286156, 'top2': 0.29825831529887437, 'top3': 0.2982222264110502}",0.7094354984789913,0.11197468489454974,0.008760546333009399,0.4488768227000044,0.042720602432902106,student,0.0, +6,30,1.2,22.6,"[0.2724591555035009, 0.2829514427188626, 0.005805749140485103, 0.0011238169277639535]","[0.35042379468295565, 0.45451658602563566, 0.01047669215596468, 0.0006093541967623729]",67074,23,False,,,,"[0.11131163266174687, 0.8857376563778061, 0.0016940485074520217, 0.0012566624529949536]",mixture,"{'mixture': 0.284854314782343, 'uniform': 0.041772513841434294, 'top1': 0.2829514427188626, 'top2': 0.284905995057746, 'top3': 0.28487357388139867}",0.6698150851720401,0.1305862633741253,0.011866765588902227,0.39869975876027314,0.05862639635912288,student,0.0, +7,30,0.0,6.2,,,,,,,,,,,,0.7055667561203255,0.11683635188920495,0.011193001351058354,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/log.txt b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..12baa7a3792b83d00a9533fe26f3079e7d9e202f --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/log.txt @@ -0,0 +1,103 @@ +[04:25:06][ 0s][START] target=drosophila-vnc variant=macs-v11-norestart-acq-random budget=5.0% seed=1 | {"task": "mitochondria", "target": "drosophila-vnc", "variant": "macs-v11-norestart-acq-random", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "random", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 0, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[04:25:06][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[04:25:06][ 0s][DATA] drosophila-vnc: train=686 val=147 test=147 patches +[04:25:08][ 2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, multiclass=2.60 +[04:25:09][ 2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[04:25:10][ 3s][WEIGHTS] mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020 | eff.sources=1.77 | MMD2 mix=0.1975 uniform=0.3743 +[04:25:10][ 3s][WEIGHTS] per-source: fly-mito=0.039, kasthuri=0.943, lucchi=0.014, multiclass=0.004 | min-MMD source (NeuroADDA rule) = kasthuri +[04:25:13][ 7s][TEACHERS] target-val Dice: fly-mito=0.278, kasthuri=0.183, lucchi=0.058, multiclass=0.017 +[04:25:13][ 7s][TEACHERS] ensemble target-val Dice: mixture=0.187, mixture@raw=0.185, mixture@logit=0.193, per_source=0.184, per_source@raw=0.184, per_source@logit=0.189, uniform=0.070, uniform@raw=0.088, uniform@logit=0.065 +[04:25:13][ 7s][WEIGHTS] using mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020 +[04:25:13][ 7s][INIT] student initialised from teacher kasthuri (weight 0.838) +[04:25:15][ 8s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[04:25:15][ 8s][KD] distilling the mixture ensemble (prob space) into the student +[04:25:19][ 12s][KD] soft labels for 686 patches from 4 teachers in 4s +[04:25:25][ 19s][KD] epoch 1/5 | patches=686 | loss=0.2005 | 6s +[04:25:32][ 25s][KD] epoch 2/5 | patches=686 | loss=0.1322 | 6s +[04:25:38][ 32s][KD] epoch 3/5 | patches=686 | loss=0.1258 | 6s +[04:25:44][ 38s][KD] epoch 4/5 | patches=686 | loss=0.1216 | 6s +[04:25:51][ 44s][KD] epoch 5/5 | patches=686 | loss=0.1204 | 6s +[04:25:51][ 45s][ROUND0] before any target label: val_dice=0.1917 val_vi=0.2213 +[04:25:51][ 45s][AL] patches=686 (pool 686 after dropping blank; free blank negatives: True) | budget=34 (5.0%) | rounds=6 | per_round=5 | acquisition=random | first_round=stratified +[04:25:52][ 46s][AL] round-1 uncertainty quality on 686 patches: {'spearman_uncertainty_error': -0.03470887967342099, 'auroc_top20_error': 0.482987650099299} +[04:25:56][ 50s][AL] round 1: stratified cold start: 3 patches from the membrane-rich half (ensemble prevalence >= 0.047) + 2 from the membrane-poor half of 686 candidates +[04:25:57][ 50s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.303, uniform=0.018, top1=0.312, top2=0.305, top3=0.304 -> using top1 +[04:25:57][ 51s][WEIGHTS] round 1: student kept (labeled Dice 0.122 vs best teacher fly-mito 0.312); distilling the corrected ensemble into it +[04:25:57][ 51s][WEIGHTS] round 1: teacher Dice on 5 labeled patches + 0 blank (15368 fg px): fly-mito=0.312, kasthuri=0.097, lucchi=0.000, multiclass=0.003 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[04:25:58][ 52s][KD] soft labels for 681 patches from 4 teachers in 1s +[04:26:05][ 58s][KD] epoch 1/2 | patches=681 | loss=0.2886 | 7s +[04:26:11][ 65s][KD] epoch 2/2 | patches=681 | loss=0.2383 | 6s +[04:26:12][ 66s][CONS] pseudo-labels for 681 unlabeled patches from the current ensemble in 1s +[04:26:13][ 66s][FT] round 1/6 epoch 1/5 | patches=5 | loss=1.2861 +[04:26:13][ 67s][FT] round 1/6 epoch 2/5 | patches=5 | loss=1.2453 +[04:26:13][ 67s][FT] round 1/6 epoch 3/5 | patches=5 | loss=1.1900 +[04:26:14][ 67s][FT] round 1/6 epoch 4/5 | patches=5 | loss=1.1676 +[04:26:14][ 68s][FT] round 1/6 epoch 5/5 | patches=5 | loss=1.1582 +[04:26:15][ 68s][AL] round 1/6 | labeled=5 | selection_seconds=5.3000 round_seconds=23.3000 restart_student_dice=0.1221 restart_teacher_dice=0.3119 val_dice=0.2704 val_vi=0.3045 val_ece=0.0198 +[04:26:16][ 69s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.228, uniform=0.029, top1=0.337, top2=0.229, top3=0.229 -> using top1 +[04:26:16][ 69s][WEIGHTS] round 2: teacher Dice on 10 labeled patches + 0 blank (29992 fg px): fly-mito=0.337, kasthuri=0.208, lucchi=0.000, multiclass=0.001 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[04:26:17][ 71s][CONS] round 2: on the 5 new patches student Dice=0.402 vs ensemble 0.361 -> pseudo-labels from the student | consistency x0.00 +[04:26:18][ 71s][FT] round 2/6 epoch 1/5 | patches=10 | loss=0.9670 +[04:26:18][ 72s][FT] round 2/6 epoch 2/5 | patches=10 | loss=0.9125 +[04:26:19][ 72s][FT] round 2/6 epoch 3/5 | patches=10 | loss=0.8827 +[04:26:19][ 73s][FT] round 2/6 epoch 4/5 | patches=10 | loss=0.8156 +[04:26:19][ 73s][FT] round 2/6 epoch 5/5 | patches=10 | loss=0.7817 +[04:26:20][ 74s][AL] round 2/6 | labeled=10 | selection_seconds=0.1000 round_seconds=4.7000 pseudo_student_dice=0.4019 pseudo_ensemble_dice=0.3610 consistency_scale=0.0000 val_dice=0.3955 val_vi=0.2676 val_ece=0.0101 +[04:26:29][ 83s][WEIGHTS] round 3: teacher agreement with the student on 671 unlabeled patches: fly-mito=0.583, kasthuri=0.506, lucchi=0.004, multiclass=0.000 | blended with labeled Dice at 0.50 +[04:26:37][ 91s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.379, uniform=0.031, top1=0.478, top2=0.380, top3=0.379 -> using top1 +[04:26:37][ 91s][WEIGHTS] round 3: teacher Dice on 15 labeled patches + 0 blank (53089 fg px): fly-mito=0.478, kasthuri=0.362, lucchi=0.002, multiclass=0.001 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[04:26:39][ 93s][CONS] round 3: on the 5 new patches student Dice=0.659 vs ensemble 0.427 -> pseudo-labels from the student | consistency x0.53 +[04:26:39][ 93s][FT] round 3/6 epoch 1/5 | patches=15 | loss=0.7948 +[04:26:40][ 94s][FT] round 3/6 epoch 2/5 | patches=15 | loss=0.7941 +[04:26:41][ 94s][FT] round 3/6 epoch 3/5 | patches=15 | loss=0.7283 +[04:26:41][ 95s][FT] round 3/6 epoch 4/5 | patches=15 | loss=0.7077 +[04:26:42][ 96s][FT] round 3/6 epoch 5/5 | patches=15 | loss=0.6730 +[04:26:42][ 96s][AL] round 3/6 | labeled=15 | selection_seconds=0.1000 round_seconds=22.0000 pseudo_student_dice=0.6589 pseudo_ensemble_dice=0.4271 consistency_scale=0.5300 val_dice=0.5028 val_vi=0.1864 val_ece=0.0084 +[04:26:52][ 105s][WEIGHTS] round 4: teacher agreement with the student on 666 unlabeled patches: fly-mito=0.484, kasthuri=0.500, lucchi=0.004, multiclass=0.000 | blended with labeled Dice at 0.50 +[04:26:59][ 113s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.373, uniform=0.035, top1=0.431, top2=0.373, top3=0.373 -> using top1 +[04:26:59][ 113s][WEIGHTS] round 4: teacher Dice on 20 labeled patches + 0 blank (74091 fg px): fly-mito=0.431, kasthuri=0.363, lucchi=0.006, multiclass=0.001 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000 +[04:27:01][ 115s][CONS] round 4: on the 5 new patches student Dice=0.649 vs ensemble 0.391 -> pseudo-labels from the student | consistency x0.50 +[04:27:02][ 116s][FT] round 4/6 epoch 1/5 | patches=20 | loss=0.5785 +[04:27:03][ 116s][FT] round 4/6 epoch 2/5 | patches=20 | loss=0.5677 +[04:27:04][ 117s][FT] round 4/6 epoch 3/5 | patches=20 | loss=0.4914 +[04:27:04][ 118s][FT] round 4/6 epoch 4/5 | patches=20 | loss=0.5238 +[04:27:05][ 119s][FT] round 4/6 epoch 5/5 | patches=20 | loss=0.4591 +[04:27:05][ 119s][AL] round 4/6 | labeled=20 | selection_seconds=0.1000 round_seconds=22.6000 pseudo_student_dice=0.6493 pseudo_ensemble_dice=0.3907 consistency_scale=0.4980 val_dice=0.6575 val_vi=0.1353 val_ece=0.0086 +[04:27:14][ 128s][WEIGHTS] round 5: teacher agreement with the student on 661 unlabeled patches: fly-mito=0.371, kasthuri=0.460, lucchi=0.009, multiclass=0.001 | blended with labeled Dice at 0.50 +[04:27:22][ 136s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.385, uniform=0.054, top1=0.382, top2=0.385, top3=0.385 -> using mixture +[04:27:22][ 136s][WEIGHTS] round 5: teacher Dice on 25 labeled patches + 0 blank (90327 fg px): fly-mito=0.372, kasthuri=0.382, lucchi=0.008, multiclass=0.004 | weights: fly-mito=0.111, kasthuri=0.888, lucchi=0.001, multiclass=0.000 +[04:27:24][ 137s][CONS] round 5: on the 5 new patches student Dice=0.829 vs ensemble 0.593 -> pseudo-labels from the student | consistency x1.00 +[04:27:25][ 138s][FT] round 5/6 epoch 1/5 | patches=25 | loss=0.4421 +[04:27:25][ 139s][FT] round 5/6 epoch 2/5 | patches=25 | loss=0.3746 +[04:27:26][ 140s][FT] round 5/6 epoch 3/5 | patches=25 | loss=0.3486 +[04:27:27][ 141s][FT] round 5/6 epoch 4/5 | patches=25 | loss=0.3556 +[04:27:28][ 142s][FT] round 5/6 epoch 5/5 | patches=25 | loss=0.3895 +[04:27:29][ 142s][AL] round 5/6 | labeled=25 | selection_seconds=0.1000 round_seconds=22.6000 pseudo_student_dice=0.8290 pseudo_ensemble_dice=0.5930 consistency_scale=1.0000 val_dice=0.6519 val_vi=0.1272 val_ece=0.0108 +[04:27:38][ 151s][WEIGHTS] round 6: teacher agreement with the student on 656 unlabeled patches: fly-mito=0.399, kasthuri=0.472, lucchi=0.010, multiclass=0.001 | blended with labeled Dice at 0.50 +[04:27:45][ 159s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.376, uniform=0.049, top1=0.376, top2=0.376, top3=0.376 -> using mixture +[04:27:45][ 159s][WEIGHTS] round 6: teacher Dice on 30 labeled patches + 0 blank (102597 fg px): fly-mito=0.376, kasthuri=0.374, lucchi=0.008, multiclass=0.003 | weights: fly-mito=0.125, kasthuri=0.873, lucchi=0.001, multiclass=0.001 +[04:27:47][ 161s][CONS] round 6: on the 5 new patches student Dice=0.783 vs ensemble 0.000 -> pseudo-labels from the student | consistency x0.94 +[04:27:48][ 162s][FT] round 6/6 epoch 1/5 | patches=30 | loss=0.3306 +[04:27:49][ 163s][FT] round 6/6 epoch 2/5 | patches=30 | loss=0.3561 +[04:27:50][ 163s][FT] round 6/6 epoch 3/5 | patches=30 | loss=0.3140 +[04:27:51][ 164s][FT] round 6/6 epoch 4/5 | patches=30 | loss=0.3094 +[04:27:52][ 165s][FT] round 6/6 epoch 5/5 | patches=30 | loss=0.3001 +[04:27:52][ 166s][AL] round 6/6 | labeled=30 | selection_seconds=0.1000 round_seconds=23.1000 pseudo_student_dice=0.7834 pseudo_ensemble_dice=0.0001 consistency_scale=0.9450 val_dice=0.6547 val_vi=0.1273 val_ece=0.0109 +[04:27:53][ 167s][CONS] final consolidation: pseudo-labels refreshed from the student +[04:27:54][ 168s][FT] final 1/10 epoch 1/1 | patches=30 | loss=0.2706 +[04:27:55][ 169s][FT] final 2/10 epoch 1/1 | patches=30 | loss=0.2997 +[04:27:56][ 169s][FT] final 3/10 epoch 1/1 | patches=30 | loss=0.3040 +[04:27:57][ 170s][FT] final 4/10 epoch 1/1 | patches=30 | loss=0.2837 +[04:27:58][ 171s][FT] final 5/10 epoch 1/1 | patches=30 | loss=0.2883 +[04:27:58][ 172s][FT] final 6/10 epoch 1/1 | patches=30 | loss=0.2767 +[04:27:59][ 173s][FT] final 7/10 epoch 1/1 | patches=30 | loss=0.2819 +[04:28:00][ 174s][FT] final 8/10 epoch 1/1 | patches=30 | loss=0.2961 +[04:28:01][ 175s][FT] final 9/10 epoch 1/1 | patches=30 | loss=0.2780 +[04:28:02][ 176s][FT] final 10/10 epoch 1/1 | patches=30 | loss=0.3077 +[04:28:02][ 176s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=9.9000 val_dice=0.6747 val_vi=0.1178 val_ece=0.0097 +[04:28:02][ 176s][WEIGHTS] final: fly-mito=0.125, kasthuri=0.873, lucchi=0.001, multiclass=0.001 +[04:28:05][ 178s][TEST] raw student (no EMA, no TTA): patch_dice=0.7961 patch_vi=0.1242 +[04:28:05][ 179s][TEST] final model without TTA: patch_dice=0.7961 patch_vi=0.1242 +[04:28:05][ 179s][RESOURCES] {"phases": {"setup": {"seconds": 1.6, "peak_cpu_rss_mb": 1025.6, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 6.31, "peak_cpu_rss_mb": 1806.9, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 36.04, "peak_cpu_rss_mb": 2105.9, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4168.0, "kd_epochs": 5}, "active_learning": {"seconds": 131.45, "peak_cpu_rss_mb": 2382.2, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4178.0, "rounds": 6}, "inference": {"seconds": 2.73, "peak_cpu_rss_mb": 2382.2, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4178.0, "patches": 147, "images": 3, "patches_per_second": 53.8, "seconds_per_image": 0.91}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 2382.2, "total_seconds": 178.1} +[04:28:05][ 179s][TEST] patch_dice=0.7957 patch_iou=0.7131 patch_precision=0.7525 patch_recall=0.9245 patch_vi=0.1160 image_dice=0.8688 image_iou=0.7684 image_precision=0.7991 image_recall=0.9525 image_vi=0.1422 cal_ece=0.0046 cal_nll=0.0431 cal_brier=0.0125 cal_mean_confidence=0.9878 cal_accuracy=0.9832 cal_predicted_prevalence=0.0705 cal_true_prevalence=0.0594 cal_mean_probability=0.0712 +[04:28:05][ 179s][DONE] drosophila-vnc macs-v11-norestart-acq-random b5 s1 | test patch Dice=0.7957 VI=0.1160 | image Dice=0.8688 VI=0.1422 | 179s diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/result.json b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/result.json new file mode 100644 index 0000000000000000000000000000000000000000..c30730b8b19d530ecec1b58c90f6be8c10b320b1 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/result.json @@ -0,0 +1,667 @@ +{ + "target": "drosophila-vnc", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 1, + "config": { + "task": "mitochondria", + "target": "drosophila-vnc", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 1, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "random", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 0, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "drosophila-vnc", + "--variants", + "macs-v11-norestart-acq-random", + "--budgets", + "5", + "--seeds", + "1", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "multiclass" + ], + "alphas": [ + 0.11690531690536884, + 0.8375898290916663, + 0.02560239431814963, + 0.01990245968481518 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "multiclass" + ], + "mixture": [ + 0.11690531690536884, + 0.8375898290916663, + 0.02560239431814963, + 0.01990245968481518 + ], + "per_source": [ + 0.038599381604120485, + 0.943285983973492, + 0.013712262708195049, + 0.004402371714192633 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.1917814945701421, + "mmd2_mixture": 0.19749308036725388, + "mmd2_uniform": 0.3743480458439246, + "mmd2_single_source": [ + 0.7042945447726864, + 0.2070680114579429, + 0.6307945959829456, + 0.6534353423959632 + ], + "median_squared_distance": 0.1964580928585633, + "gammas": [ + 1.2725360221224549, + 2.5450720442449097, + 5.0901440884898195, + 10.180288176979639, + 20.360576353959278 + ], + "max_weight": 0.8375898290916663, + "entropy": 0.5711585797111782, + "effective_sources": 1.7703169170005038, + "min_weight": 0.01990245968481518 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6050756121533214, + 0.3110169227475098, + 0.7002952924749369, + 0.8048259551971948 + ], + "standardized": [ + -0.0012381647254944845, + -1.5993047903844642, + 0.5162347028009382, + 1.0843082523090217 + ], + "tau": 0.5, + "max_weight": 0.943285983973492, + "entropy": 0.2634007298678481, + "effective_sources": 1.3013481035311127, + "min_weight": 0.004402371714192633 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "kasthuri-pp", + "seconds": 1.4 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.2784035818310388, + 0.18341782772166676, + 0.057702367443012846, + 0.01681466600081078 + ], + "ensemble_val_dice": { + "mixture": 0.186522633793688, + "mixture@raw": 0.18548325188563394, + "mixture@logit": 0.19296315829990748, + "per_source": 0.18415887092754818, + "per_source@raw": 0.18350562678267812, + "per_source@logit": 0.1890579186582894, + "uniform": 0.06967222752121038, + "uniform@raw": 0.0884686387663988, + "uniform@logit": 0.06545324142703345 + }, + "val_patches": 147 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 3.7969236373901367, + 2.597255229949951 + ], + "top_source": "kasthuri-pp", + "weight_stats": { + "max_weight": 0.8375898290916663, + "entropy": 0.5711585797111782, + "effective_sources": 1.7703169170005038, + "min_weight": 0.01990245968481518 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.21081851067789195, + "top_source": "kasthuri-pp", + "top_source_is_best_teacher": false + }, + "init_teacher": "kasthuri-pp", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 36.0, + "round0_val": { + "val_dice": 0.1916880550436862, + "val_vi": 0.22128545106748224, + "val_ece": 0.022343801622776858 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 5, + "selection_seconds": 5.3, + "round_seconds": 23.3, + "teacher_dice_labeled": [ + 0.3118739095489208, + 0.0968435913812229, + 0.0004512053628980276, + 0.00288 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 15368, + "evidence_nonempty_patches": 5, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.12206444565811032, + "restart_teacher_dice": 0.3118739095489208, + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.3032701474337657, + "uniform": 0.018090264826557255, + "top1": 0.3118739095489208, + "top2": 0.3048324685603729, + "top3": 0.3042586054721977 + }, + "val_dice": 0.27044419233563916, + "val_vi": 0.304520179729652, + "val_ece": 0.019762184201704298 + }, + { + "round": 2, + "n_labeled": 10, + "selection_seconds": 0.1, + "round_seconds": 4.7, + "teacher_dice_labeled": [ + 0.3370766113617249, + 0.20785448050939587, + 0.00023106885851983892, + 0.001479435841798994 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 29992, + "evidence_nonempty_patches": 10, + "first_correction": false, + "restart": "", + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.2284639770019887, + "uniform": 0.028564987762462964, + "top1": 0.3370766113617249, + "top2": 0.2287742341421212, + "top3": 0.22868552543573198 + }, + "pseudo_student_dice": 0.4019437835222003, + "pseudo_ensemble_dice": 0.36104428904428904, + "pseudo_source": "student", + "consistency_scale": 0.0, + "val_dice": 0.395526844758151, + "val_vi": 0.2676356453097999, + "val_ece": 0.01013508053922452 + }, + { + "round": 3, + "n_labeled": 15, + "selection_seconds": 0.1, + "round_seconds": 22.0, + "teacher_dice_labeled": [ + 0.47845312015423197, + 0.36233065584123325, + 0.0023601996475788446, + 0.001159658464534431 + ], + "teacher_dice_pool": [ + 0.5826728360901532, + 0.5055378509654219, + 0.0036168323423117956, + 0.00028675870280224237 + ], + "evidence_fg_pixels": 53089, + "evidence_nonempty_patches": 15, + "first_correction": false, + "restart": "", + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.3793255215978283, + "uniform": 0.03080981377991152, + "top1": 0.47845312015423197, + "top2": 0.3795297534477732, + "top3": 0.37938025962169 + }, + "pseudo_student_dice": 0.6588509676155887, + "pseudo_ensemble_dice": 0.42708432425086634, + "pseudo_source": "student", + "consistency_scale": 0.53, + "val_dice": 0.5028134313637753, + "val_vi": 0.18638684223811713, + "val_ece": 0.008429511681865535 + }, + { + "round": 4, + "n_labeled": 20, + "selection_seconds": 0.1, + "round_seconds": 22.6, + "teacher_dice_labeled": [ + 0.43103000518725343, + 0.36277141788004924, + 0.005746551389353614, + 0.0008508898528234805 + ], + "teacher_dice_pool": [ + 0.4837028345554021, + 0.49975188054577785, + 0.004083124007295061, + 0.00024411224105223264 + ], + "evidence_fg_pixels": 74091, + "evidence_nonempty_patches": 20, + "first_correction": false, + "restart": "", + "weights": [ + 1.0, + 0.0, + 0.0, + 0.0 + ], + "ensemble_choice": "top1", + "candidate_dice": { + "mixture": 0.373334630363959, + "uniform": 0.03538747088305604, + "top1": 0.43103000518725343, + "top2": 0.37340299844945096, + "top3": 0.37335681419917816 + }, + "pseudo_student_dice": 0.6492734286625866, + "pseudo_ensemble_dice": 0.3906538327265473, + "pseudo_source": "student", + "consistency_scale": 0.498, + "val_dice": 0.6574593702393732, + "val_vi": 0.13533919737072217, + "val_ece": 0.008640088643453371 + }, + { + "round": 5, + "n_labeled": 25, + "selection_seconds": 0.1, + "round_seconds": 22.6, + "teacher_dice_labeled": [ + 0.3717128901337308, + 0.3823762412712888, + 0.007664640876645464, + 0.0036914687359017786 + ], + "teacher_dice_pool": [ + 0.3705947151012748, + 0.45955121435857227, + 0.009201777039825764, + 0.0009343618515797726 + ], + "evidence_fg_pixels": 90327, + "evidence_nonempty_patches": 25, + "first_correction": false, + "restart": "", + "weights": [ + 0.11134689678576087, + 0.8875352175350215, + 0.0006398570780882527, + 0.0004780286011293887 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.3852127875716316, + "uniform": 0.054485108960520695, + "top1": 0.3823762412712888, + "top2": 0.385270066563859, + "top3": 0.38524165066760796 + }, + "pseudo_student_dice": 0.8290223463687151, + "pseudo_ensemble_dice": 0.5929518365361226, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.6519417313245782, + "val_vi": 0.12718020187778287, + "val_ece": 0.010796660762073279 + }, + { + "round": 6, + "n_labeled": 30, + "selection_seconds": 0.1, + "round_seconds": 23.1, + "teacher_dice_labeled": [ + 0.37641538950337, + 0.37352917480538395, + 0.008009229163548612, + 0.003443806587578693 + ], + "teacher_dice_pool": [ + 0.39919716938468497, + 0.47226146654321266, + 0.009783755993179285, + 0.0012095196263971187 + ], + "evidence_fg_pixels": 102597, + "evidence_nonempty_patches": 28, + "first_correction": false, + "restart": "", + "weights": [ + 0.12546358417446743, + 0.8733335622071413, + 0.0006902340557613062, + 0.0005126195626298668 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.37627428788395567, + "uniform": 0.04938321963255888, + "top1": 0.37641538950337, + "top2": 0.37633076172426516, + "top3": 0.376285460732609 + }, + "pseudo_student_dice": 0.7833869929169349, + "pseudo_ensemble_dice": 5.972288580984233e-05, + "pseudo_source": "student", + "consistency_scale": 0.945, + "val_dice": 0.6547061788185442, + "val_vi": 0.12729252241884847, + "val_ece": 0.01092049264393179 + }, + { + "round": 7, + "n_labeled": 30, + "selection_seconds": 0.0, + "round_seconds": 9.9, + "final_consolidation": true, + "val_dice": 0.6747453638476713, + "val_vi": 0.11778407187789014, + "val_ece": 0.009716465947843022 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.03470887967342099, + "auroc_top20_error": 0.482987650099299 + }, + "al_seconds": 131.5, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.12546358417446743, + 0.8733335622071413, + 0.0006902340557613062, + 0.0005126195626298668 + ], + "weight_stats_final": { + "max_weight": 0.8733335622071413, + "entropy": 0.38761949128992906, + "effective_sources": 1.4734690098128682, + "min_weight": 0.0005126195626298668 + }, + "top_source_final": "kasthuri-pp", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.7961039268265155, + "patch_iou": 0.7097687229145664, + "patch_precision": 0.7584498601881322, + "patch_recall": 0.9158267507576621, + "patch_vi": 0.1241707856182842, + "n_patches": 147, + "cal_ece": 0.01067422514685362, + "cal_nll": 0.05508691845768595, + "cal_brier": 0.014360713568413223, + "cal_mean_confidence": 0.9927506494665814, + "cal_accuracy": 0.9820764243197279, + "cal_predicted_prevalence": 0.0682530824829932, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.06837096344262414 + }, + "test_notta": { + "patch_dice": 0.7961039268265155, + "patch_iou": 0.7097687229145664, + "patch_precision": 0.7584498601881322, + "patch_recall": 0.9158267507576621, + "patch_vi": 0.1241707856182842, + "n_patches": 147, + "cal_ece": 0.01067422514685362, + "cal_nll": 0.05508691845768595, + "cal_brier": 0.014360713568413223, + "cal_mean_confidence": 0.9927506494665814, + "cal_accuracy": 0.9820764243197279, + "cal_predicted_prevalence": 0.0682530824829932, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.06837096344262414 + }, + "test": { + "patch_dice": 0.795692102630151, + "patch_iou": 0.7131124461490234, + "patch_precision": 0.752512679245492, + "patch_recall": 0.9245430931406642, + "patch_vi": 0.11597981232523923, + "n_patches": 147, + "image_dice": 0.8687959458062932, + "image_iou": 0.7683731492929882, + "image_precision": 0.79911133488317, + "image_recall": 0.952452859311531, + "image_vi": 0.14220201389233786, + "n_images": 3, + "cal_ece": 0.004598624949523295, + "cal_nll": 0.04306924903348975, + "cal_brier": 0.012508827748498425, + "cal_mean_confidence": 0.9878069916271571, + "cal_accuracy": 0.9832389987244898, + "cal_predicted_prevalence": 0.07049186862244898, + "cal_true_prevalence": 0.05939094387755102, + "cal_mean_probability": 0.07124218746769279 + }, + "eval_seconds": 2.7, + "total_seconds": 178.8, + "resources": { + "phases": { + "setup": { + "seconds": 1.6, + "peak_cpu_rss_mb": 1025.6, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 6.31, + "peak_cpu_rss_mb": 1806.9, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 36.04, + "peak_cpu_rss_mb": 2105.9, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4168.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 131.45, + "peak_cpu_rss_mb": 2382.2, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4178.0, + "rounds": 6 + }, + "inference": { + "seconds": 2.73, + "peak_cpu_rss_mb": 2382.2, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4178.0, + "patches": 147, + "images": 3, + "patches_per_second": 53.8, + "seconds_per_image": 0.91 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 2382.2, + "total_seconds": 178.1 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/drosophila-vnc/macs-v11-norestart-acq-random_b5_s1.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/rounds.csv b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..8211ed4a0b49ca6fdd640ff1a307e2b0ceea36cf --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-random/budget-5/seed-1/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,5,5.3,23.3,"[0.3118739095489208, 0.0968435913812229, 0.0004512053628980276, 0.00288]",,15368,5,True,kept,0.12206444565811032,0.3118739095489208,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.3032701474337657, 'uniform': 0.018090264826557255, 'top1': 0.3118739095489208, 'top2': 0.3048324685603729, 'top3': 0.3042586054721977}",0.27044419233563916,0.304520179729652,0.019762184201704298,,,,, +2,10,0.1,4.7,"[0.3370766113617249, 0.20785448050939587, 0.00023106885851983892, 0.001479435841798994]",,29992,10,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.2284639770019887, 'uniform': 0.028564987762462964, 'top1': 0.3370766113617249, 'top2': 0.2287742341421212, 'top3': 0.22868552543573198}",0.395526844758151,0.2676356453097999,0.01013508053922452,0.4019437835222003,0.36104428904428904,student,0.0, +3,15,0.1,22.0,"[0.47845312015423197, 0.36233065584123325, 0.0023601996475788446, 0.001159658464534431]","[0.5826728360901532, 0.5055378509654219, 0.0036168323423117956, 0.00028675870280224237]",53089,15,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.3793255215978283, 'uniform': 0.03080981377991152, 'top1': 0.47845312015423197, 'top2': 0.3795297534477732, 'top3': 0.37938025962169}",0.5028134313637753,0.18638684223811713,0.008429511681865535,0.6588509676155887,0.42708432425086634,student,0.53, +4,20,0.1,22.6,"[0.43103000518725343, 0.36277141788004924, 0.005746551389353614, 0.0008508898528234805]","[0.4837028345554021, 0.49975188054577785, 0.004083124007295061, 0.00024411224105223264]",74091,20,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.373334630363959, 'uniform': 0.03538747088305604, 'top1': 0.43103000518725343, 'top2': 0.37340299844945096, 'top3': 0.37335681419917816}",0.6574593702393732,0.13533919737072217,0.008640088643453371,0.6492734286625866,0.3906538327265473,student,0.498, +5,25,0.1,22.6,"[0.3717128901337308, 0.3823762412712888, 0.007664640876645464, 0.0036914687359017786]","[0.3705947151012748, 0.45955121435857227, 0.009201777039825764, 0.0009343618515797726]",90327,25,False,,,,"[0.11134689678576087, 0.8875352175350215, 0.0006398570780882527, 0.0004780286011293887]",mixture,"{'mixture': 0.3852127875716316, 'uniform': 0.054485108960520695, 'top1': 0.3823762412712888, 'top2': 0.385270066563859, 'top3': 0.38524165066760796}",0.6519417313245782,0.12718020187778287,0.010796660762073279,0.8290223463687151,0.5929518365361226,student,1.0, +6,30,0.1,23.1,"[0.37641538950337, 0.37352917480538395, 0.008009229163548612, 0.003443806587578693]","[0.39919716938468497, 0.47226146654321266, 0.009783755993179285, 0.0012095196263971187]",102597,28,False,,,,"[0.12546358417446743, 0.8733335622071413, 0.0006902340557613062, 0.0005126195626298668]",mixture,"{'mixture': 0.37627428788395567, 'uniform': 0.04938321963255888, 'top1': 0.37641538950337, 'top2': 0.37633076172426516, 'top3': 0.376285460732609}",0.6547061788185442,0.12729252241884847,0.01092049264393179,0.7833869929169349,5.972288580984233e-05,student,0.945, +7,30,0.0,9.9,,,,,,,,,,,,0.6747453638476713,0.11778407187789014,0.009716465947843022,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..a51f56304ca8371d62cf644e43a23ed2bdaee728 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt @@ -0,0 +1,108 @@ +[07:27:23][ 0s][START] target=kasthuri-pp variant=macs-v11-norestart-acq-laplace budget=5.0% seed=2 | {"task": "mitochondria", "target": "kasthuri-pp", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "laplace", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[07:27:23][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[07:27:23][ 0s][DATA] kasthuri-pp: train=12023 val=2376 test=2376 patches +[07:27:24][ 1s][CALIB] source-val temperatures: fly-mito=1.38, lucchi=3.80, dros-vnc=2.77, multiclass=2.60 +[07:27:24][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[07:27:25][ 2s][WEIGHTS] mixture: fly-mito=0.055, lucchi=0.143, dros-vnc=0.678, multiclass=0.124 | eff.sources=2.61 | MMD2 mix=0.1257 uniform=0.2386 +[07:27:25][ 2s][WEIGHTS] per-source: fly-mito=0.004, lucchi=0.028, dros-vnc=0.948, multiclass=0.020 | min-MMD source (NeuroADDA rule) = dros-vnc +[07:27:38][ 15s][TEACHERS] target-val Dice: fly-mito=0.194, lucchi=0.137, dros-vnc=0.349, multiclass=0.152 +[07:27:38][ 15s][TEACHERS] ensemble target-val Dice: mixture=0.341, mixture@raw=0.346, mixture@logit=0.321, per_source=0.348, per_source@raw=0.348, per_source@logit=0.346, uniform=0.188, uniform@raw=0.183, uniform@logit=0.207 +[07:27:38][ 15s][WEIGHTS] using mixture: fly-mito=0.055, lucchi=0.143, dros-vnc=0.678, multiclass=0.124 +[07:27:38][ 15s][INIT] student initialised from teacher dros-vnc (weight 0.678) +[07:27:39][ 16s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[07:27:39][ 16s][KD] distilling the mixture ensemble (prob space) into the student +[07:28:14][ 51s][KD] soft labels for 6000 patches from 4 teachers in 34s +[07:29:07][ 104s][KD] epoch 1/5 | patches=6000 | loss=0.1778 | 53s +[07:30:02][ 159s][KD] epoch 2/5 | patches=6000 | loss=0.1510 | 54s +[07:30:56][ 213s][KD] epoch 3/5 | patches=6000 | loss=0.1461 | 54s +[07:31:49][ 266s][KD] epoch 4/5 | patches=6000 | loss=0.1440 | 53s +[07:32:43][ 320s][KD] epoch 5/5 | patches=6000 | loss=0.1430 | 54s +[07:32:45][ 322s][ROUND0] before any target label: val_dice=0.2118 val_vi=0.1496 +[07:32:45][ 322s][AL] patches=12023 (pool 10327 after dropping blank; free blank negatives: True) | budget=601 (5.0%) | rounds=6 | per_round=100 | acquisition=laplace | first_round=stratified +[07:33:10][ 347s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': -0.36017774401252334, 'auroc_top20_error': 0.360394235162642} +[07:33:22][ 359s][AL] round 1: stratified cold start: 50 patches from the membrane-rich half (ensemble prevalence >= 0.034) + 50 from the membrane-poor half of 2000 candidates +[07:33:28][ 365s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.221, uniform=0.011, top1=0.223, top2=0.223, top3=0.222 -> using mixture +[07:33:28][ 365s][WEIGHTS] round 1: student kept (labeled Dice 0.013 vs best teacher dros-vnc 0.223); distilling the corrected ensemble into it +[07:33:28][ 365s][WEIGHTS] round 1: teacher Dice on 100 labeled patches + 100 blank (267058 fg px): fly-mito=0.038, lucchi=0.009, dros-vnc=0.223, multiclass=0.000 | weights: fly-mito=0.012, lucchi=0.023, dros-vnc=0.946, multiclass=0.019 +[07:34:03][ 400s][KD] soft labels for 6000 patches from 4 teachers in 34s +[07:34:57][ 454s][KD] epoch 1/2 | patches=6000 | loss=0.0535 | 54s +[07:35:51][ 508s][KD] epoch 2/2 | patches=6000 | loss=0.0528 | 54s +[07:36:03][ 520s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 12s +[07:36:07][ 524s][FT] round 1/6 epoch 1/5 | patches=200 | loss=0.7622 +[07:36:11][ 528s][FT] round 1/6 epoch 2/5 | patches=200 | loss=0.5386 +[07:36:15][ 532s][FT] round 1/6 epoch 3/5 | patches=200 | loss=0.4307 +[07:36:19][ 536s][FT] round 1/6 epoch 4/5 | patches=200 | loss=0.3971 +[07:36:23][ 540s][FT] round 1/6 epoch 5/5 | patches=200 | loss=0.3589 +[07:36:25][ 542s][AL] round 1/6 | labeled=100 | selection_seconds=37.5000 round_seconds=218.1000 restart_student_dice=0.0133 restart_teacher_dice=0.2227 val_dice=0.7285 val_vi=0.0893 val_ece=0.0073 +[07:36:50][ 567s][AL] round 2: 2834/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:37:00][ 577s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.130, uniform=0.008, top1=0.132, top2=0.133, top3=0.131 -> using mixture +[07:37:00][ 577s][WEIGHTS] round 2: teacher Dice on 200 labeled patches + 200 blank (1115399 fg px): fly-mito=0.084, lucchi=0.012, dros-vnc=0.132, multiclass=0.007 | weights: fly-mito=0.043, lucchi=0.054, dros-vnc=0.858, multiclass=0.045 +[07:37:08][ 585s][CONS] round 2: on the 100 new patches student Dice=0.552 vs ensemble 0.100 -> pseudo-labels from the student | consistency x0.17 +[07:37:15][ 592s][FT] round 2/6 epoch 1/5 | patches=400 | loss=0.2396 +[07:37:23][ 600s][FT] round 2/6 epoch 2/5 | patches=400 | loss=0.2152 +[07:37:30][ 607s][FT] round 2/6 epoch 3/5 | patches=400 | loss=0.2085 +[07:37:37][ 614s][FT] round 2/6 epoch 4/5 | patches=400 | loss=0.1984 +[07:37:44][ 621s][FT] round 2/6 epoch 5/5 | patches=400 | loss=0.2184 +[07:37:46][ 623s][AL] round 2/6 | labeled=200 | selection_seconds=25.5000 round_seconds=79.6000 pseudo_student_dice=0.5515 pseudo_ensemble_dice=0.0995 consistency_scale=0.1720 val_dice=0.8179 val_vi=0.0591 val_ece=0.0027 +[07:38:12][ 649s][AL] round 3: 2464/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:38:44][ 681s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.112, lucchi=0.015, dros-vnc=0.285, multiclass=0.001 | blended with labeled Dice at 0.50 +[07:39:13][ 710s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.207, uniform=0.012, top1=0.208, top2=0.209, top3=0.208 -> using mixture +[07:39:13][ 710s][WEIGHTS] round 3: teacher Dice on 300 labeled patches + 300 blank (1632966 fg px): fly-mito=0.098, lucchi=0.021, dros-vnc=0.208, multiclass=0.004 | weights: fly-mito=0.025, lucchi=0.030, dros-vnc=0.923, multiclass=0.022 +[07:39:21][ 718s][CONS] round 3: on the 100 new patches student Dice=0.781 vs ensemble 0.130 -> pseudo-labels from the student | consistency x0.94 +[07:39:31][ 728s][FT] round 3/6 epoch 1/5 | patches=600 | loss=0.2248 +[07:39:42][ 739s][FT] round 3/6 epoch 2/5 | patches=600 | loss=0.1817 +[07:39:53][ 750s][FT] round 3/6 epoch 3/5 | patches=600 | loss=0.1835 +[07:40:03][ 760s][FT] round 3/6 epoch 4/5 | patches=600 | loss=0.1582 +[07:40:14][ 771s][FT] round 3/6 epoch 5/5 | patches=600 | loss=0.1784 +[07:40:16][ 773s][AL] round 3/6 | labeled=300 | selection_seconds=25.4000 round_seconds=147.7000 pseudo_student_dice=0.7809 pseudo_ensemble_dice=0.1295 consistency_scale=0.9360 val_dice=0.8398 val_vi=0.0550 val_ece=0.0030 +[07:40:41][ 798s][AL] round 4: 2492/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:41:16][ 833s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.113, lucchi=0.014, dros-vnc=0.303, multiclass=0.001 | blended with labeled Dice at 0.50 +[07:41:47][ 864s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.222, uniform=0.013, top1=0.223, top2=0.224, top3=0.223 -> using mixture +[07:41:47][ 864s][WEIGHTS] round 4: teacher Dice on 400 labeled patches + 400 blank (2188516 fg px): fly-mito=0.097, lucchi=0.018, dros-vnc=0.223, multiclass=0.004 | weights: fly-mito=0.022, lucchi=0.025, dros-vnc=0.934, multiclass=0.019 +[07:41:55][ 872s][CONS] round 4: on the 100 new patches student Dice=0.815 vs ensemble 0.176 -> pseudo-labels from the student | consistency x1.00 +[07:42:09][ 886s][FT] round 4/6 epoch 1/5 | patches=800 | loss=0.1741 +[07:42:23][ 900s][FT] round 4/6 epoch 2/5 | patches=800 | loss=0.1575 +[07:42:37][ 914s][FT] round 4/6 epoch 3/5 | patches=800 | loss=0.1496 +[07:42:51][ 928s][FT] round 4/6 epoch 4/5 | patches=800 | loss=0.1385 +[07:43:05][ 942s][FT] round 4/6 epoch 5/5 | patches=800 | loss=0.1491 +[07:43:07][ 944s][AL] round 4/6 | labeled=400 | selection_seconds=25.5000 round_seconds=169.0000 pseudo_student_dice=0.8146 pseudo_ensemble_dice=0.1760 consistency_scale=1.0000 val_dice=0.8727 val_vi=0.0459 val_ece=0.0024 +[07:43:32][ 969s][AL] round 5: 2323/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:44:09][ 1006s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.113, lucchi=0.015, dros-vnc=0.264, multiclass=0.001 | blended with labeled Dice at 0.50 +[07:44:42][ 1039s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.207, uniform=0.013, top1=0.209, top2=0.209, top3=0.208 -> using mixture +[07:44:42][ 1039s][WEIGHTS] round 5: teacher Dice on 500 labeled patches + 500 blank (2518881 fg px): fly-mito=0.094, lucchi=0.018, dros-vnc=0.209, multiclass=0.003 | weights: fly-mito=0.024, lucchi=0.029, dros-vnc=0.925, multiclass=0.022 +[07:44:50][ 1047s][CONS] round 5: on the 100 new patches student Dice=0.794 vs ensemble 0.214 -> pseudo-labels from the student | consistency x0.98 +[07:45:08][ 1065s][FT] round 5/6 epoch 1/5 | patches=1000 | loss=0.1510 +[07:45:25][ 1082s][FT] round 5/6 epoch 2/5 | patches=1000 | loss=0.1710 +[07:45:42][ 1099s][FT] round 5/6 epoch 3/5 | patches=1000 | loss=0.1518 +[07:46:00][ 1117s][FT] round 5/6 epoch 4/5 | patches=1000 | loss=0.1546 +[07:46:17][ 1134s][FT] round 5/6 epoch 5/5 | patches=1000 | loss=0.1450 +[07:46:19][ 1136s][AL] round 5/6 | labeled=500 | selection_seconds=25.5000 round_seconds=190.5000 pseudo_student_dice=0.7943 pseudo_ensemble_dice=0.2140 consistency_scale=0.9810 val_dice=0.8634 val_vi=0.0461 val_ece=0.0023 +[07:46:45][ 1162s][AL] round 6: 2522/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[07:47:21][ 1198s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.112, lucchi=0.012, dros-vnc=0.286, multiclass=0.002 | blended with labeled Dice at 0.50 +[07:47:55][ 1232s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.235, uniform=0.012, top1=0.236, top2=0.236, top3=0.235 -> using mixture +[07:47:55][ 1232s][WEIGHTS] round 6: teacher Dice on 512 labeled patches + 512 blank (2489206 fg px): fly-mito=0.093, lucchi=0.015, dros-vnc=0.236, multiclass=0.004 | weights: fly-mito=0.018, lucchi=0.022, dros-vnc=0.943, multiclass=0.017 +[07:48:03][ 1240s][CONS] round 6: on the 100 new patches student Dice=0.945 vs ensemble 0.367 -> pseudo-labels from the student | consistency x1.00 +[07:48:24][ 1261s][FT] round 6/6 epoch 1/5 | patches=1200 | loss=0.1372 +[07:48:45][ 1282s][FT] round 6/6 epoch 2/5 | patches=1200 | loss=0.1167 +[07:49:05][ 1302s][FT] round 6/6 epoch 3/5 | patches=1200 | loss=0.1188 +[07:49:26][ 1323s][FT] round 6/6 epoch 4/5 | patches=1200 | loss=0.1208 +[07:49:47][ 1344s][FT] round 6/6 epoch 5/5 | patches=1200 | loss=0.1142 +[07:49:49][ 1346s][AL] round 6/6 | labeled=600 | selection_seconds=25.4000 round_seconds=207.8000 pseudo_student_dice=0.9455 pseudo_ensemble_dice=0.3667 consistency_scale=1.0000 val_dice=0.8758 val_vi=0.0413 val_ece=0.0024 +[07:49:52][ 1349s][CONS] final consolidation: pseudo-labels refreshed from the student +[07:50:13][ 1370s][FT] final 1/10 epoch 1/1 | patches=1200 | loss=0.1081 +[07:50:34][ 1391s][FT] final 2/10 epoch 1/1 | patches=1200 | loss=0.1131 +[07:50:55][ 1412s][FT] final 3/10 epoch 1/1 | patches=1200 | loss=0.1010 +[07:51:16][ 1433s][FT] final 4/10 epoch 1/1 | patches=1200 | loss=0.1056 +[07:51:36][ 1453s][FT] final 5/10 epoch 1/1 | patches=1200 | loss=0.0999 +[07:51:57][ 1474s][FT] final 6/10 epoch 1/1 | patches=1200 | loss=0.0932 +[07:52:18][ 1495s][FT] final 7/10 epoch 1/1 | patches=1200 | loss=0.0827 +[07:52:39][ 1516s][FT] final 8/10 epoch 1/1 | patches=1200 | loss=0.0815 +[07:53:00][ 1537s][FT] final 9/10 epoch 1/1 | patches=1200 | loss=0.0790 +[07:53:21][ 1558s][FT] final 10/10 epoch 1/1 | patches=1200 | loss=0.0742 +[07:53:23][ 1560s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=211.4000 val_dice=0.8811 val_vi=0.0406 val_ece=0.0026 +[07:53:23][ 1560s][WEIGHTS] final: fly-mito=0.018, lucchi=0.022, dros-vnc=0.943, multiclass=0.017 +[07:53:55][ 1592s][TEST] raw student (no EMA, no TTA): patch_dice=0.8717 patch_vi=0.0270 +[07:53:59][ 1596s][TEST] final model without TTA: patch_dice=0.8717 patch_vi=0.0270 +[07:53:59][ 1596s][RESOURCES] {"phases": {"setup": {"seconds": 0.59, "peak_cpu_rss_mb": 1028.3, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 15.46, "peak_cpu_rss_mb": 1793.0, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 303.55, "peak_cpu_rss_mb": 3685.8, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 1237.79, "peak_cpu_rss_mb": 3685.8, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4170.0, "rounds": 6}, "inference": {"seconds": 36.81, "peak_cpu_rss_mb": 3685.8, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4172.0, "patches": 2376, "images": 24, "patches_per_second": 64.5, "seconds_per_image": 1.534}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3685.8, "total_seconds": 1594.2} +[07:53:59][ 1596s][TEST] patch_dice=0.8806 patch_iou=0.8521 patch_precision=0.9547 patch_recall=0.8873 patch_vi=0.0259 image_dice=0.9483 image_iou=0.9021 image_precision=0.9598 image_recall=0.9376 image_vi=0.0308 cal_ece=0.0014 cal_nll=0.0117 cal_brier=0.0024 cal_mean_confidence=0.9983 cal_accuracy=0.9969 cal_predicted_prevalence=0.0262 cal_true_prevalence=0.0268 cal_mean_probability=0.0260 +[07:53:59][ 1596s][DONE] kasthuri-pp macs-v11-norestart-acq-laplace b5 s2 | test patch Dice=0.8806 VI=0.0259 | image Dice=0.9483 VI=0.0308 | 1596s diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..be3ff8af2e67dbb525c28f4eb40e17230a0da57c --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json @@ -0,0 +1,667 @@ +{ + "target": "kasthuri-pp", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 2, + "config": { + "task": "mitochondria", + "target": "kasthuri-pp", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "laplace", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "kasthuri-pp", + "--variants", + "macs-v11-norestart-acq-laplace", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "lucchi-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.0552621669329701, + 0.14304343546870246, + 0.6777952504109969, + 0.12389914718733061 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "lucchi-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.0552621669329701, + 0.14304343546870246, + 0.6777952504109969, + 0.12389914718733061 + ], + "per_source": [ + 0.00394004854372213, + 0.027549943457875075, + 0.9484681548320345, + 0.02004185316636836 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.1161079474391782, + "mmd2_mixture": 0.12571317215793731, + "mmd2_uniform": 0.23862471954971437, + "mmd2_single_source": [ + 0.7126139881719412, + 0.4366429148044758, + 0.1798181763334552, + 0.4482686385790514 + ], + "median_squared_distance": 0.1734670188179156, + "gammas": [ + 1.4411961518887884, + 2.8823923037775767, + 5.764784607555153, + 11.529569215110307, + 23.059138430220614 + ], + "max_weight": 0.6777952504109969, + "entropy": 0.9605224718759118, + "effective_sources": 2.6130613679073416, + "min_weight": 0.0552621669329701 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6491308049576032, + 0.5568981487675557, + 0.3890679978127308, + 0.5719877547192425 + ], + "standardized": [ + 1.1318865389009094, + 0.1594828192631669, + -1.6099410463963124, + 0.31857168823223964 + ], + "tau": 0.5, + "max_weight": 0.9484681548320345, + "entropy": 0.24930992499892873, + "effective_sources": 1.2831396485053355, + "min_weight": 0.00394004854372213 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "drosophila-vnc", + "seconds": 1.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.19408169161460087, + 0.13693054106355262, + 0.34856214071670877, + 0.15175765833135202 + ], + "ensemble_val_dice": { + "mixture": 0.3408904698785591, + "mixture@raw": 0.3461558372224687, + "mixture@logit": 0.3209424494814286, + "per_source": 0.3475944498943022, + "per_source@raw": 0.34822194749797575, + "per_source@logit": 0.3464301210911742, + "uniform": 0.18819756340126403, + "uniform@raw": 0.18329087899026675, + "uniform@logit": 0.2066047847592749 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 3.7969236373901367, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "drosophila-vnc", + "weight_stats": { + "max_weight": 0.6777952504109969, + "entropy": 0.9605224718759118, + "effective_sources": 2.6130613679073416, + "min_weight": 0.0552621669329701 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.19999999999999998, + "spearman_weight_taxonomy": 0.2581988897471611, + "top_source": "drosophila-vnc", + "top_source_is_best_teacher": true + }, + "init_teacher": "drosophila-vnc", + "student_init": "top-source", + "blank_fraction": 0.14106296265491142, + "kd_seconds": 303.6, + "round0_val": { + "val_dice": 0.2118068600167677, + "val_vi": 0.14964552136814735, + "val_ece": 0.02150047831005253 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 100, + "selection_seconds": 37.5, + "round_seconds": 218.1, + "teacher_dice_labeled": [ + 0.03829273777525091, + 0.008551357968456415, + 0.22265360583181462, + 0.00019730768085296483 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 267058, + "evidence_nonempty_patches": 74, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.013334523583695058, + "restart_teacher_dice": 0.22265360583181462, + "weights": [ + 0.012200911710615017, + 0.0234567119495013, + 0.9456533644481262, + 0.018689011891757392 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.22146026492223692, + "uniform": 0.011353831964774486, + "top1": 0.22265360583181462, + "top2": 0.22287410068430968, + "top3": 0.2221372032521384 + }, + "val_dice": 0.7284880907378732, + "val_vi": 0.08925383861862662, + "val_ece": 0.007318781441033324 + }, + { + "round": 2, + "n_labeled": 200, + "selection_seconds": 25.5, + "round_seconds": 79.6, + "teacher_dice_labeled": [ + 0.0843977025985284, + 0.012172782655249998, + 0.13224230106627227, + 0.0066752619085354665 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1115399, + "evidence_nonempty_patches": 172, + "first_correction": false, + "restart": "", + "weights": [ + 0.043330876352989676, + 0.054471414405189024, + 0.8575402913759741, + 0.04465741786584716 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.12992613885762752, + "uniform": 0.008046029319777243, + "top1": 0.13224230106627227, + "top2": 0.13316217464277688, + "top3": 0.13142508151376586 + }, + "pseudo_student_dice": 0.5515170063829185, + "pseudo_ensemble_dice": 0.09952690898409115, + "pseudo_source": "student", + "consistency_scale": 0.172, + "val_dice": 0.8179183255309906, + "val_vi": 0.05907835029705699, + "val_ece": 0.002730643722707998 + }, + { + "round": 3, + "n_labeled": 300, + "selection_seconds": 25.4, + "round_seconds": 147.7, + "teacher_dice_labeled": [ + 0.09774431100649172, + 0.02100781303993087, + 0.20822007653296637, + 0.0037858675867297508 + ], + "teacher_dice_pool": [ + 0.11166637835235022, + 0.014642946673106777, + 0.28479381985635444, + 0.0009620469771313624 + ], + "evidence_fg_pixels": 1632966, + "evidence_nonempty_patches": 266, + "first_correction": false, + "restart": "", + "weights": [ + 0.02493772672979201, + 0.02996639548739577, + 0.9232464064547601, + 0.021849471328052025 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.20698885628003225, + "uniform": 0.011817528245685245, + "top1": 0.20822007653296637, + "top2": 0.2087118275421432, + "top3": 0.20773206921661225 + }, + "pseudo_student_dice": 0.7809243690957529, + "pseudo_ensemble_dice": 0.1295378560580478, + "pseudo_source": "student", + "consistency_scale": 0.936, + "val_dice": 0.8397631826742316, + "val_vi": 0.05498086933118604, + "val_ece": 0.00298188285613413 + }, + { + "round": 4, + "n_labeled": 400, + "selection_seconds": 25.5, + "round_seconds": 169.0, + "teacher_dice_labeled": [ + 0.09741935827619103, + 0.018257986269871015, + 0.22313621046835663, + 0.0035449807697112965 + ], + "teacher_dice_pool": [ + 0.11298328490431746, + 0.013772936029469702, + 0.30295471895352916, + 0.0013155154647071317 + ], + "evidence_fg_pixels": 2188516, + "evidence_nonempty_patches": 363, + "first_correction": false, + "restart": "", + "weights": [ + 0.021660446784121497, + 0.025404693616115942, + 0.9339408533881395, + 0.018994006211623046 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.22211558136819576, + "uniform": 0.012848261537052798, + "top1": 0.22313621046835663, + "top2": 0.22353551488564827, + "top3": 0.22274276271997542 + }, + "pseudo_student_dice": 0.8146274749309655, + "pseudo_ensemble_dice": 0.17598222940226171, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8727027758814474, + "val_vi": 0.04586085054725591, + "val_ece": 0.0023572107581083693 + }, + { + "round": 5, + "n_labeled": 500, + "selection_seconds": 25.5, + "round_seconds": 190.5, + "teacher_dice_labeled": [ + 0.09423271841560225, + 0.01786067875926255, + 0.20863073975641946, + 0.0032668848418224545 + ], + "teacher_dice_pool": [ + 0.11321393596398284, + 0.015169177507659712, + 0.2639679558408339, + 0.0014020792250121128 + ], + "evidence_fg_pixels": 2518881, + "evidence_nonempty_patches": 451, + "first_correction": false, + "restart": "", + "weights": [ + 0.024031619475181773, + 0.028983011307454676, + 0.9252901596473508, + 0.02169520957001273 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.20744316954220898, + "uniform": 0.01252951462867678, + "top1": 0.20863073975641946, + "top2": 0.2090463951847489, + "top3": 0.2081601937542924 + }, + "pseudo_student_dice": 0.7943479375839712, + "pseudo_ensemble_dice": 0.21403595365511507, + "pseudo_source": "student", + "consistency_scale": 0.981, + "val_dice": 0.8633937643280325, + "val_vi": 0.0460846659403376, + "val_ece": 0.0023077556545834403 + }, + { + "round": 6, + "n_labeled": 600, + "selection_seconds": 25.4, + "round_seconds": 207.8, + "teacher_dice_labeled": [ + 0.09276645918051137, + 0.014889898090616777, + 0.23552871081062293, + 0.0035663265755556104 + ], + "teacher_dice_pool": [ + 0.11218089728921189, + 0.011678871931780969, + 0.28592648339309323, + 0.0019156104138320128 + ], + "evidence_fg_pixels": 2489206, + "evidence_nonempty_patches": 468, + "first_correction": false, + "restart": "", + "weights": [ + 0.018437500364764308, + 0.021904253452186863, + 0.9427167721353997, + 0.01694147404764911 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.23450584653940748, + "uniform": 0.011503959466380224, + "top1": 0.23552871081062293, + "top2": 0.23586462884397835, + "top3": 0.23511377156938024 + }, + "pseudo_student_dice": 0.9454673857874526, + "pseudo_ensemble_dice": 0.36671536038318975, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8758495345285738, + "val_vi": 0.041280698120005795, + "val_ece": 0.00239807538318637 + }, + { + "round": 7, + "n_labeled": 600, + "selection_seconds": 0.0, + "round_seconds": 211.4, + "final_consolidation": true, + "val_dice": 0.8810958794996602, + "val_vi": 0.04057494837156638, + "val_ece": 0.0026333786376941874 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.36017774401252334, + "auroc_top20_error": 0.360394235162642 + }, + "al_seconds": 1237.8, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.018437500364764308, + 0.021904253452186863, + 0.9427167721353997, + 0.01694147404764911 + ], + "weight_stats_final": { + "max_weight": 0.9427167721353997, + "entropy": 0.2820229765274581, + "effective_sources": 1.325809182127911, + "min_weight": 0.01694147404764911 + }, + "top_source_final": "drosophila-vnc", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8716612471825786, + "patch_iou": 0.8425408245737226, + "patch_precision": 0.9459390937256462, + "patch_recall": 0.8854203269001327, + "patch_vi": 0.027044653248263214, + "n_patches": 2376, + "cal_ece": 0.0019236981596546764, + "cal_nll": 0.01505363365568678, + "cal_brier": 0.0027411552110365505, + "cal_mean_confidence": 0.9986372529786782, + "cal_accuracy": 0.9967135548190236, + "cal_predicted_prevalence": 0.026243473142887205, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.025931264006348577 + }, + "test_notta": { + "patch_dice": 0.8716612471825786, + "patch_iou": 0.8425408245737226, + "patch_precision": 0.9459390937256462, + "patch_recall": 0.8854203269001327, + "patch_vi": 0.027044653248263214, + "n_patches": 2376, + "cal_ece": 0.0019236981596546764, + "cal_nll": 0.01505363365568678, + "cal_brier": 0.0027411552110365505, + "cal_mean_confidence": 0.9986372529786782, + "cal_accuracy": 0.9967135548190236, + "cal_predicted_prevalence": 0.026243473142887205, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.025931264006348577 + }, + "test": { + "patch_dice": 0.8806073948945004, + "patch_iou": 0.8520803541413138, + "patch_precision": 0.9547011432703486, + "patch_recall": 0.887258433513542, + "patch_vi": 0.025885128225993523, + "n_patches": 2376, + "image_dice": 0.948303957395134, + "image_iou": 0.9020892014257859, + "image_precision": 0.9597966206412246, + "image_recall": 0.9375817043623197, + "image_vi": 0.03077187452953857, + "n_images": 24, + "cal_ece": 0.0014176467257451166, + "cal_nll": 0.011710549894120235, + "cal_brier": 0.0024250120921702435, + "cal_mean_confidence": 0.9983486264448106, + "cal_accuracy": 0.9969309797190656, + "cal_predicted_prevalence": 0.026230731797138047, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.02602820112381996 + }, + "eval_seconds": 36.8, + "total_seconds": 1596.3, + "resources": { + "phases": { + "setup": { + "seconds": 0.59, + "peak_cpu_rss_mb": 1028.3, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 15.46, + "peak_cpu_rss_mb": 1793.0, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 303.55, + "peak_cpu_rss_mb": 3685.8, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 1237.79, + "peak_cpu_rss_mb": 3685.8, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4170.0, + "rounds": 6 + }, + "inference": { + "seconds": 36.81, + "peak_cpu_rss_mb": 3685.8, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4172.0, + "patches": 2376, + "images": 24, + "patches_per_second": 64.5, + "seconds_per_image": 1.534 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3685.8, + "total_seconds": 1594.2 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/kasthuri-pp/macs-v11-norestart-acq-laplace_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..8c7e2a5d38e94621e5d9199d45e48a4e3f6f4255 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,100,37.5,218.1,"[0.03829273777525091, 0.008551357968456415, 0.22265360583181462, 0.00019730768085296483]",,267058,74,True,kept,0.013334523583695058,0.22265360583181462,"[0.012200911710615017, 0.0234567119495013, 0.9456533644481262, 0.018689011891757392]",mixture,"{'mixture': 0.22146026492223692, 'uniform': 0.011353831964774486, 'top1': 0.22265360583181462, 'top2': 0.22287410068430968, 'top3': 0.2221372032521384}",0.7284880907378732,0.08925383861862662,0.007318781441033324,,,,, +2,200,25.5,79.6,"[0.0843977025985284, 0.012172782655249998, 0.13224230106627227, 0.0066752619085354665]",,1115399,172,False,,,,"[0.043330876352989676, 0.054471414405189024, 0.8575402913759741, 0.04465741786584716]",mixture,"{'mixture': 0.12992613885762752, 'uniform': 0.008046029319777243, 'top1': 0.13224230106627227, 'top2': 0.13316217464277688, 'top3': 0.13142508151376586}",0.8179183255309906,0.05907835029705699,0.002730643722707998,0.5515170063829185,0.09952690898409115,student,0.172, +3,300,25.4,147.7,"[0.09774431100649172, 0.02100781303993087, 0.20822007653296637, 0.0037858675867297508]","[0.11166637835235022, 0.014642946673106777, 0.28479381985635444, 0.0009620469771313624]",1632966,266,False,,,,"[0.02493772672979201, 0.02996639548739577, 0.9232464064547601, 0.021849471328052025]",mixture,"{'mixture': 0.20698885628003225, 'uniform': 0.011817528245685245, 'top1': 0.20822007653296637, 'top2': 0.2087118275421432, 'top3': 0.20773206921661225}",0.8397631826742316,0.05498086933118604,0.00298188285613413,0.7809243690957529,0.1295378560580478,student,0.936, +4,400,25.5,169.0,"[0.09741935827619103, 0.018257986269871015, 0.22313621046835663, 0.0035449807697112965]","[0.11298328490431746, 0.013772936029469702, 0.30295471895352916, 0.0013155154647071317]",2188516,363,False,,,,"[0.021660446784121497, 0.025404693616115942, 0.9339408533881395, 0.018994006211623046]",mixture,"{'mixture': 0.22211558136819576, 'uniform': 0.012848261537052798, 'top1': 0.22313621046835663, 'top2': 0.22353551488564827, 'top3': 0.22274276271997542}",0.8727027758814474,0.04586085054725591,0.0023572107581083693,0.8146274749309655,0.17598222940226171,student,1.0, +5,500,25.5,190.5,"[0.09423271841560225, 0.01786067875926255, 0.20863073975641946, 0.0032668848418224545]","[0.11321393596398284, 0.015169177507659712, 0.2639679558408339, 0.0014020792250121128]",2518881,451,False,,,,"[0.024031619475181773, 0.028983011307454676, 0.9252901596473508, 0.02169520957001273]",mixture,"{'mixture': 0.20744316954220898, 'uniform': 0.01252951462867678, 'top1': 0.20863073975641946, 'top2': 0.2090463951847489, 'top3': 0.2081601937542924}",0.8633937643280325,0.0460846659403376,0.0023077556545834403,0.7943479375839712,0.21403595365511507,student,0.981, +6,600,25.4,207.8,"[0.09276645918051137, 0.014889898090616777, 0.23552871081062293, 0.0035663265755556104]","[0.11218089728921189, 0.011678871931780969, 0.28592648339309323, 0.0019156104138320128]",2489206,468,False,,,,"[0.018437500364764308, 0.021904253452186863, 0.9427167721353997, 0.01694147404764911]",mixture,"{'mixture': 0.23450584653940748, 'uniform': 0.011503959466380224, 'top1': 0.23552871081062293, 'top2': 0.23586462884397835, 'top3': 0.23511377156938024}",0.8758495345285738,0.041280698120005795,0.00239807538318637,0.9454673857874526,0.36671536038318975,student,1.0, +7,600,0.0,211.4,,,,,,,,,,,,0.8810958794996602,0.04057494837156638,0.0026333786376941874,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..15c2b588875bad40e34d450c530f9a57cd7c3447 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/log.txt @@ -0,0 +1,108 @@ +[11:23:13][ 0s][START] target=kasthuri-pp variant=macs-v11-norestart-acq-laplace budget=5.0% seed=3 | {"task": "mitochondria", "target": "kasthuri-pp", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "laplace", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[11:23:13][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[11:23:13][ 0s][DATA] kasthuri-pp: train=12023 val=2376 test=2376 patches +[11:23:14][ 1s][CALIB] source-val temperatures: fly-mito=1.38, lucchi=3.80, dros-vnc=2.77, multiclass=2.60 +[11:23:14][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[11:23:15][ 2s][WEIGHTS] mixture: fly-mito=0.055, lucchi=0.143, dros-vnc=0.678, multiclass=0.124 | eff.sources=2.61 | MMD2 mix=0.1257 uniform=0.2386 +[11:23:15][ 2s][WEIGHTS] per-source: fly-mito=0.004, lucchi=0.028, dros-vnc=0.948, multiclass=0.020 | min-MMD source (NeuroADDA rule) = dros-vnc +[11:23:28][ 15s][TEACHERS] target-val Dice: fly-mito=0.194, lucchi=0.137, dros-vnc=0.349, multiclass=0.152 +[11:23:28][ 15s][TEACHERS] ensemble target-val Dice: mixture=0.341, mixture@raw=0.346, mixture@logit=0.321, per_source=0.348, per_source@raw=0.348, per_source@logit=0.346, uniform=0.188, uniform@raw=0.183, uniform@logit=0.207 +[11:23:28][ 15s][WEIGHTS] using mixture: fly-mito=0.055, lucchi=0.143, dros-vnc=0.678, multiclass=0.124 +[11:23:28][ 15s][INIT] student initialised from teacher dros-vnc (weight 0.678) +[11:23:29][ 16s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[11:23:29][ 16s][KD] distilling the mixture ensemble (prob space) into the student +[11:24:04][ 51s][KD] soft labels for 6000 patches from 4 teachers in 34s +[11:24:59][ 106s][KD] epoch 1/5 | patches=6000 | loss=0.1796 | 55s +[11:25:52][ 159s][KD] epoch 2/5 | patches=6000 | loss=0.1520 | 54s +[11:26:46][ 213s][KD] epoch 3/5 | patches=6000 | loss=0.1470 | 54s +[11:28:27][ 314s][KD] epoch 4/5 | patches=6000 | loss=0.1441 | 101s +[11:30:15][ 422s][KD] epoch 5/5 | patches=6000 | loss=0.1430 | 107s +[11:30:18][ 425s][ROUND0] before any target label: val_dice=0.2180 val_vi=0.1501 +[11:30:18][ 425s][AL] patches=12023 (pool 10327 after dropping blank; free blank negatives: True) | budget=601 (5.0%) | rounds=6 | per_round=100 | acquisition=laplace | first_round=stratified +[11:31:14][ 481s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': -0.36780975100244245, 'auroc_top20_error': 0.3513228515625} +[11:31:39][ 506s][AL] round 1: stratified cold start: 50 patches from the membrane-rich half (ensemble prevalence >= 0.034) + 50 from the membrane-poor half of 2000 candidates +[11:31:50][ 517s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.292, uniform=0.033, top1=0.292, top2=0.292, top3=0.292 -> using mixture +[11:31:52][ 518s][WEIGHTS] round 1: student kept (labeled Dice 0.048 vs best teacher dros-vnc 0.292); distilling the corrected ensemble into it +[11:31:52][ 518s][WEIGHTS] round 1: teacher Dice on 100 labeled patches + 100 blank (264499 fg px): fly-mito=0.038, lucchi=0.021, dros-vnc=0.292, multiclass=0.011 | weights: fly-mito=0.006, lucchi=0.014, dros-vnc=0.970, multiclass=0.011 +[11:33:04][ 591s][KD] soft labels for 6000 patches from 4 teachers in 72s +[11:34:51][ 698s][KD] epoch 1/2 | patches=6000 | loss=0.0434 | 108s +[11:36:37][ 804s][KD] epoch 2/2 | patches=6000 | loss=0.0426 | 106s +[11:37:01][ 828s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 23s +[11:37:08][ 835s][FT] round 1/6 epoch 1/5 | patches=200 | loss=0.9069 +[11:37:16][ 843s][FT] round 1/6 epoch 2/5 | patches=200 | loss=0.5702 +[11:37:23][ 850s][FT] round 1/6 epoch 3/5 | patches=200 | loss=0.4298 +[11:37:31][ 858s][FT] round 1/6 epoch 4/5 | patches=200 | loss=0.3568 +[11:37:38][ 865s][FT] round 1/6 epoch 5/5 | patches=200 | loss=0.3961 +[11:37:42][ 869s][AL] round 1/6 | labeled=100 | selection_seconds=80.4000 round_seconds=440.0000 restart_student_dice=0.0475 restart_teacher_dice=0.2924 val_dice=0.8030 val_vi=0.0753 val_ece=0.0010 +[11:38:38][ 924s][AL] round 2: 2139/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:38:59][ 946s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.197, uniform=0.028, top1=0.199, top2=0.198, top3=0.198 -> using mixture +[11:38:59][ 946s][WEIGHTS] round 2: teacher Dice on 200 labeled patches + 200 blank (642529 fg px): fly-mito=0.047, lucchi=0.048, dros-vnc=0.199, multiclass=0.013 | weights: fly-mito=0.016, lucchi=0.043, dros-vnc=0.915, multiclass=0.026 +[11:39:13][ 960s][CONS] round 2: on the 100 new patches student Dice=0.877 vs ensemble 0.126 -> pseudo-labels from the student | consistency x1.00 +[11:39:28][ 975s][FT] round 2/6 epoch 1/5 | patches=400 | loss=0.2787 +[11:39:42][ 989s][FT] round 2/6 epoch 2/5 | patches=400 | loss=0.2590 +[11:39:57][ 1004s][FT] round 2/6 epoch 3/5 | patches=400 | loss=0.2339 +[11:40:12][ 1019s][FT] round 2/6 epoch 4/5 | patches=400 | loss=0.2013 +[11:40:27][ 1034s][FT] round 2/6 epoch 5/5 | patches=400 | loss=0.2166 +[11:40:30][ 1037s][AL] round 2/6 | labeled=200 | selection_seconds=55.9000 round_seconds=164.8000 pseudo_student_dice=0.8770 pseudo_ensemble_dice=0.1257 consistency_scale=1.0000 val_dice=0.8291 val_vi=0.0618 val_ece=0.0039 +[11:41:25][ 1092s][AL] round 3: 2446/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:42:34][ 1161s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.114, lucchi=0.013, dros-vnc=0.303, multiclass=0.002 | blended with labeled Dice at 0.50 +[11:43:49][ 1236s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.227, uniform=0.018, top1=0.228, top2=0.228, top3=0.227 -> using mixture +[11:43:49][ 1236s][WEIGHTS] round 3: teacher Dice on 300 labeled patches + 300 blank (1098795 fg px): fly-mito=0.083, lucchi=0.021, dros-vnc=0.228, multiclass=0.005 | weights: fly-mito=0.018, lucchi=0.025, dros-vnc=0.939, multiclass=0.018 +[11:44:00][ 1247s][CONS] round 3: on the 100 new patches student Dice=0.822 vs ensemble 0.084 -> pseudo-labels from the student | consistency x1.00 +[11:44:21][ 1268s][FT] round 3/6 epoch 1/5 | patches=600 | loss=0.1953 +[11:44:42][ 1289s][FT] round 3/6 epoch 2/5 | patches=600 | loss=0.1731 +[11:45:03][ 1309s][FT] round 3/6 epoch 3/5 | patches=600 | loss=0.1763 +[11:45:21][ 1328s][FT] round 3/6 epoch 4/5 | patches=600 | loss=0.1600 +[11:45:42][ 1349s][FT] round 3/6 epoch 5/5 | patches=600 | loss=0.1707 +[11:45:45][ 1352s][AL] round 3/6 | labeled=300 | selection_seconds=54.7000 round_seconds=311.8000 pseudo_student_dice=0.8216 pseudo_ensemble_dice=0.0836 consistency_scale=1.0000 val_dice=0.8311 val_vi=0.0579 val_ece=0.0040 +[11:46:41][ 1408s][AL] round 4: 2504/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:47:55][ 1482s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.110, lucchi=0.011, dros-vnc=0.288, multiclass=0.001 | blended with labeled Dice at 0.50 +[11:49:14][ 1561s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.229, uniform=0.018, top1=0.230, top2=0.230, top3=0.229 -> using mixture +[11:49:14][ 1561s][WEIGHTS] round 4: teacher Dice on 400 labeled patches + 400 blank (1530549 fg px): fly-mito=0.080, lucchi=0.023, dros-vnc=0.230, multiclass=0.003 | weights: fly-mito=0.017, lucchi=0.025, dros-vnc=0.940, multiclass=0.018 +[11:49:29][ 1575s][CONS] round 4: on the 100 new patches student Dice=0.858 vs ensemble 0.213 -> pseudo-labels from the student | consistency x1.00 +[11:49:57][ 1604s][FT] round 4/6 epoch 1/5 | patches=800 | loss=0.1402 +[11:50:25][ 1632s][FT] round 4/6 epoch 2/5 | patches=800 | loss=0.1328 +[11:50:54][ 1661s][FT] round 4/6 epoch 3/5 | patches=800 | loss=0.1419 +[11:51:23][ 1690s][FT] round 4/6 epoch 4/5 | patches=800 | loss=0.1426 +[11:51:52][ 1719s][FT] round 4/6 epoch 5/5 | patches=800 | loss=0.1410 +[11:51:56][ 1723s][AL] round 4/6 | labeled=400 | selection_seconds=55.9000 round_seconds=366.7000 pseudo_student_dice=0.8578 pseudo_ensemble_dice=0.2135 consistency_scale=1.0000 val_dice=0.6930 val_vi=0.0950 val_ece=0.0103 +[11:52:50][ 1777s][AL] round 5: 3237/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[11:54:08][ 1855s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.088, lucchi=0.009, dros-vnc=0.308, multiclass=0.000 | blended with labeled Dice at 0.50 +[11:55:14][ 1921s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.250, uniform=0.016, top1=0.250, top2=0.251, top3=0.250 -> using mixture +[11:55:14][ 1921s][WEIGHTS] round 5: teacher Dice on 500 labeled patches + 500 blank (2059997 fg px): fly-mito=0.073, lucchi=0.019, dros-vnc=0.250, multiclass=0.002 | weights: fly-mito=0.013, lucchi=0.020, dros-vnc=0.952, multiclass=0.015 +[11:55:28][ 1935s][CONS] round 5: on the 100 new patches student Dice=0.671 vs ensemble 0.249 -> pseudo-labels from the student | consistency x0.57 +[11:56:03][ 1970s][FT] round 5/6 epoch 1/5 | patches=1000 | loss=0.1204 +[11:56:38][ 2005s][FT] round 5/6 epoch 2/5 | patches=1000 | loss=0.1131 +[11:57:14][ 2041s][FT] round 5/6 epoch 3/5 | patches=1000 | loss=0.1260 +[11:57:49][ 2076s][FT] round 5/6 epoch 4/5 | patches=1000 | loss=0.1066 +[11:58:25][ 2112s][FT] round 5/6 epoch 5/5 | patches=1000 | loss=0.1011 +[11:58:28][ 2115s][AL] round 5/6 | labeled=500 | selection_seconds=54.7000 round_seconds=389.0000 pseudo_student_dice=0.6707 pseudo_ensemble_dice=0.2487 consistency_scale=0.5690 val_dice=0.7672 val_vi=0.0775 val_ece=0.0082 +[11:59:22][ 2169s][AL] round 6: 2992/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[12:00:39][ 2246s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.093, lucchi=0.014, dros-vnc=0.305, multiclass=0.001 | blended with labeled Dice at 0.50 +[12:01:58][ 2325s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.255, uniform=0.016, top1=0.256, top2=0.256, top3=0.255 -> using mixture +[12:01:58][ 2325s][WEIGHTS] round 6: teacher Dice on 512 labeled patches + 512 blank (2082676 fg px): fly-mito=0.074, lucchi=0.021, dros-vnc=0.256, multiclass=0.003 | weights: fly-mito=0.013, lucchi=0.019, dros-vnc=0.954, multiclass=0.014 +[12:02:12][ 2339s][CONS] round 6: on the 100 new patches student Dice=0.895 vs ensemble 0.256 -> pseudo-labels from the student | consistency x1.00 +[12:02:54][ 2381s][FT] round 6/6 epoch 1/5 | patches=1200 | loss=0.1111 +[12:03:37][ 2424s][FT] round 6/6 epoch 2/5 | patches=1200 | loss=0.1149 +[12:04:19][ 2466s][FT] round 6/6 epoch 3/5 | patches=1200 | loss=0.1021 +[12:05:02][ 2509s][FT] round 6/6 epoch 4/5 | patches=1200 | loss=0.0941 +[12:05:44][ 2551s][FT] round 6/6 epoch 5/5 | patches=1200 | loss=0.1079 +[12:05:47][ 2554s][AL] round 6/6 | labeled=600 | selection_seconds=54.3000 round_seconds=435.5000 pseudo_student_dice=0.8952 pseudo_ensemble_dice=0.2560 consistency_scale=1.0000 val_dice=0.8360 val_vi=0.0573 val_ece=0.0029 +[12:05:53][ 2560s][CONS] final consolidation: pseudo-labels refreshed from the student +[12:06:36][ 2603s][FT] final 1/10 epoch 1/1 | patches=1200 | loss=0.0935 +[12:07:16][ 2643s][FT] final 2/10 epoch 1/1 | patches=1200 | loss=0.0845 +[12:07:54][ 2681s][FT] final 3/10 epoch 1/1 | patches=1200 | loss=0.0828 +[12:08:36][ 2723s][FT] final 4/10 epoch 1/1 | patches=1200 | loss=0.0879 +[12:09:18][ 2765s][FT] final 5/10 epoch 1/1 | patches=1200 | loss=0.0823 +[12:10:01][ 2808s][FT] final 6/10 epoch 1/1 | patches=1200 | loss=0.0824 +[12:10:43][ 2850s][FT] final 7/10 epoch 1/1 | patches=1200 | loss=0.0711 +[12:11:26][ 2892s][FT] final 8/10 epoch 1/1 | patches=1200 | loss=0.0739 +[12:12:07][ 2934s][FT] final 9/10 epoch 1/1 | patches=1200 | loss=0.0623 +[12:12:50][ 2977s][FT] final 10/10 epoch 1/1 | patches=1200 | loss=0.0652 +[12:12:54][ 2981s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=423.1000 val_dice=0.8612 val_vi=0.0500 val_ece=0.0025 +[12:12:54][ 2981s][WEIGHTS] final: fly-mito=0.013, lucchi=0.019, dros-vnc=0.954, multiclass=0.014 +[12:13:58][ 3045s][TEST] raw student (no EMA, no TTA): patch_dice=0.8603 patch_vi=0.0306 +[12:14:06][ 3053s][TEST] final model without TTA: patch_dice=0.8603 patch_vi=0.0306 +[12:14:06][ 3053s][RESOURCES] {"phases": {"setup": {"seconds": 0.58, "peak_cpu_rss_mb": 1028.8, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 15.31, "peak_cpu_rss_mb": 1793.7, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 405.7, "peak_cpu_rss_mb": 3686.1, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 2555.43, "peak_cpu_rss_mb": 3686.1, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4170.0, "rounds": 6}, "inference": {"seconds": 72.52, "peak_cpu_rss_mb": 3686.1, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4172.0, "patches": 2376, "images": 24, "patches_per_second": 32.8, "seconds_per_image": 3.022}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3686.1, "total_seconds": 3049.5} +[12:14:06][ 3053s][TEST] patch_dice=0.8626 patch_iou=0.8293 patch_precision=0.9730 patch_recall=0.8453 patch_vi=0.0300 image_dice=0.9320 image_iou=0.8731 image_precision=0.9772 image_recall=0.8913 image_vi=0.0359 cal_ece=0.0010 cal_nll=0.0139 cal_brier=0.0030 cal_mean_confidence=0.9969 cal_accuracy=0.9963 cal_predicted_prevalence=0.0245 cal_true_prevalence=0.0268 cal_mean_probability=0.0232 +[12:14:06][ 3053s][DONE] kasthuri-pp macs-v11-norestart-acq-laplace b5 s3 | test patch Dice=0.8626 VI=0.0300 | image Dice=0.9320 VI=0.0359 | 3053s diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..c1c144e8bcb72614f884dafc67a9c61f1597835f --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/result.json @@ -0,0 +1,667 @@ +{ + "target": "kasthuri-pp", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 3, + "config": { + "task": "mitochondria", + "target": "kasthuri-pp", + "variant": "macs-v11-norestart-acq-laplace", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "laplace", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "kasthuri-pp", + "--variants", + "macs-v11-norestart-acq-laplace", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "lucchi-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.0552621669329701, + 0.14304343546870246, + 0.6777952504109969, + 0.12389914718733061 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "lucchi-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.0552621669329701, + 0.14304343546870246, + 0.6777952504109969, + 0.12389914718733061 + ], + "per_source": [ + 0.00394004854372213, + 0.027549943457875075, + 0.9484681548320345, + 0.02004185316636836 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.1161079474391782, + "mmd2_mixture": 0.12571317215793731, + "mmd2_uniform": 0.23862471954971437, + "mmd2_single_source": [ + 0.7126139881719412, + 0.4366429148044758, + 0.1798181763334552, + 0.4482686385790514 + ], + "median_squared_distance": 0.1734670188179156, + "gammas": [ + 1.4411961518887884, + 2.8823923037775767, + 5.764784607555153, + 11.529569215110307, + 23.059138430220614 + ], + "max_weight": 0.6777952504109969, + "entropy": 0.9605224718759118, + "effective_sources": 2.6130613679073416, + "min_weight": 0.0552621669329701 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6491308049576032, + 0.5568981487675557, + 0.3890679978127308, + 0.5719877547192425 + ], + "standardized": [ + 1.1318865389009094, + 0.1594828192631669, + -1.6099410463963124, + 0.31857168823223964 + ], + "tau": 0.5, + "max_weight": 0.9484681548320345, + "entropy": 0.24930992499892873, + "effective_sources": 1.2831396485053355, + "min_weight": 0.00394004854372213 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "drosophila-vnc", + "seconds": 1.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.19408169161460087, + 0.13693054106355262, + 0.34856214071670877, + 0.15175765833135202 + ], + "ensemble_val_dice": { + "mixture": 0.3408904698785591, + "mixture@raw": 0.3461558372224687, + "mixture@logit": 0.3209424494814286, + "per_source": 0.3475944498943022, + "per_source@raw": 0.34822194749797575, + "per_source@logit": 0.3464301210911742, + "uniform": 0.18819756340126403, + "uniform@raw": 0.18329087899026675, + "uniform@logit": 0.2066047847592749 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 3.7969236373901367, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "drosophila-vnc", + "weight_stats": { + "max_weight": 0.6777952504109969, + "entropy": 0.9605224718759118, + "effective_sources": 2.6130613679073416, + "min_weight": 0.0552621669329701 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.19999999999999998, + "spearman_weight_taxonomy": 0.2581988897471611, + "top_source": "drosophila-vnc", + "top_source_is_best_teacher": true + }, + "init_teacher": "drosophila-vnc", + "student_init": "top-source", + "blank_fraction": 0.14106296265491142, + "kd_seconds": 405.7, + "round0_val": { + "val_dice": 0.21798321217103253, + "val_vi": 0.15009799279698172, + "val_ece": 0.027540151789761623 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 100, + "selection_seconds": 80.4, + "round_seconds": 440.0, + "teacher_dice_labeled": [ + 0.038404221088852455, + 0.02077413830317999, + 0.2923980901071123, + 0.01126342508308318 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 264499, + "evidence_nonempty_patches": 79, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.04754171572408753, + "restart_teacher_dice": 0.2923980901071123, + "weights": [ + 0.006234923104471137, + 0.013530213405948727, + 0.9695787084892556, + 0.01065615500032451 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.29178575192159695, + "uniform": 0.03321569980814974, + "top1": 0.2923980901071123, + "top2": 0.2924863719344832, + "top3": 0.2921281235134678 + }, + "val_dice": 0.8029529345803188, + "val_vi": 0.07525777826253324, + "val_ece": 0.0010212449010216776 + }, + { + "round": 2, + "n_labeled": 200, + "selection_seconds": 55.9, + "round_seconds": 164.8, + "teacher_dice_labeled": [ + 0.046842362578081775, + 0.04780564953184647, + 0.1990651318389544, + 0.013492415424126526 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 642529, + "evidence_nonempty_patches": 175, + "first_correction": false, + "restart": "", + "weights": [ + 0.016280407867034646, + 0.04254894581878906, + 0.9150208260981681, + 0.026149820216008094 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.19741945730861185, + "uniform": 0.028403682100728907, + "top1": 0.1990651318389544, + "top2": 0.19797312729440208, + "top3": 0.1981771848716092 + }, + "pseudo_student_dice": 0.8770335682830072, + "pseudo_ensemble_dice": 0.12569697209148423, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8291435588647924, + "val_vi": 0.06179662626331432, + "val_ece": 0.0038847594329730973 + }, + { + "round": 3, + "n_labeled": 300, + "selection_seconds": 54.7, + "round_seconds": 311.8, + "teacher_dice_labeled": [ + 0.08255865604715446, + 0.02138812939535003, + 0.22783268924966504, + 0.004926919744107637 + ], + "teacher_dice_pool": [ + 0.11391794434011832, + 0.01325136837388421, + 0.30251711300878714, + 0.0018761484957749866 + ], + "evidence_fg_pixels": 1098795, + "evidence_nonempty_patches": 274, + "first_correction": false, + "restart": "", + "weights": [ + 0.01789980199801367, + 0.025132056464211197, + 0.9385036102974098, + 0.018464531240365296 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.2268272887304805, + "uniform": 0.018125503025893913, + "top1": 0.22783268924966504, + "top2": 0.2281070769553864, + "top3": 0.22737868569280834 + }, + "pseudo_student_dice": 0.8216393245823658, + "pseudo_ensemble_dice": 0.0836239049898745, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8310764803956745, + "val_vi": 0.057929555122978094, + "val_ece": 0.003957649912988966 + }, + { + "round": 4, + "n_labeled": 400, + "selection_seconds": 55.9, + "round_seconds": 366.7, + "teacher_dice_labeled": [ + 0.08048327496300407, + 0.022645864518324753, + 0.2296239813331129, + 0.0034469322707550393 + ], + "teacher_dice_pool": [ + 0.10958378623833324, + 0.010918176448017415, + 0.2883436901932895, + 0.001028443900657843 + ], + "evidence_fg_pixels": 1530549, + "evidence_nonempty_patches": 373, + "first_correction": false, + "restart": "", + "weights": [ + 0.0172451478010864, + 0.02503353958397879, + 0.9398258577311639, + 0.017895454883770926 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.2285886997434258, + "uniform": 0.018449128714215023, + "top1": 0.2296239813331129, + "top2": 0.22991655601049882, + "top3": 0.22914610374983752 + }, + "pseudo_student_dice": 0.8578352787612109, + "pseudo_ensemble_dice": 0.2134634851771511, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.6930196957330317, + "val_vi": 0.09496863254914646, + "val_ece": 0.010334223900637703 + }, + { + "round": 5, + "n_labeled": 500, + "selection_seconds": 54.7, + "round_seconds": 389.0, + "teacher_dice_labeled": [ + 0.07254406541407415, + 0.019096794061193467, + 0.2503883663051782, + 0.002487304098069243 + ], + "teacher_dice_pool": [ + 0.08826533136457557, + 0.00872585261709801, + 0.30841154064631243, + 0.00028425006196710506 + ], + "evidence_fg_pixels": 2059997, + "evidence_nonempty_patches": 473, + "first_correction": false, + "restart": "", + "weights": [ + 0.013115347543832544, + 0.019893045448484043, + 0.9523978348534398, + 0.014593772154243501 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.24956061427908754, + "uniform": 0.016428420249561627, + "top1": 0.2503883663051782, + "top2": 0.250631761884242, + "top3": 0.250038841633492 + }, + "pseudo_student_dice": 0.6706546731372539, + "pseudo_ensemble_dice": 0.2487381368920132, + "pseudo_source": "student", + "consistency_scale": 0.569, + "val_dice": 0.7672179735080855, + "val_vi": 0.07754526117524091, + "val_ece": 0.008222561514588736 + }, + { + "round": 6, + "n_labeled": 600, + "selection_seconds": 54.3, + "round_seconds": 435.5, + "teacher_dice_labeled": [ + 0.0741110863552717, + 0.02149826324908519, + 0.25555708480767686, + 0.002904886068703804 + ], + "teacher_dice_pool": [ + 0.09321674429540634, + 0.01355926016091147, + 0.3053232753965797, + 0.0013553642420735324 + ], + "evidence_fg_pixels": 2082676, + "evidence_nonempty_patches": 489, + "first_correction": false, + "restart": "", + "weights": [ + 0.012672719673732857, + 0.019382744109807604, + 0.9540044443362244, + 0.013940091880235197 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.2547049826942088, + "uniform": 0.01605497272283999, + "top1": 0.25555708480767686, + "top2": 0.2557923110518476, + "top3": 0.25517447573533975 + }, + "pseudo_student_dice": 0.8952162120359283, + "pseudo_ensemble_dice": 0.25597677767595495, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8360289452175707, + "val_vi": 0.0572751561654105, + "val_ece": 0.002901497062769228 + }, + { + "round": 7, + "n_labeled": 600, + "selection_seconds": 0.0, + "round_seconds": 423.1, + "final_consolidation": true, + "val_dice": 0.8611782530184082, + "val_vi": 0.04995436072194648, + "val_ece": 0.002535585236144839 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.36780975100244245, + "auroc_top20_error": 0.3513228515625 + }, + "al_seconds": 2555.4, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.012672719673732857, + 0.019382744109807604, + 0.9540044443362244, + 0.013940091880235197 + ], + "weight_stats_final": { + "max_weight": 0.9540044443362244, + "entropy": 0.23627863966260976, + "effective_sources": 1.2665271657069612, + "min_weight": 0.012672719673732857 + }, + "top_source_final": "drosophila-vnc", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8602503933131584, + "patch_iou": 0.825430895251587, + "patch_precision": 0.9672073859368605, + "patch_recall": 0.8463672148694931, + "patch_vi": 0.03058341851820067, + "n_patches": 2376, + "cal_ece": 0.001038001270790534, + "cal_nll": 0.016237718009692405, + "cal_brier": 0.0031186559088795794, + "cal_mean_confidence": 0.9971927973034305, + "cal_accuracy": 0.9962474681712963, + "cal_predicted_prevalence": 0.024482701362584174, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.02324717608279715 + }, + "test_notta": { + "patch_dice": 0.8602503933131584, + "patch_iou": 0.825430895251587, + "patch_precision": 0.9672073859368605, + "patch_recall": 0.8463672148694931, + "patch_vi": 0.03058341851820067, + "n_patches": 2376, + "cal_ece": 0.001038001270790534, + "cal_nll": 0.016237718009692405, + "cal_brier": 0.0031186559088795794, + "cal_mean_confidence": 0.9971927973034305, + "cal_accuracy": 0.9962474681712963, + "cal_predicted_prevalence": 0.024482701362584174, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.02324717608279715 + }, + "test": { + "patch_dice": 0.8625932890893546, + "patch_iou": 0.829279225277325, + "patch_precision": 0.9729915517125354, + "patch_recall": 0.845281203254042, + "patch_vi": 0.0299755587179031, + "n_patches": 2376, + "image_dice": 0.9319722259491418, + "image_iou": 0.8731114823384939, + "image_precision": 0.9771762362818, + "image_recall": 0.891326389697674, + "image_vi": 0.035926113257777614, + "n_images": 24, + "cal_ece": 0.0009590394516339326, + "cal_nll": 0.01394730970038897, + "cal_brier": 0.00297940856574426, + "cal_mean_confidence": 0.9968601800667, + "cal_accuracy": 0.9963304924242424, + "cal_predicted_prevalence": 0.024481879340277776, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.023225710490421464 + }, + "eval_seconds": 72.5, + "total_seconds": 3053.3, + "resources": { + "phases": { + "setup": { + "seconds": 0.58, + "peak_cpu_rss_mb": 1028.8, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 15.31, + "peak_cpu_rss_mb": 1793.7, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 405.7, + "peak_cpu_rss_mb": 3686.1, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 2555.43, + "peak_cpu_rss_mb": 3686.1, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4170.0, + "rounds": 6 + }, + "inference": { + "seconds": 72.52, + "peak_cpu_rss_mb": 3686.1, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4172.0, + "patches": 2376, + "images": 24, + "patches_per_second": 32.8, + "seconds_per_image": 3.022 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3686.1, + "total_seconds": 3049.5 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/kasthuri-pp/macs-v11-norestart-acq-laplace_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..dcd1aa9cf81818ac6fe6659dcf30b92d29360767 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-laplace/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,100,80.4,440.0,"[0.038404221088852455, 0.02077413830317999, 0.2923980901071123, 0.01126342508308318]",,264499,79,True,kept,0.04754171572408753,0.2923980901071123,"[0.006234923104471137, 0.013530213405948727, 0.9695787084892556, 0.01065615500032451]",mixture,"{'mixture': 0.29178575192159695, 'uniform': 0.03321569980814974, 'top1': 0.2923980901071123, 'top2': 0.2924863719344832, 'top3': 0.2921281235134678}",0.8029529345803188,0.07525777826253324,0.0010212449010216776,,,,, +2,200,55.9,164.8,"[0.046842362578081775, 0.04780564953184647, 0.1990651318389544, 0.013492415424126526]",,642529,175,False,,,,"[0.016280407867034646, 0.04254894581878906, 0.9150208260981681, 0.026149820216008094]",mixture,"{'mixture': 0.19741945730861185, 'uniform': 0.028403682100728907, 'top1': 0.1990651318389544, 'top2': 0.19797312729440208, 'top3': 0.1981771848716092}",0.8291435588647924,0.06179662626331432,0.0038847594329730973,0.8770335682830072,0.12569697209148423,student,1.0, +3,300,54.7,311.8,"[0.08255865604715446, 0.02138812939535003, 0.22783268924966504, 0.004926919744107637]","[0.11391794434011832, 0.01325136837388421, 0.30251711300878714, 0.0018761484957749866]",1098795,274,False,,,,"[0.01789980199801367, 0.025132056464211197, 0.9385036102974098, 0.018464531240365296]",mixture,"{'mixture': 0.2268272887304805, 'uniform': 0.018125503025893913, 'top1': 0.22783268924966504, 'top2': 0.2281070769553864, 'top3': 0.22737868569280834}",0.8310764803956745,0.057929555122978094,0.003957649912988966,0.8216393245823658,0.0836239049898745,student,1.0, +4,400,55.9,366.7,"[0.08048327496300407, 0.022645864518324753, 0.2296239813331129, 0.0034469322707550393]","[0.10958378623833324, 0.010918176448017415, 0.2883436901932895, 0.001028443900657843]",1530549,373,False,,,,"[0.0172451478010864, 0.02503353958397879, 0.9398258577311639, 0.017895454883770926]",mixture,"{'mixture': 0.2285886997434258, 'uniform': 0.018449128714215023, 'top1': 0.2296239813331129, 'top2': 0.22991655601049882, 'top3': 0.22914610374983752}",0.6930196957330317,0.09496863254914646,0.010334223900637703,0.8578352787612109,0.2134634851771511,student,1.0, +5,500,54.7,389.0,"[0.07254406541407415, 0.019096794061193467, 0.2503883663051782, 0.002487304098069243]","[0.08826533136457557, 0.00872585261709801, 0.30841154064631243, 0.00028425006196710506]",2059997,473,False,,,,"[0.013115347543832544, 0.019893045448484043, 0.9523978348534398, 0.014593772154243501]",mixture,"{'mixture': 0.24956061427908754, 'uniform': 0.016428420249561627, 'top1': 0.2503883663051782, 'top2': 0.250631761884242, 'top3': 0.250038841633492}",0.7672179735080855,0.07754526117524091,0.008222561514588736,0.6706546731372539,0.2487381368920132,student,0.569, +6,600,54.3,435.5,"[0.0741110863552717, 0.02149826324908519, 0.25555708480767686, 0.002904886068703804]","[0.09321674429540634, 0.01355926016091147, 0.3053232753965797, 0.0013553642420735324]",2082676,489,False,,,,"[0.012672719673732857, 0.019382744109807604, 0.9540044443362244, 0.013940091880235197]",mixture,"{'mixture': 0.2547049826942088, 'uniform': 0.01605497272283999, 'top1': 0.25555708480767686, 'top2': 0.2557923110518476, 'top3': 0.25517447573533975}",0.8360289452175707,0.0572751561654105,0.002901497062769228,0.8952162120359283,0.25597677767595495,student,1.0, +7,600,0.0,423.1,,,,,,,,,,,,0.8611782530184082,0.04995436072194648,0.002535585236144839,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/log.txt b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..fb30496b249036787c759e21a492033e66e9207a --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/log.txt @@ -0,0 +1,103 @@ +[04:28:08][ 0s][START] target=kasthuri-pp variant=macs-v11-norestart-acq-random budget=5.0% seed=1 | {"task": "mitochondria", "target": "kasthuri-pp", "variant": "macs-v11-norestart-acq-random", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "random", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 0, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[04:28:08][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[04:28:08][ 0s][DATA] kasthuri-pp: train=12023 val=2376 test=2376 patches +[04:28:09][ 1s][CALIB] source-val temperatures: fly-mito=1.38, lucchi=3.80, dros-vnc=2.77, multiclass=2.60 +[04:28:09][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[04:28:10][ 2s][WEIGHTS] mixture: fly-mito=0.055, lucchi=0.143, dros-vnc=0.678, multiclass=0.124 | eff.sources=2.61 | MMD2 mix=0.1257 uniform=0.2386 +[04:28:10][ 2s][WEIGHTS] per-source: fly-mito=0.004, lucchi=0.028, dros-vnc=0.948, multiclass=0.020 | min-MMD source (NeuroADDA rule) = dros-vnc +[04:28:23][ 15s][TEACHERS] target-val Dice: fly-mito=0.194, lucchi=0.137, dros-vnc=0.349, multiclass=0.152 +[04:28:23][ 15s][TEACHERS] ensemble target-val Dice: mixture=0.341, mixture@raw=0.346, mixture@logit=0.321, per_source=0.348, per_source@raw=0.348, per_source@logit=0.346, uniform=0.188, uniform@raw=0.183, uniform@logit=0.207 +[04:28:23][ 15s][WEIGHTS] using mixture: fly-mito=0.055, lucchi=0.143, dros-vnc=0.678, multiclass=0.124 +[04:28:23][ 15s][INIT] student initialised from teacher dros-vnc (weight 0.678) +[04:28:25][ 18s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[04:28:25][ 18s][KD] distilling the mixture ensemble (prob space) into the student +[04:29:00][ 52s][KD] soft labels for 6000 patches from 4 teachers in 34s +[04:29:54][ 106s][KD] epoch 1/5 | patches=6000 | loss=0.1770 | 54s +[04:30:49][ 161s][KD] epoch 2/5 | patches=6000 | loss=0.1512 | 55s +[04:31:42][ 215s][KD] epoch 3/5 | patches=6000 | loss=0.1468 | 54s +[04:32:37][ 269s][KD] epoch 4/5 | patches=6000 | loss=0.1451 | 54s +[04:33:31][ 323s][KD] epoch 5/5 | patches=6000 | loss=0.1438 | 54s +[04:33:33][ 325s][ROUND0] before any target label: val_dice=0.2346 val_vi=0.1498 +[04:33:33][ 325s][AL] patches=12023 (pool 10327 after dropping blank; free blank negatives: True) | budget=601 (5.0%) | rounds=6 | per_round=100 | acquisition=random | first_round=stratified +[04:33:44][ 337s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.01973645136544973, 'auroc_top20_error': 0.5132556640625} +[04:33:56][ 349s][AL] round 1: stratified cold start: 50 patches from the membrane-rich half (ensemble prevalence >= 0.034) + 50 from the membrane-poor half of 2000 candidates +[04:34:02][ 354s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.211, uniform=0.004, top1=0.213, top2=0.213, top3=0.212 -> using mixture +[04:34:03][ 355s][WEIGHTS] round 1: student kept (labeled Dice 0.049 vs best teacher dros-vnc 0.213); distilling the corrected ensemble into it +[04:34:03][ 355s][WEIGHTS] round 1: teacher Dice on 100 labeled patches + 100 blank (273263 fg px): fly-mito=0.037, lucchi=0.008, dros-vnc=0.213, multiclass=0.002 | weights: fly-mito=0.013, lucchi=0.026, dros-vnc=0.940, multiclass=0.021 +[04:34:37][ 390s][KD] soft labels for 6000 patches from 4 teachers in 34s +[04:35:32][ 444s][KD] epoch 1/2 | patches=6000 | loss=0.0596 | 55s +[04:36:26][ 499s][KD] epoch 2/2 | patches=6000 | loss=0.0586 | 55s +[04:36:38][ 511s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 12s +[04:36:42][ 515s][FT] round 1/6 epoch 1/5 | patches=200 | loss=0.8875 +[04:36:46][ 518s][FT] round 1/6 epoch 2/5 | patches=200 | loss=0.5811 +[04:36:50][ 522s][FT] round 1/6 epoch 3/5 | patches=200 | loss=0.4305 +[04:36:54][ 526s][FT] round 1/6 epoch 4/5 | patches=200 | loss=0.4578 +[04:36:58][ 530s][FT] round 1/6 epoch 5/5 | patches=200 | loss=0.4209 +[04:37:00][ 532s][AL] round 1/6 | labeled=100 | selection_seconds=23.7000 round_seconds=205.0000 restart_student_dice=0.0487 restart_teacher_dice=0.2126 val_dice=0.7216 val_vi=0.0951 val_ece=0.0058 +[04:37:09][ 542s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.263, uniform=0.009, top1=0.264, top2=0.264, top3=0.263 -> using mixture +[04:37:09][ 542s][WEIGHTS] round 2: teacher Dice on 200 labeled patches + 200 blank (539570 fg px): fly-mito=0.038, lucchi=0.009, dros-vnc=0.264, multiclass=0.002 | weights: fly-mito=0.008, lucchi=0.016, dros-vnc=0.963, multiclass=0.013 +[04:37:18][ 550s][CONS] round 2: on the 100 new patches student Dice=0.805 vs ensemble 0.311 -> pseudo-labels from the student | consistency x1.00 +[04:37:25][ 558s][FT] round 2/6 epoch 1/5 | patches=400 | loss=0.3257 +[04:37:32][ 565s][FT] round 2/6 epoch 2/5 | patches=400 | loss=0.2668 +[04:37:40][ 572s][FT] round 2/6 epoch 3/5 | patches=400 | loss=0.2630 +[04:37:47][ 579s][FT] round 2/6 epoch 4/5 | patches=400 | loss=0.2804 +[04:37:54][ 587s][FT] round 2/6 epoch 5/5 | patches=400 | loss=0.2662 +[04:37:56][ 589s][AL] round 2/6 | labeled=200 | selection_seconds=0.2000 round_seconds=54.7000 pseudo_student_dice=0.8046 pseudo_ensemble_dice=0.3111 consistency_scale=1.0000 val_dice=0.7895 val_vi=0.0714 val_ece=0.0063 +[04:38:28][ 621s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.105, lucchi=0.009, dros-vnc=0.304, multiclass=0.000 | blended with labeled Dice at 0.50 +[04:38:57][ 650s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.284, uniform=0.014, top1=0.284, top2=0.284, top3=0.284 -> using mixture +[04:38:57][ 650s][WEIGHTS] round 3: teacher Dice on 300 labeled patches + 300 blank (873679 fg px): fly-mito=0.073, lucchi=0.011, dros-vnc=0.284, multiclass=0.001 | weights: fly-mito=0.010, lucchi=0.013, dros-vnc=0.967, multiclass=0.010 +[04:39:06][ 658s][CONS] round 3: on the 100 new patches student Dice=0.850 vs ensemble 0.264 -> pseudo-labels from the student | consistency x1.00 +[04:39:16][ 669s][FT] round 3/6 epoch 1/5 | patches=600 | loss=0.2241 +[04:39:27][ 679s][FT] round 3/6 epoch 2/5 | patches=600 | loss=0.2142 +[04:39:38][ 690s][FT] round 3/6 epoch 3/5 | patches=600 | loss=0.2152 +[04:39:48][ 701s][FT] round 3/6 epoch 4/5 | patches=600 | loss=0.1816 +[04:39:59][ 712s][FT] round 3/6 epoch 5/5 | patches=600 | loss=0.1626 +[04:40:01][ 714s][AL] round 3/6 | labeled=300 | selection_seconds=0.2000 round_seconds=122.9000 pseudo_student_dice=0.8501 pseudo_ensemble_dice=0.2644 consistency_scale=1.0000 val_dice=0.7672 val_vi=0.0758 val_ece=0.0087 +[04:40:35][ 748s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.100, lucchi=0.011, dros-vnc=0.319, multiclass=0.001 | blended with labeled Dice at 0.50 +[04:41:06][ 779s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.301, uniform=0.014, top1=0.301, top2=0.301, top3=0.301 -> using mixture +[04:41:06][ 779s][WEIGHTS] round 4: teacher Dice on 400 labeled patches + 400 blank (1105137 fg px): fly-mito=0.070, lucchi=0.012, dros-vnc=0.301, multiclass=0.001 | weights: fly-mito=0.008, lucchi=0.011, dros-vnc=0.972, multiclass=0.009 +[04:41:15][ 787s][CONS] round 4: on the 100 new patches student Dice=0.889 vs ensemble 0.352 -> pseudo-labels from the student | consistency x1.00 +[04:41:29][ 802s][FT] round 4/6 epoch 1/5 | patches=800 | loss=0.1736 +[04:41:43][ 816s][FT] round 4/6 epoch 2/5 | patches=800 | loss=0.1973 +[04:41:57][ 830s][FT] round 4/6 epoch 3/5 | patches=800 | loss=0.1819 +[04:42:11][ 844s][FT] round 4/6 epoch 4/5 | patches=800 | loss=0.1630 +[04:42:25][ 858s][FT] round 4/6 epoch 5/5 | patches=800 | loss=0.1492 +[04:42:27][ 860s][AL] round 4/6 | labeled=400 | selection_seconds=0.2000 round_seconds=144.1000 pseudo_student_dice=0.8893 pseudo_ensemble_dice=0.3524 consistency_scale=1.0000 val_dice=0.8221 val_vi=0.0658 val_ece=0.0055 +[04:43:04][ 896s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.107, lucchi=0.013, dros-vnc=0.277, multiclass=0.002 | blended with labeled Dice at 0.50 +[04:43:37][ 929s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.280, uniform=0.013, top1=0.281, top2=0.281, top3=0.280 -> using mixture +[04:43:37][ 929s][WEIGHTS] round 5: teacher Dice on 500 labeled patches + 500 blank (1380594 fg px): fly-mito=0.074, lucchi=0.013, dros-vnc=0.281, multiclass=0.001 | weights: fly-mito=0.010, lucchi=0.014, dros-vnc=0.965, multiclass=0.011 +[04:43:45][ 938s][CONS] round 5: on the 100 new patches student Dice=0.874 vs ensemble 0.288 -> pseudo-labels from the student | consistency x1.00 +[04:44:03][ 955s][FT] round 5/6 epoch 1/5 | patches=1000 | loss=0.1683 +[04:44:20][ 972s][FT] round 5/6 epoch 2/5 | patches=1000 | loss=0.1588 +[04:44:38][ 990s][FT] round 5/6 epoch 3/5 | patches=1000 | loss=0.1714 +[04:44:55][ 1007s][FT] round 5/6 epoch 4/5 | patches=1000 | loss=0.1618 +[04:45:13][ 1025s][FT] round 5/6 epoch 5/5 | patches=1000 | loss=0.1541 +[04:45:15][ 1027s][AL] round 5/6 | labeled=500 | selection_seconds=0.2000 round_seconds=165.4000 pseudo_student_dice=0.8735 pseudo_ensemble_dice=0.2876 consistency_scale=1.0000 val_dice=0.8266 val_vi=0.0635 val_ece=0.0059 +[04:45:52][ 1064s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.109, lucchi=0.011, dros-vnc=0.291, multiclass=0.001 | blended with labeled Dice at 0.50 +[04:46:25][ 1097s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.283, uniform=0.012, top1=0.283, top2=0.283, top3=0.283 -> using mixture +[04:46:25][ 1097s][WEIGHTS] round 6: teacher Dice on 512 labeled patches + 512 blank (1368974 fg px): fly-mito=0.074, lucchi=0.013, dros-vnc=0.283, multiclass=0.001 | weights: fly-mito=0.010, lucchi=0.014, dros-vnc=0.966, multiclass=0.011 +[04:46:33][ 1106s][CONS] round 6: on the 100 new patches student Dice=0.887 vs ensemble 0.213 -> pseudo-labels from the student | consistency x1.00 +[04:46:54][ 1127s][FT] round 6/6 epoch 1/5 | patches=1200 | loss=0.1421 +[04:47:15][ 1148s][FT] round 6/6 epoch 2/5 | patches=1200 | loss=0.1477 +[04:47:36][ 1168s][FT] round 6/6 epoch 3/5 | patches=1200 | loss=0.1464 +[04:47:57][ 1189s][FT] round 6/6 epoch 4/5 | patches=1200 | loss=0.1336 +[04:48:18][ 1210s][FT] round 6/6 epoch 5/5 | patches=1200 | loss=0.1181 +[04:48:20][ 1212s][AL] round 6/6 | labeled=600 | selection_seconds=0.2000 round_seconds=183.0000 pseudo_student_dice=0.8867 pseudo_ensemble_dice=0.2133 consistency_scale=1.0000 val_dice=0.8770 val_vi=0.0473 val_ece=0.0026 +[04:48:23][ 1215s][CONS] final consolidation: pseudo-labels refreshed from the student +[04:48:44][ 1236s][FT] final 1/10 epoch 1/1 | patches=1200 | loss=0.1167 +[04:49:05][ 1257s][FT] final 2/10 epoch 1/1 | patches=1200 | loss=0.1251 +[04:49:25][ 1278s][FT] final 3/10 epoch 1/1 | patches=1200 | loss=0.1314 +[04:49:46][ 1299s][FT] final 4/10 epoch 1/1 | patches=1200 | loss=0.1191 +[04:50:07][ 1320s][FT] final 5/10 epoch 1/1 | patches=1200 | loss=0.1281 +[04:50:28][ 1340s][FT] final 6/10 epoch 1/1 | patches=1200 | loss=0.1175 +[04:50:49][ 1361s][FT] final 7/10 epoch 1/1 | patches=1200 | loss=0.1058 +[04:51:10][ 1382s][FT] final 8/10 epoch 1/1 | patches=1200 | loss=0.1030 +[04:51:31][ 1403s][FT] final 9/10 epoch 1/1 | patches=1200 | loss=0.0958 +[04:51:51][ 1424s][FT] final 10/10 epoch 1/1 | patches=1200 | loss=0.0961 +[04:51:54][ 1426s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=211.9000 val_dice=0.8812 val_vi=0.0453 val_ece=0.0028 +[04:51:54][ 1426s][WEIGHTS] final: fly-mito=0.010, lucchi=0.014, dros-vnc=0.966, multiclass=0.011 +[04:52:26][ 1458s][TEST] raw student (no EMA, no TTA): patch_dice=0.8677 patch_vi=0.0291 +[04:52:30][ 1462s][TEST] final model without TTA: patch_dice=0.8677 patch_vi=0.0291 +[04:52:30][ 1462s][RESOURCES] {"phases": {"setup": {"seconds": 0.62, "peak_cpu_rss_mb": 1027.9, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 16.68, "peak_cpu_rss_mb": 1793.0, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 305.15, "peak_cpu_rss_mb": 3684.8, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 1101.03, "peak_cpu_rss_mb": 3684.8, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4170.0, "rounds": 6}, "inference": {"seconds": 36.34, "peak_cpu_rss_mb": 3684.8, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4172.0, "patches": 2376, "images": 24, "patches_per_second": 65.4, "seconds_per_image": 1.514}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3684.8, "total_seconds": 1459.8} +[04:52:30][ 1462s][TEST] patch_dice=0.8740 patch_iou=0.8457 patch_precision=0.9589 patch_recall=0.8762 patch_vi=0.0283 image_dice=0.9395 image_iou=0.8866 image_precision=0.9532 image_recall=0.9270 image_vi=0.0348 cal_ece=0.0017 cal_nll=0.0146 cal_brier=0.0029 cal_mean_confidence=0.9981 cal_accuracy=0.9964 cal_predicted_prevalence=0.0262 cal_true_prevalence=0.0268 cal_mean_probability=0.0257 +[04:52:30][ 1462s][DONE] kasthuri-pp macs-v11-norestart-acq-random b5 s1 | test patch Dice=0.8740 VI=0.0283 | image Dice=0.9395 VI=0.0348 | 1462s diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/result.json b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/result.json new file mode 100644 index 0000000000000000000000000000000000000000..cf323b4b3b603502e3193592bf909817309a857d --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/result.json @@ -0,0 +1,667 @@ +{ + "target": "kasthuri-pp", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 1, + "config": { + "task": "mitochondria", + "target": "kasthuri-pp", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 1, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "random", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 0, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "kasthuri-pp", + "--variants", + "macs-v11-norestart-acq-random", + "--budgets", + "5", + "--seeds", + "1", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "lucchi-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.0552621669329701, + 0.14304343546870246, + 0.6777952504109969, + 0.12389914718733061 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "lucchi-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.0552621669329701, + 0.14304343546870246, + 0.6777952504109969, + 0.12389914718733061 + ], + "per_source": [ + 0.00394004854372213, + 0.027549943457875075, + 0.9484681548320345, + 0.02004185316636836 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.1161079474391782, + "mmd2_mixture": 0.12571317215793731, + "mmd2_uniform": 0.23862471954971437, + "mmd2_single_source": [ + 0.7126139881719412, + 0.4366429148044758, + 0.1798181763334552, + 0.4482686385790514 + ], + "median_squared_distance": 0.1734670188179156, + "gammas": [ + 1.4411961518887884, + 2.8823923037775767, + 5.764784607555153, + 11.529569215110307, + 23.059138430220614 + ], + "max_weight": 0.6777952504109969, + "entropy": 0.9605224718759118, + "effective_sources": 2.6130613679073416, + "min_weight": 0.0552621669329701 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6491308049576032, + 0.5568981487675557, + 0.3890679978127308, + 0.5719877547192425 + ], + "standardized": [ + 1.1318865389009094, + 0.1594828192631669, + -1.6099410463963124, + 0.31857168823223964 + ], + "tau": 0.5, + "max_weight": 0.9484681548320345, + "entropy": 0.24930992499892873, + "effective_sources": 1.2831396485053355, + "min_weight": 0.00394004854372213 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "drosophila-vnc", + "seconds": 1.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.19408169161460087, + 0.13693054106355262, + 0.34856214071670877, + 0.15175765833135202 + ], + "ensemble_val_dice": { + "mixture": 0.3408904698785591, + "mixture@raw": 0.3461558372224687, + "mixture@logit": 0.3209424494814286, + "per_source": 0.3475944498943022, + "per_source@raw": 0.34822194749797575, + "per_source@logit": 0.3464301210911742, + "uniform": 0.18819756340126403, + "uniform@raw": 0.18329087899026675, + "uniform@logit": 0.2066047847592749 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 3.7969236373901367, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "drosophila-vnc", + "weight_stats": { + "max_weight": 0.6777952504109969, + "entropy": 0.9605224718759118, + "effective_sources": 2.6130613679073416, + "min_weight": 0.0552621669329701 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.19999999999999998, + "spearman_weight_taxonomy": 0.2581988897471611, + "top_source": "drosophila-vnc", + "top_source_is_best_teacher": true + }, + "init_teacher": "drosophila-vnc", + "student_init": "top-source", + "blank_fraction": 0.14106296265491142, + "kd_seconds": 305.1, + "round0_val": { + "val_dice": 0.23463455252684254, + "val_vi": 0.14981963507387186, + "val_ece": 0.02902207695205794 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 100, + "selection_seconds": 23.7, + "round_seconds": 205.0, + "teacher_dice_labeled": [ + 0.03735964808141487, + 0.007780861810074168, + 0.212574608937917, + 0.0024969741906641515 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 273263, + "evidence_nonempty_patches": 79, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.04866841141249433, + "restart_teacher_dice": 0.212574608937917, + "weights": [ + 0.013290741238442873, + 0.025593529845871647, + 0.9400884565379644, + 0.021027272377721215 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.21133112103917567, + "uniform": 0.004250103060439019, + "top1": 0.212574608937917, + "top2": 0.21283805568810535, + "top3": 0.21197631837906725 + }, + "val_dice": 0.7216407820685099, + "val_vi": 0.09507628822221623, + "val_ece": 0.005837026129384508 + }, + { + "round": 2, + "n_labeled": 200, + "selection_seconds": 0.2, + "round_seconds": 54.7, + "teacher_dice_labeled": [ + 0.038193294843739733, + 0.00857742737470785, + 0.2635635667069667, + 0.0016351710767849852 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 539570, + "evidence_nonempty_patches": 162, + "first_correction": false, + "restart": "", + "weights": [ + 0.008245380569876698, + 0.01587196323906002, + 0.9630569526271485, + 0.012825703563914716 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.26273982934992696, + "uniform": 0.008529724069786306, + "top1": 0.2635635667069667, + "top2": 0.26369897963188876, + "top3": 0.2631587214571706 + }, + "pseudo_student_dice": 0.8046019777449922, + "pseudo_ensemble_dice": 0.3110821821300579, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.7895356584262208, + "val_vi": 0.07138506388659246, + "val_ece": 0.006280976401575323 + }, + { + "round": 3, + "n_labeled": 300, + "selection_seconds": 0.2, + "round_seconds": 122.9, + "teacher_dice_labeled": [ + 0.07335595767462577, + 0.010637602062809264, + 0.28429099027221527, + 0.0008178643731797476 + ], + "teacher_dice_pool": [ + 0.10454404004270158, + 0.009155936872986034, + 0.3044675375919843, + 0.0002790818589145627 + ], + "evidence_fg_pixels": 873679, + "evidence_nonempty_patches": 248, + "first_correction": false, + "restart": "", + "weights": [ + 0.009563189930162986, + 0.013220882492319393, + 0.9668355236967202, + 0.01038040388079742 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.2836514433196089, + "uniform": 0.013774333209289592, + "top1": 0.28429099027221527, + "top2": 0.2844644732772197, + "top3": 0.28402392913164665 + }, + "pseudo_student_dice": 0.8501465400865926, + "pseudo_ensemble_dice": 0.26442834918772795, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.7672489471848805, + "val_vi": 0.07582779875337967, + "val_ece": 0.008725395963820633 + }, + { + "round": 4, + "n_labeled": 400, + "selection_seconds": 0.2, + "round_seconds": 144.1, + "teacher_dice_labeled": [ + 0.06999789620282618, + 0.011878293696383484, + 0.30113624146757817, + 0.0011106206396201228 + ], + "teacher_dice_pool": [ + 0.10008671078894942, + 0.01102401984706239, + 0.31867381830243924, + 0.0008765013981614792 + ], + "evidence_fg_pixels": 1105137, + "evidence_nonempty_patches": 326, + "first_correction": false, + "restart": "", + "weights": [ + 0.007854959550464124, + 0.011370347766777403, + 0.9719314667027957, + 0.00884322597996266 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.30059273212246784, + "uniform": 0.013980460464784912, + "top1": 0.30113624146757817, + "top2": 0.30127431069900223, + "top3": 0.30089605041109635 + }, + "pseudo_student_dice": 0.8893358305310802, + "pseudo_ensemble_dice": 0.35243771176615135, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8221327410162531, + "val_vi": 0.06578217646954726, + "val_ece": 0.005516991115877433 + }, + { + "round": 5, + "n_labeled": 500, + "selection_seconds": 0.2, + "round_seconds": 165.4, + "teacher_dice_labeled": [ + 0.07366598787362917, + 0.013108144050432476, + 0.2806613225815625, + 0.001440072005102871 + ], + "teacher_dice_pool": [ + 0.10724873065315944, + 0.01294148097130922, + 0.27677553327568677, + 0.0015694659656771235 + ], + "evidence_fg_pixels": 1380594, + "evidence_nonempty_patches": 403, + "first_correction": false, + "restart": "", + "weights": [ + 0.009930918226728373, + 0.014029107831741373, + 0.9652267139452294, + 0.010813259996300818 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.27995127956233884, + "uniform": 0.013045028327123214, + "top1": 0.2806613225815625, + "top2": 0.28085336508089176, + "top3": 0.28035509293999183 + }, + "pseudo_student_dice": 0.8735029763078015, + "pseudo_ensemble_dice": 0.28763235980780527, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8266331286032274, + "val_vi": 0.06350459301537013, + "val_ece": 0.005871369330097335 + }, + { + "round": 6, + "n_labeled": 600, + "selection_seconds": 0.2, + "round_seconds": 183.0, + "teacher_dice_labeled": [ + 0.07399190372900116, + 0.012717102991331934, + 0.2831522419474163, + 0.0011427493742722891 + ], + "teacher_dice_pool": [ + 0.10929364091674894, + 0.010875862291462184, + 0.29142294701513916, + 0.0007928564817581522 + ], + "evidence_fg_pixels": 1368974, + "evidence_nonempty_patches": 405, + "first_correction": false, + "restart": "", + "weights": [ + 0.009727061960847284, + 0.013642960618649553, + 0.9661044896371229, + 0.010525487783380399 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.28250846960702325, + "uniform": 0.011679061095255239, + "top1": 0.2831522419474163, + "top2": 0.2833506486693861, + "top3": 0.28288299758426366 + }, + "pseudo_student_dice": 0.8866523235027264, + "pseudo_ensemble_dice": 0.2132830104823841, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8770394082761853, + "val_vi": 0.04733762687921725, + "val_ece": 0.0025933877356800316 + }, + { + "round": 7, + "n_labeled": 600, + "selection_seconds": 0.0, + "round_seconds": 211.9, + "final_consolidation": true, + "val_dice": 0.8812399369231715, + "val_vi": 0.04526571929124888, + "val_ece": 0.002832710175876692 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.01973645136544973, + "auroc_top20_error": 0.5132556640625 + }, + "al_seconds": 1101.0, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.009727061960847284, + 0.013642960618649553, + 0.9661044896371229, + 0.010525487783380399 + ], + "weight_stats_final": { + "max_weight": 0.9661044896371229, + "entropy": 0.18490113847132433, + "effective_sources": 1.2030994939930688, + "min_weight": 0.009727061960847284 + }, + "top_source_final": "drosophila-vnc", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8677192541266396, + "patch_iou": 0.8381120628455865, + "patch_precision": 0.9502317100834548, + "patch_recall": 0.8759811111502647, + "patch_vi": 0.029050105178595263, + "n_patches": 2376, + "cal_ece": 0.00206566840244836, + "cal_nll": 0.017401240515757976, + "cal_brier": 0.003046062804347499, + "cal_mean_confidence": 0.9984195884624231, + "cal_accuracy": 0.9963539200599747, + "cal_predicted_prevalence": 0.02619291877104377, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.025654720743633477 + }, + "test_notta": { + "patch_dice": 0.8677192541266396, + "patch_iou": 0.8381120628455865, + "patch_precision": 0.9502317100834548, + "patch_recall": 0.8759811111502647, + "patch_vi": 0.029050105178595263, + "n_patches": 2376, + "cal_ece": 0.00206566840244836, + "cal_nll": 0.017401240515757976, + "cal_brier": 0.003046062804347499, + "cal_mean_confidence": 0.9984195884624231, + "cal_accuracy": 0.9963539200599747, + "cal_predicted_prevalence": 0.02619291877104377, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.025654720743633477 + }, + "test": { + "patch_dice": 0.8739638226818567, + "patch_iou": 0.8457292127082943, + "patch_precision": 0.958925008538769, + "patch_recall": 0.8761645322807718, + "patch_vi": 0.02834765761364493, + "n_patches": 2376, + "image_dice": 0.9395188434319831, + "image_iou": 0.8866158015181593, + "image_precision": 0.9532327765847892, + "image_recall": 0.926951017809151, + "image_vi": 0.03484235983963741, + "n_images": 24, + "cal_ece": 0.0017045598168420944, + "cal_nll": 0.014563070119785687, + "cal_brier": 0.002889060931314359, + "cal_mean_confidence": 0.9981106782932732, + "cal_accuracy": 0.996406118476431, + "cal_predicted_prevalence": 0.02622703269675926, + "cal_true_prevalence": 0.026826286958122895, + "cal_mean_probability": 0.025747317206493585 + }, + "eval_seconds": 36.3, + "total_seconds": 1462.4, + "resources": { + "phases": { + "setup": { + "seconds": 0.62, + "peak_cpu_rss_mb": 1027.9, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 16.68, + "peak_cpu_rss_mb": 1793.0, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 305.15, + "peak_cpu_rss_mb": 3684.8, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 1101.03, + "peak_cpu_rss_mb": 3684.8, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4170.0, + "rounds": 6 + }, + "inference": { + "seconds": 36.34, + "peak_cpu_rss_mb": 3684.8, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4172.0, + "patches": 2376, + "images": 24, + "patches_per_second": 65.4, + "seconds_per_image": 1.514 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3684.8, + "total_seconds": 1459.8 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/kasthuri-pp/macs-v11-norestart-acq-random_b5_s1.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/rounds.csv b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..1b2141bb3fb1df250a289d5502b039572dd421d4 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-random/budget-5/seed-1/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,100,23.7,205.0,"[0.03735964808141487, 0.007780861810074168, 0.212574608937917, 0.0024969741906641515]",,273263,79,True,kept,0.04866841141249433,0.212574608937917,"[0.013290741238442873, 0.025593529845871647, 0.9400884565379644, 0.021027272377721215]",mixture,"{'mixture': 0.21133112103917567, 'uniform': 0.004250103060439019, 'top1': 0.212574608937917, 'top2': 0.21283805568810535, 'top3': 0.21197631837906725}",0.7216407820685099,0.09507628822221623,0.005837026129384508,,,,, +2,200,0.2,54.7,"[0.038193294843739733, 0.00857742737470785, 0.2635635667069667, 0.0016351710767849852]",,539570,162,False,,,,"[0.008245380569876698, 0.01587196323906002, 0.9630569526271485, 0.012825703563914716]",mixture,"{'mixture': 0.26273982934992696, 'uniform': 0.008529724069786306, 'top1': 0.2635635667069667, 'top2': 0.26369897963188876, 'top3': 0.2631587214571706}",0.7895356584262208,0.07138506388659246,0.006280976401575323,0.8046019777449922,0.3110821821300579,student,1.0, +3,300,0.2,122.9,"[0.07335595767462577, 0.010637602062809264, 0.28429099027221527, 0.0008178643731797476]","[0.10454404004270158, 0.009155936872986034, 0.3044675375919843, 0.0002790818589145627]",873679,248,False,,,,"[0.009563189930162986, 0.013220882492319393, 0.9668355236967202, 0.01038040388079742]",mixture,"{'mixture': 0.2836514433196089, 'uniform': 0.013774333209289592, 'top1': 0.28429099027221527, 'top2': 0.2844644732772197, 'top3': 0.28402392913164665}",0.7672489471848805,0.07582779875337967,0.008725395963820633,0.8501465400865926,0.26442834918772795,student,1.0, +4,400,0.2,144.1,"[0.06999789620282618, 0.011878293696383484, 0.30113624146757817, 0.0011106206396201228]","[0.10008671078894942, 0.01102401984706239, 0.31867381830243924, 0.0008765013981614792]",1105137,326,False,,,,"[0.007854959550464124, 0.011370347766777403, 0.9719314667027957, 0.00884322597996266]",mixture,"{'mixture': 0.30059273212246784, 'uniform': 0.013980460464784912, 'top1': 0.30113624146757817, 'top2': 0.30127431069900223, 'top3': 0.30089605041109635}",0.8221327410162531,0.06578217646954726,0.005516991115877433,0.8893358305310802,0.35243771176615135,student,1.0, +5,500,0.2,165.4,"[0.07366598787362917, 0.013108144050432476, 0.2806613225815625, 0.001440072005102871]","[0.10724873065315944, 0.01294148097130922, 0.27677553327568677, 0.0015694659656771235]",1380594,403,False,,,,"[0.009930918226728373, 0.014029107831741373, 0.9652267139452294, 0.010813259996300818]",mixture,"{'mixture': 0.27995127956233884, 'uniform': 0.013045028327123214, 'top1': 0.2806613225815625, 'top2': 0.28085336508089176, 'top3': 0.28035509293999183}",0.8266331286032274,0.06350459301537013,0.005871369330097335,0.8735029763078015,0.28763235980780527,student,1.0, +6,600,0.2,183.0,"[0.07399190372900116, 0.012717102991331934, 0.2831522419474163, 0.0011427493742722891]","[0.10929364091674894, 0.010875862291462184, 0.29142294701513916, 0.0007928564817581522]",1368974,405,False,,,,"[0.009727061960847284, 0.013642960618649553, 0.9661044896371229, 0.010525487783380399]",mixture,"{'mixture': 0.28250846960702325, 'uniform': 0.011679061095255239, 'top1': 0.2831522419474163, 'top2': 0.2833506486693861, 'top3': 0.28288299758426366}",0.8770394082761853,0.04733762687921725,0.0025933877356800316,0.8866523235027264,0.2132830104823841,student,1.0, +7,600,0.0,211.9,,,,,,,,,,,,0.8812399369231715,0.04526571929124888,0.002832710175876692,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..e46b25d901cc1ed3f6b730200d960357d7e2c8e3 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt @@ -0,0 +1,107 @@ +[04:59:59][ 0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-margin budget=5.0% seed=1 | {"task": "mitochondria", "target": "lucchi-pp", "variant": "macs-v11-norestart-acq-margin", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "margin", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[04:59:59][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[04:59:59][ 0s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches +[05:00:00][ 1s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60 +[05:00:00][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[05:00:01][ 2s][WEIGHTS] mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 | eff.sources=1.44 | MMD2 mix=0.0050 uniform=0.2092 +[05:00:01][ 2s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass +[05:00:14][ 15s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833 +[05:00:14][ 15s][TEACHERS] ensemble target-val Dice: mixture=0.834, mixture@raw=0.833, mixture@logit=0.835, per_source=0.834, per_source@raw=0.833, per_source@logit=0.834, uniform=0.331, uniform@raw=0.184, uniform@logit=0.220 +[05:00:14][ 15s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 +[05:00:14][ 15s][INIT] student initialised from teacher multiclass (weight 0.920) +[05:00:15][ 16s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[05:00:15][ 16s][KD] distilling the mixture ensemble (prob space) into the student +[05:00:50][ 51s][KD] soft labels for 6000 patches from 4 teachers in 34s +[05:01:44][ 105s][KD] epoch 1/5 | patches=6000 | loss=0.1185 | 54s +[05:02:37][ 158s][KD] epoch 2/5 | patches=6000 | loss=0.1064 | 53s +[05:03:29][ 210s][KD] epoch 3/5 | patches=6000 | loss=0.1043 | 52s +[05:04:20][ 261s][KD] epoch 4/5 | patches=6000 | loss=0.1046 | 52s +[05:05:13][ 314s][KD] epoch 5/5 | patches=6000 | loss=0.1029 | 53s +[05:05:15][ 316s][ROUND0] before any target label: val_dice=0.8819 val_vi=0.0627 +[05:05:15][ 316s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=margin | first_round=stratified +[05:05:27][ 328s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.16758785175575056, 'auroc_top20_error': 0.5127568359375} +[05:05:39][ 340s][AL] round 1: stratified cold start: 34 patches from the membrane-rich half (ensemble prevalence >= 0.074) + 33 from the membrane-poor half of 2000 candidates +[05:05:41][ 342s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.932, uniform=0.354, top1=0.932, top2=0.932, top3=0.932 -> using mixture +[05:05:42][ 343s][WEIGHTS] round 1: student kept (labeled Dice 0.941 vs best teacher multiclass 0.932); distilling the corrected ensemble into it +[05:05:42][ 343s][WEIGHTS] round 1: teacher Dice on 67 labeled patches + 0 blank (352197 fg px): fly-mito=0.288, kasthuri=0.000, dros-vnc=0.023, multiclass=0.932 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[05:06:16][ 377s][KD] soft labels for 6000 patches from 4 teachers in 34s +[05:07:09][ 430s][KD] epoch 1/2 | patches=6000 | loss=0.0471 | 53s +[05:08:01][ 482s][KD] epoch 2/2 | patches=6000 | loss=0.0460 | 52s +[05:08:13][ 494s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 12s +[05:08:15][ 496s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3105 +[05:08:16][ 497s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3525 +[05:08:18][ 499s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.4123 +[05:08:19][ 500s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.2904 +[05:08:21][ 502s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.3520 +[05:08:23][ 504s][AL] round 1/6 | labeled=67 | selection_seconds=24.1000 round_seconds=185.7000 restart_student_dice=0.9407 restart_teacher_dice=0.9318 val_dice=0.7701 val_vi=0.1061 val_ece=0.0111 +[05:08:35][ 516s][AL] round 2: 1140/7983 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:08:38][ 519s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.883, uniform=0.267, top1=0.883, top2=0.883, top3=0.883 -> using mixture +[05:08:38][ 519s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (861194 fg px): fly-mito=0.333, kasthuri=0.001, dros-vnc=0.024, multiclass=0.883 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[05:08:40][ 521s][CONS] round 2: on the 67 new patches student Dice=0.747 vs ensemble 0.848 -> pseudo-labels from the ensemble | consistency x1.00 +[05:08:43][ 524s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.3178 +[05:08:46][ 527s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.2872 +[05:08:49][ 530s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.2655 +[05:08:51][ 532s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.2456 +[05:08:54][ 535s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.2381 +[05:08:56][ 537s][AL] round 2/6 | labeled=134 | selection_seconds=12.1000 round_seconds=31.1000 pseudo_student_dice=0.7475 pseudo_ensemble_dice=0.8477 consistency_scale=1.0000 val_dice=0.8196 val_vi=0.0802 val_ece=0.0064 +[05:09:08][ 549s][AL] round 3: 1304/7916 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:09:13][ 554s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.851, uniform=0.227, top1=0.851, top2=0.851, top3=0.851 -> using mixture +[05:09:13][ 554s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (1289614 fg px): fly-mito=0.330, kasthuri=0.001, dros-vnc=0.020, multiclass=0.851 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[05:09:18][ 559s][CONS] round 3: on the 67 new patches student Dice=0.829 vs ensemble 0.787 -> pseudo-labels from the student | consistency x1.00 +[05:09:22][ 563s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.2948 +[05:09:26][ 567s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.2708 +[05:09:30][ 571s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.2435 +[05:09:34][ 575s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.2886 +[05:09:37][ 578s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2382 +[05:09:39][ 580s][AL] round 3/6 | labeled=201 | selection_seconds=12.1000 round_seconds=41.7000 pseudo_student_dice=0.8288 pseudo_ensemble_dice=0.7868 consistency_scale=1.0000 val_dice=0.8619 val_vi=0.0665 val_ece=0.0038 +[05:09:51][ 592s][AL] round 4: 1348/7849 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:10:20][ 621s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.270, kasthuri=0.000, dros-vnc=0.027, multiclass=0.951 | blended with labeled Dice at 0.50 +[05:10:45][ 646s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.905, uniform=0.278, top1=0.905, top2=0.905, top3=0.905 -> using mixture +[05:10:45][ 646s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1720970 fg px): fly-mito=0.300, kasthuri=0.000, dros-vnc=0.021, multiclass=0.905 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[05:10:50][ 651s][CONS] round 4: on the 67 new patches student Dice=0.857 vs ensemble 0.885 -> pseudo-labels from the student | consistency x1.00 +[05:10:55][ 656s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.2343 +[05:10:59][ 660s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.2150 +[05:11:04][ 665s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.1907 +[05:11:09][ 670s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.1891 +[05:11:14][ 675s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.1930 +[05:11:16][ 677s][AL] round 4/6 | labeled=268 | selection_seconds=11.9000 round_seconds=94.9000 pseudo_student_dice=0.8569 pseudo_ensemble_dice=0.8845 consistency_scale=1.0000 val_dice=0.8439 val_vi=0.0731 val_ece=0.0055 +[05:11:28][ 689s][AL] round 5: 987/7782 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:11:57][ 718s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.287, kasthuri=0.000, dros-vnc=0.020, multiclass=0.943 | blended with labeled Dice at 0.50 +[05:12:23][ 744s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.899, uniform=0.260, top1=0.899, top2=0.899, top3=0.899 -> using mixture +[05:12:23][ 744s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (2036353 fg px): fly-mito=0.297, kasthuri=0.000, dros-vnc=0.017, multiclass=0.899 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[05:12:28][ 749s][CONS] round 5: on the 67 new patches student Dice=0.814 vs ensemble 0.830 -> pseudo-labels from the student | consistency x1.00 +[05:12:34][ 755s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2311 +[05:12:40][ 761s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.2398 +[05:12:46][ 767s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.2188 +[05:12:52][ 773s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.1788 +[05:12:58][ 779s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.2109 +[05:13:00][ 781s][AL] round 5/6 | labeled=335 | selection_seconds=11.8000 round_seconds=102.3000 pseudo_student_dice=0.8136 pseudo_ensemble_dice=0.8300 consistency_scale=1.0000 val_dice=0.8459 val_vi=0.0764 val_ece=0.0065 +[05:13:12][ 793s][AL] round 6: 1263/7715 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[05:13:42][ 823s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.284, kasthuri=0.001, dros-vnc=0.025, multiclass=0.954 | blended with labeled Dice at 0.50 +[05:14:08][ 849s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.904, uniform=0.269, top1=0.904, top2=0.904, top3=0.904 -> using mixture +[05:14:08][ 849s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (2426256 fg px): fly-mito=0.294, kasthuri=0.001, dros-vnc=0.019, multiclass=0.904 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[05:14:13][ 854s][CONS] round 6: on the 67 new patches student Dice=0.873 vs ensemble 0.850 -> pseudo-labels from the student | consistency x1.00 +[05:14:21][ 862s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2097 +[05:14:28][ 869s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.1937 +[05:14:35][ 876s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.1851 +[05:14:42][ 883s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.1730 +[05:14:50][ 891s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.2006 +[05:14:52][ 893s][AL] round 6/6 | labeled=402 | selection_seconds=11.7000 round_seconds=109.2000 pseudo_student_dice=0.8730 pseudo_ensemble_dice=0.8500 consistency_scale=1.0000 val_dice=0.8672 val_vi=0.0640 val_ece=0.0034 +[05:14:55][ 896s][CONS] final consolidation: pseudo-labels refreshed from the student +[05:15:02][ 903s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.1883 +[05:15:09][ 910s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.1490 +[05:15:17][ 918s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.1485 +[05:15:24][ 925s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.1486 +[05:15:31][ 932s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.1525 +[05:15:38][ 939s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1429 +[05:15:46][ 947s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1334 +[05:15:53][ 954s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1259 +[05:16:00][ 961s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1220 +[05:16:07][ 968s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1220 +[05:16:09][ 970s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=75.6000 val_dice=0.8793 val_vi=0.0605 val_ece=0.0043 +[05:16:09][ 970s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[05:16:33][ 994s][TEST] raw student (no EMA, no TTA): patch_dice=0.8950 patch_vi=0.0425 +[05:16:36][ 997s][TEST] final model without TTA: patch_dice=0.8950 patch_vi=0.0425 +[05:16:36][ 997s][RESOURCES] {"phases": {"setup": {"seconds": 0.58, "peak_cpu_rss_mb": 1027.9, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 15.5, "peak_cpu_rss_mb": 1801.2, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 297.83, "peak_cpu_rss_mb": 3685.2, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 654.04, "peak_cpu_rss_mb": 3685.2, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 26.92, "peak_cpu_rss_mb": 3685.2, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 65.0, "seconds_per_image": 0.538}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3685.2, "total_seconds": 994.9} +[05:16:36][ 997s][TEST] patch_dice=0.9142 patch_iou=0.8860 patch_precision=0.9416 patch_recall=0.9382 patch_vi=0.0403 image_dice=0.9556 image_iou=0.9155 image_precision=0.9633 image_recall=0.9487 image_vi=0.0459 cal_ece=0.0019 cal_nll=0.0168 cal_brier=0.0034 cal_mean_confidence=0.9976 cal_accuracy=0.9957 cal_predicted_prevalence=0.0445 cal_true_prevalence=0.0452 cal_mean_probability=0.0447 +[05:16:36][ 997s][DONE] lucchi-pp macs-v11-norestart-acq-margin b5 s1 | test patch Dice=0.9142 VI=0.0403 | image Dice=0.9556 VI=0.0459 | 997s diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json new file mode 100644 index 0000000000000000000000000000000000000000..12e68aa147657a8c237d230df3742bf6a92eff4f --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json @@ -0,0 +1,662 @@ +{ + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 1, + "config": { + "task": "mitochondria", + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 1, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "margin", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "lucchi-pp", + "--variants", + "macs-v11-norestart-acq-margin", + "--budgets", + "5", + "--seeds", + "1", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "per_source": [ + 0.01302141100636151, + 0.021594387588085767, + 0.004953657914530629, + 0.960430543491022 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.001413550071584186, + "mmd2_mixture": 0.005039751415091165, + "mmd2_uniform": 0.2092225964739336, + "mmd2_single_source": [ + 0.5680514029229977, + 0.4644134839747518, + 0.6002496450435827, + 0.004698779437425582 + ], + "median_squared_distance": 0.18927357450912807, + "gammas": [ + 1.3208394285803657, + 2.6416788571607315, + 5.283357714321463, + 10.566715428642926, + 21.133430857285852 + ], + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6675364398352843, + 0.5899254221485877, + 0.8158221669373606, + 0.007664827003586616 + ], + "standardized": [ + 0.48001954758469056, + 0.22710032562990826, + 0.9632539109975666, + -1.6703737842121655 + ], + "tau": 0.5, + "max_weight": 0.960430543491022, + "entropy": 0.20441768713797392, + "effective_sources": 1.226810469408674, + "min_weight": 0.004953657914530629 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "multiclass", + "seconds": 1.1 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.243175358344919, + 0.16070093323689605, + 0.16243110642819403, + 0.8333930692567719 + ], + "ensemble_val_dice": { + "mixture": 0.8341160757031219, + "mixture@raw": 0.8334634835118209, + "mixture@logit": 0.8348662448964319, + "per_source": 0.833502984539706, + "per_source@raw": 0.833444326075437, + "per_source@logit": 0.8337659337218213, + "uniform": 0.33138683445274353, + "uniform@raw": 0.18373960615593887, + "uniform@logit": 0.2195153840155939 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "multiclass", + "weight_stats": { + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.2581988897471611, + "top_source": "multiclass", + "top_source_is_best_teacher": true + }, + "init_teacher": "multiclass", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 297.8, + "round0_val": { + "val_dice": 0.8819387695596455, + "val_vi": 0.06273517778285083, + "val_ece": 0.01808873124832979 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 67, + "selection_seconds": 24.1, + "round_seconds": 185.7, + "teacher_dice_labeled": [ + 0.28764289107382435, + 2.8393119779215102e-06, + 0.02286523788150808, + 0.931847537434372 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 352197, + "evidence_nonempty_patches": 62, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9407486646169541, + "restart_teacher_dice": 0.931847537434372, + "weights": [ + 5.639154488319541e-05, + 3.102040970640405e-06, + 1.9079057817406194e-06, + 0.9999385985083644 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.931847537434372, + "uniform": 0.35405039348013917, + "top1": 0.931847537434372, + "top2": 0.931847537434372, + "top3": 0.931847537434372 + }, + "val_dice": 0.770057019385612, + "val_vi": 0.10607701602242692, + "val_ece": 0.01113145532193277 + }, + { + "round": 2, + "n_labeled": 134, + "selection_seconds": 12.1, + "round_seconds": 31.1, + "teacher_dice_labeled": [ + 0.33312202368027116, + 0.0011083013511992285, + 0.02410092110532639, + 0.8833937340910062 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 861194, + "evidence_nonempty_patches": 126, + "first_correction": false, + "restart": "", + "weights": [ + 0.0001442513849956721, + 5.091445811611562e-06, + 3.135566847038331e-06, + 0.9998475216023457 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8833957503463934, + "uniform": 0.2672716395099217, + "top1": 0.8833937340910062, + "top2": 0.8833957503463934, + "top3": 0.8833957503463934 + }, + "pseudo_student_dice": 0.7474593431407749, + "pseudo_ensemble_dice": 0.8476539212179403, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8196379716322061, + "val_vi": 0.08018595032909451, + "val_ece": 0.006369059716871562 + }, + { + "round": 3, + "n_labeled": 201, + "selection_seconds": 12.1, + "round_seconds": 41.7, + "teacher_dice_labeled": [ + 0.3303838796661986, + 0.0007402532673849062, + 0.019718253825561582, + 0.8513378216891808 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1289614, + "evidence_nonempty_patches": 191, + "first_correction": false, + "restart": "", + "weights": [ + 0.00019338502509015358, + 6.989356154959776e-06, + 4.135012622737572e-06, + 0.9997954906061322 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8513417719890296, + "uniform": 0.22672250796858034, + "top1": 0.8513378216891808, + "top2": 0.8513414296788804, + "top3": 0.8513414296788804 + }, + "pseudo_student_dice": 0.8287858308200103, + "pseudo_ensemble_dice": 0.7868366198786637, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8618775826113046, + "val_vi": 0.06652593370670218, + "val_ece": 0.003847508910868401 + }, + { + "round": 4, + "n_labeled": 268, + "selection_seconds": 11.9, + "round_seconds": 94.9, + "teacher_dice_labeled": [ + 0.2998075469687518, + 0.00041909380758614286, + 0.020968897534247473, + 0.9054507826871723 + ], + "teacher_dice_pool": [ + 0.2702924441102034, + 0.0002834243480523204, + 0.027147308313248724, + 0.9511278837226628 + ], + "evidence_fg_pixels": 1720970, + "evidence_nonempty_patches": 257, + "first_correction": false, + "restart": "", + "weights": [ + 8.292240772420515e-05, + 4.055858375919286e-06, + 2.437522900096428e-06, + 0.9999105842109998 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9054511955817448, + "uniform": 0.2778366680308455, + "top1": 0.9054507826871723, + "top2": 0.9054515979747473, + "top3": 0.9054513440924151 + }, + "pseudo_student_dice": 0.8568621442055456, + "pseudo_ensemble_dice": 0.8845378373465105, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8439009991942928, + "val_vi": 0.07312091627194423, + "val_ece": 0.005526697548911061 + }, + { + "round": 5, + "n_labeled": 335, + "selection_seconds": 11.8, + "round_seconds": 102.3, + "teacher_dice_labeled": [ + 0.296830503425059, + 0.00042464354617365897, + 0.016892946983330177, + 0.89907961442796 + ], + "teacher_dice_pool": [ + 0.286922745913222, + 0.00038042309825451326, + 0.020411324441041663, + 0.9432745427947116 + ], + "evidence_fg_pixels": 2036353, + "evidence_nonempty_patches": 319, + "first_correction": false, + "restart": "", + "weights": [ + 8.57849329621882e-05, + 4.3228994324122156e-06, + 2.49410839693374e-06, + 0.9999073980592085 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8990799803348473, + "uniform": 0.26021542639168865, + "top1": 0.89907961442796, + "top2": 0.8990803790064736, + "top3": 0.8990802590775561 + }, + "pseudo_student_dice": 0.8135710636987691, + "pseudo_ensemble_dice": 0.8299705604519567, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8459314333171605, + "val_vi": 0.07638859021473715, + "val_ece": 0.0064629955666189835 + }, + { + "round": 6, + "n_labeled": 402, + "selection_seconds": 11.7, + "round_seconds": 109.2, + "teacher_dice_labeled": [ + 0.2940400731578844, + 0.0005404487036796806, + 0.01858962741407579, + 0.9041076431186614 + ], + "teacher_dice_pool": [ + 0.2839103565656452, + 0.0006873655059327154, + 0.02500074273899836, + 0.9541014461743027 + ], + "evidence_fg_pixels": 2426256, + "evidence_nonempty_patches": 385, + "first_correction": false, + "restart": "", + "weights": [ + 7.933390060976356e-05, + 4.115707894799194e-06, + 2.4124059459263477e-06, + 0.9999141379855495 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9041081011357273, + "uniform": 0.2693799768416871, + "top1": 0.9041076431186614, + "top2": 0.904108326361762, + "top3": 0.9041083455547834 + }, + "pseudo_student_dice": 0.8729559237245252, + "pseudo_ensemble_dice": 0.8499822602581402, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8672241877982702, + "val_vi": 0.06396485778756292, + "val_ece": 0.0033871397515547164 + }, + { + "round": 7, + "n_labeled": 402, + "selection_seconds": 0.0, + "round_seconds": 75.6, + "final_consolidation": true, + "val_dice": 0.8793257517796769, + "val_vi": 0.06054513313514344, + "val_ece": 0.0043245230591371225 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.16758785175575056, + "auroc_top20_error": 0.5127568359375 + }, + "al_seconds": 654.0, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 7.933390060976356e-05, + 4.115707894799194e-06, + 2.4124059459263477e-06, + 0.9999141379855495 + ], + "weight_stats_final": { + "max_weight": 0.9999141379855495, + "entropy": 0.0009171585763597139, + "effective_sources": 1.0009175792948988, + "min_weight": 2.4124059459263477e-06 + }, + "top_source_final": "multiclass", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.8949585803283605, + "patch_iou": 0.865657464151904, + "patch_precision": 0.9189963943675663, + "patch_recall": 0.938098837664604, + "patch_vi": 0.04252129109746946, + "n_patches": 1750, + "cal_ece": 0.002869218246236941, + "cal_nll": 0.020062396247747204, + "cal_brier": 0.0038749820454695863, + "cal_mean_confidence": 0.9981043704495272, + "cal_accuracy": 0.9952488839285715, + "cal_predicted_prevalence": 0.04489620535714286, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04487601272824127 + }, + "test_notta": { + "patch_dice": 0.8949585803283605, + "patch_iou": 0.865657464151904, + "patch_precision": 0.9189963943675663, + "patch_recall": 0.938098837664604, + "patch_vi": 0.04252129109746946, + "n_patches": 1750, + "cal_ece": 0.002869218246236941, + "cal_nll": 0.020062396247747204, + "cal_brier": 0.0038749820454695863, + "cal_mean_confidence": 0.9981043704495272, + "cal_accuracy": 0.9952488839285715, + "cal_predicted_prevalence": 0.04489620535714286, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04487601272824127 + }, + "test": { + "patch_dice": 0.9141819338331755, + "patch_iou": 0.8859919092010415, + "patch_precision": 0.9416269089557219, + "patch_recall": 0.9382400828930674, + "patch_vi": 0.0402645834958631, + "n_patches": 1750, + "image_dice": 0.9556097829663196, + "image_iou": 0.9154691932939815, + "image_precision": 0.9632519593336933, + "image_recall": 0.9486664806902545, + "image_vi": 0.04585087873541439, + "n_images": 50, + "cal_ece": 0.0019103181766823621, + "cal_nll": 0.01680154621481725, + "cal_brier": 0.0034478053485783727, + "cal_mean_confidence": 0.9975585181489799, + "cal_accuracy": 0.9956919642857143, + "cal_predicted_prevalence": 0.04447098214285714, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04471203136909009 + }, + "eval_seconds": 26.9, + "total_seconds": 997.1, + "resources": { + "phases": { + "setup": { + "seconds": 0.58, + "peak_cpu_rss_mb": 1027.9, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 15.5, + "peak_cpu_rss_mb": 1801.2, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 297.83, + "peak_cpu_rss_mb": 3685.2, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 654.04, + "peak_cpu_rss_mb": 3685.2, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4176.0, + "rounds": 6 + }, + "inference": { + "seconds": 26.92, + "peak_cpu_rss_mb": 3685.2, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4176.0, + "patches": 1750, + "images": 50, + "patches_per_second": 65.0, + "seconds_per_image": 0.538 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3685.2, + "total_seconds": 994.9 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-margin_b5_s1.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..49e283f840b07489f72731012fb5d12da08cdbd4 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-margin/budget-5/seed-1/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,67,24.1,185.7,"[0.28764289107382435, 2.8393119779215102e-06, 0.02286523788150808, 0.931847537434372]",,352197,62,True,kept,0.9407486646169541,0.931847537434372,"[5.639154488319541e-05, 3.102040970640405e-06, 1.9079057817406194e-06, 0.9999385985083644]",mixture,"{'mixture': 0.931847537434372, 'uniform': 0.35405039348013917, 'top1': 0.931847537434372, 'top2': 0.931847537434372, 'top3': 0.931847537434372}",0.770057019385612,0.10607701602242692,0.01113145532193277,,,,, +2,134,12.1,31.1,"[0.33312202368027116, 0.0011083013511992285, 0.02410092110532639, 0.8833937340910062]",,861194,126,False,,,,"[0.0001442513849956721, 5.091445811611562e-06, 3.135566847038331e-06, 0.9998475216023457]",mixture,"{'mixture': 0.8833957503463934, 'uniform': 0.2672716395099217, 'top1': 0.8833937340910062, 'top2': 0.8833957503463934, 'top3': 0.8833957503463934}",0.8196379716322061,0.08018595032909451,0.006369059716871562,0.7474593431407749,0.8476539212179403,ensemble,1.0, +3,201,12.1,41.7,"[0.3303838796661986, 0.0007402532673849062, 0.019718253825561582, 0.8513378216891808]",,1289614,191,False,,,,"[0.00019338502509015358, 6.989356154959776e-06, 4.135012622737572e-06, 0.9997954906061322]",mixture,"{'mixture': 0.8513417719890296, 'uniform': 0.22672250796858034, 'top1': 0.8513378216891808, 'top2': 0.8513414296788804, 'top3': 0.8513414296788804}",0.8618775826113046,0.06652593370670218,0.003847508910868401,0.8287858308200103,0.7868366198786637,student,1.0, +4,268,11.9,94.9,"[0.2998075469687518, 0.00041909380758614286, 0.020968897534247473, 0.9054507826871723]","[0.2702924441102034, 0.0002834243480523204, 0.027147308313248724, 0.9511278837226628]",1720970,257,False,,,,"[8.292240772420515e-05, 4.055858375919286e-06, 2.437522900096428e-06, 0.9999105842109998]",mixture,"{'mixture': 0.9054511955817448, 'uniform': 0.2778366680308455, 'top1': 0.9054507826871723, 'top2': 0.9054515979747473, 'top3': 0.9054513440924151}",0.8439009991942928,0.07312091627194423,0.005526697548911061,0.8568621442055456,0.8845378373465105,student,1.0, +5,335,11.8,102.3,"[0.296830503425059, 0.00042464354617365897, 0.016892946983330177, 0.89907961442796]","[0.286922745913222, 0.00038042309825451326, 0.020411324441041663, 0.9432745427947116]",2036353,319,False,,,,"[8.57849329621882e-05, 4.3228994324122156e-06, 2.49410839693374e-06, 0.9999073980592085]",mixture,"{'mixture': 0.8990799803348473, 'uniform': 0.26021542639168865, 'top1': 0.89907961442796, 'top2': 0.8990803790064736, 'top3': 0.8990802590775561}",0.8459314333171605,0.07638859021473715,0.0064629955666189835,0.8135710636987691,0.8299705604519567,student,1.0, +6,402,11.7,109.2,"[0.2940400731578844, 0.0005404487036796806, 0.01858962741407579, 0.9041076431186614]","[0.2839103565656452, 0.0006873655059327154, 0.02500074273899836, 0.9541014461743027]",2426256,385,False,,,,"[7.933390060976356e-05, 4.115707894799194e-06, 2.4124059459263477e-06, 0.9999141379855495]",mixture,"{'mixture': 0.9041081011357273, 'uniform': 0.2693799768416871, 'top1': 0.9041076431186614, 'top2': 0.904108326361762, 'top3': 0.9041083455547834}",0.8672241877982702,0.06396485778756292,0.0033871397515547164,0.8729559237245252,0.8499822602581402,student,1.0, +7,402,0.0,75.6,,,,,,,,,,,,0.8793257517796769,0.06054513313514344,0.0043245230591371225,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/log.txt b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..31081c5618f6d353dd61f7d5e51e1d272fc470bb --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/log.txt @@ -0,0 +1,107 @@ +[04:08:04][ 0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-mcdropout budget=5.0% seed=1 | {"task": "mitochondria", "target": "lucchi-pp", "variant": "macs-v11-norestart-acq-mcdropout", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "mcdropout", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[04:08:04][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[04:08:06][ 2s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches +[04:08:11][ 7s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60 +[04:08:13][ 10s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 3s +[04:08:14][ 11s][WEIGHTS] mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 | eff.sources=1.44 | MMD2 mix=0.0050 uniform=0.2092 +[04:08:14][ 11s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass +[04:08:28][ 24s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833 +[04:08:28][ 24s][TEACHERS] ensemble target-val Dice: mixture=0.834, mixture@raw=0.833, mixture@logit=0.835, per_source=0.834, per_source@raw=0.833, per_source@logit=0.834, uniform=0.331, uniform@raw=0.184, uniform@logit=0.220 +[04:08:28][ 24s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 +[04:08:28][ 24s][INIT] student initialised from teacher multiclass (weight 0.920) +[04:08:30][ 26s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[04:08:30][ 26s][KD] distilling the mixture ensemble (prob space) into the student +[04:09:04][ 61s][KD] soft labels for 6000 patches from 4 teachers in 34s +[04:09:58][ 115s][KD] epoch 1/5 | patches=6000 | loss=0.1185 | 53s +[04:10:51][ 167s][KD] epoch 2/5 | patches=6000 | loss=0.1064 | 53s +[04:11:44][ 220s][KD] epoch 3/5 | patches=6000 | loss=0.1044 | 53s +[04:12:36][ 272s][KD] epoch 4/5 | patches=6000 | loss=0.1035 | 52s +[04:13:27][ 324s][KD] epoch 5/5 | patches=6000 | loss=0.1037 | 52s +[04:13:29][ 326s][ROUND0] before any target label: val_dice=0.8829 val_vi=0.0629 +[04:13:29][ 326s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=mcdropout | first_round=stratified +[04:13:44][ 340s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.08059150136378555, 'auroc_top20_error': 0.460871337890625} +[04:13:56][ 352s][AL] round 1: stratified cold start: 34 patches from the membrane-rich half (ensemble prevalence >= 0.074) + 33 from the membrane-poor half of 2000 candidates +[04:13:58][ 354s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.932, uniform=0.354, top1=0.932, top2=0.932, top3=0.932 -> using mixture +[04:13:58][ 355s][WEIGHTS] round 1: student kept (labeled Dice 0.941 vs best teacher multiclass 0.932); distilling the corrected ensemble into it +[04:13:58][ 355s][WEIGHTS] round 1: teacher Dice on 67 labeled patches + 0 blank (352197 fg px): fly-mito=0.288, kasthuri=0.000, dros-vnc=0.023, multiclass=0.932 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[04:14:33][ 389s][KD] soft labels for 6000 patches from 4 teachers in 34s +[04:15:26][ 442s][KD] epoch 1/2 | patches=6000 | loss=0.0469 | 53s +[04:16:18][ 495s][KD] epoch 2/2 | patches=6000 | loss=0.0460 | 52s +[04:16:30][ 506s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 12s +[04:16:31][ 508s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3156 +[04:16:33][ 510s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3717 +[04:16:35][ 511s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3763 +[04:16:36][ 513s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.2960 +[04:16:38][ 514s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.3355 +[04:16:40][ 516s][AL] round 1/6 | labeled=67 | selection_seconds=26.5000 round_seconds=188.4000 restart_student_dice=0.9406 restart_teacher_dice=0.9318 val_dice=0.8396 val_vi=0.0774 val_ece=0.0046 +[04:16:54][ 530s][AL] round 2: 1211/7983 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:16:58][ 534s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.907, uniform=0.256, top1=0.907, top2=0.907, top3=0.907 -> using mixture +[04:16:58][ 534s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (753797 fg px): fly-mito=0.301, kasthuri=0.000, dros-vnc=0.016, multiclass=0.907 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[04:17:00][ 536s][CONS] round 2: on the 67 new patches student Dice=0.875 vs ensemble 0.884 -> pseudo-labels from the ensemble | consistency x1.00 +[04:17:03][ 539s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.3546 +[04:17:05][ 542s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.2868 +[04:17:08][ 544s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.2574 +[04:17:11][ 547s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.2519 +[04:17:13][ 550s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.2164 +[04:17:15][ 552s][AL] round 2/6 | labeled=134 | selection_seconds=14.7000 round_seconds=33.6000 pseudo_student_dice=0.8753 pseudo_ensemble_dice=0.8838 consistency_scale=1.0000 val_dice=0.8534 val_vi=0.0742 val_ece=0.0062 +[04:17:29][ 566s][AL] round 3: 1348/7916 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:17:34][ 571s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.865, uniform=0.213, top1=0.865, top2=0.865, top3=0.865 -> using mixture +[04:17:34][ 571s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (1206872 fg px): fly-mito=0.309, kasthuri=0.000, dros-vnc=0.014, multiclass=0.865 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[04:17:40][ 576s][CONS] round 3: on the 67 new patches student Dice=0.827 vs ensemble 0.794 -> pseudo-labels from the student | consistency x1.00 +[04:17:44][ 580s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.2933 +[04:17:47][ 584s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.3266 +[04:17:51][ 588s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.2655 +[04:17:55][ 592s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.2243 +[04:17:59][ 595s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2573 +[04:18:01][ 597s][AL] round 3/6 | labeled=201 | selection_seconds=14.5000 round_seconds=44.0000 pseudo_student_dice=0.8269 pseudo_ensemble_dice=0.7938 consistency_scale=1.0000 val_dice=0.8262 val_vi=0.0799 val_ece=0.0047 +[04:18:15][ 612s][AL] round 4: 1148/7849 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:18:43][ 640s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.280, kasthuri=0.000, dros-vnc=0.020, multiclass=0.953 | blended with labeled Dice at 0.50 +[04:19:08][ 665s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.899, uniform=0.275, top1=0.899, top2=0.899, top3=0.899 -> using mixture +[04:19:08][ 665s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1568189 fg px): fly-mito=0.289, kasthuri=0.000, dros-vnc=0.016, multiclass=0.899 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[04:19:14][ 670s][CONS] round 4: on the 67 new patches student Dice=0.854 vs ensemble 0.786 -> pseudo-labels from the student | consistency x1.00 +[04:19:19][ 675s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.2543 +[04:19:24][ 680s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.2242 +[04:19:29][ 685s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.2283 +[04:19:33][ 690s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.2017 +[04:19:38][ 695s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.1811 +[04:19:40][ 697s][AL] round 4/6 | labeled=268 | selection_seconds=14.3000 round_seconds=97.5000 pseudo_student_dice=0.8542 pseudo_ensemble_dice=0.7857 consistency_scale=1.0000 val_dice=0.8594 val_vi=0.0676 val_ece=0.0054 +[04:19:54][ 711s][AL] round 5: 1267/7782 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:20:24][ 740s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.290, kasthuri=0.001, dros-vnc=0.023, multiclass=0.955 | blended with labeled Dice at 0.50 +[04:20:49][ 766s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.907, uniform=0.271, top1=0.907, top2=0.907, top3=0.907 -> using mixture +[04:20:49][ 766s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (1973080 fg px): fly-mito=0.295, kasthuri=0.000, dros-vnc=0.017, multiclass=0.907 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[04:20:54][ 771s][CONS] round 5: on the 67 new patches student Dice=0.893 vs ensemble 0.908 -> pseudo-labels from the student | consistency x1.00 +[04:21:00][ 777s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2118 +[04:21:07][ 783s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.1957 +[04:21:13][ 789s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.1964 +[04:21:19][ 795s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.1742 +[04:21:25][ 801s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.1745 +[04:21:27][ 803s][AL] round 5/6 | labeled=335 | selection_seconds=14.3000 round_seconds=104.4000 pseudo_student_dice=0.8930 pseudo_ensemble_dice=0.9077 consistency_scale=1.0000 val_dice=0.8671 val_vi=0.0653 val_ece=0.0040 +[04:21:41][ 817s][AL] round 6: 1214/7715 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:22:10][ 847s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.278, kasthuri=0.000, dros-vnc=0.019, multiclass=0.954 | blended with labeled Dice at 0.50 +[04:22:37][ 873s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.905, uniform=0.269, top1=0.905, top2=0.905, top3=0.905 -> using mixture +[04:22:37][ 873s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (2355751 fg px): fly-mito=0.288, kasthuri=0.000, dros-vnc=0.014, multiclass=0.905 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[04:22:42][ 878s][CONS] round 6: on the 67 new patches student Dice=0.882 vs ensemble 0.842 -> pseudo-labels from the student | consistency x1.00 +[04:22:49][ 886s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2324 +[04:22:56][ 893s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.2326 +[04:23:04][ 900s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.1990 +[04:23:11][ 907s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.1869 +[04:23:18][ 915s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.1774 +[04:23:20][ 917s][AL] round 6/6 | labeled=402 | selection_seconds=14.1000 round_seconds=111.5000 pseudo_student_dice=0.8817 pseudo_ensemble_dice=0.8422 consistency_scale=1.0000 val_dice=0.8858 val_vi=0.0604 val_ece=0.0033 +[04:23:23][ 920s][CONS] final consolidation: pseudo-labels refreshed from the student +[04:23:30][ 927s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.1913 +[04:23:38][ 934s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.2009 +[04:23:45][ 941s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.1591 +[04:23:52][ 948s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.1540 +[04:23:59][ 956s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.1328 +[04:24:06][ 963s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1720 +[04:24:14][ 970s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1422 +[04:24:21][ 977s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1265 +[04:24:28][ 985s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1171 +[04:24:35][ 992s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1206 +[04:24:37][ 994s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=75.2000 val_dice=0.8828 val_vi=0.0602 val_ece=0.0042 +[04:24:37][ 994s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[04:25:01][ 1017s][TEST] raw student (no EMA, no TTA): patch_dice=0.9158 patch_vi=0.0387 +[04:25:04][ 1020s][TEST] final model without TTA: patch_dice=0.9158 patch_vi=0.0387 +[04:25:04][ 1020s][RESOURCES] {"phases": {"setup": {"seconds": 4.44, "peak_cpu_rss_mb": 1026.8, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 19.41, "peak_cpu_rss_mb": 1791.9, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 297.42, "peak_cpu_rss_mb": 3684.1, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 667.93, "peak_cpu_rss_mb": 3684.1, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 26.74, "peak_cpu_rss_mb": 3684.1, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 65.4, "seconds_per_image": 0.535}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3684.1, "total_seconds": 1015.9} +[04:25:04][ 1020s][TEST] patch_dice=0.9199 patch_iou=0.8930 patch_precision=0.9502 patch_recall=0.9374 patch_vi=0.0378 image_dice=0.9595 image_iou=0.9225 image_precision=0.9729 image_recall=0.9471 image_vi=0.0421 cal_ece=0.0017 cal_nll=0.0157 cal_brier=0.0032 cal_mean_confidence=0.9977 cal_accuracy=0.9960 cal_predicted_prevalence=0.0440 cal_true_prevalence=0.0452 cal_mean_probability=0.0443 +[04:25:04][ 1020s][DONE] lucchi-pp macs-v11-norestart-acq-mcdropout b5 s1 | test patch Dice=0.9199 VI=0.0378 | image Dice=0.9595 VI=0.0421 | 1020s diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/result.json b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/result.json new file mode 100644 index 0000000000000000000000000000000000000000..dcd68e245ef44adc2907e80b0f5007110296c32f --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/result.json @@ -0,0 +1,662 @@ +{ + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 1, + "config": { + "task": "mitochondria", + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 1, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "mcdropout", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "lucchi-pp", + "--variants", + "macs-v11-norestart-acq-mcdropout", + "--budgets", + "5", + "--seeds", + "1", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "per_source": [ + 0.01302141100636151, + 0.021594387588085767, + 0.004953657914530629, + 0.960430543491022 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.001413550071584186, + "mmd2_mixture": 0.005039751415091165, + "mmd2_uniform": 0.2092225964739336, + "mmd2_single_source": [ + 0.5680514029229977, + 0.4644134839747518, + 0.6002496450435827, + 0.004698779437425582 + ], + "median_squared_distance": 0.18927357450912807, + "gammas": [ + 1.3208394285803657, + 2.6416788571607315, + 5.283357714321463, + 10.566715428642926, + 21.133430857285852 + ], + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6675364398352843, + 0.5899254221485877, + 0.8158221669373606, + 0.007664827003586616 + ], + "standardized": [ + 0.48001954758469056, + 0.22710032562990826, + 0.9632539109975666, + -1.6703737842121655 + ], + "tau": 0.5, + "max_weight": 0.960430543491022, + "entropy": 0.20441768713797392, + "effective_sources": 1.226810469408674, + "min_weight": 0.004953657914530629 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "multiclass", + "seconds": 3.9 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.243175358344919, + 0.16070093323689605, + 0.16243110642819403, + 0.8333930692567719 + ], + "ensemble_val_dice": { + "mixture": 0.8341160757031219, + "mixture@raw": 0.8334634835118209, + "mixture@logit": 0.8348662448964319, + "per_source": 0.833502984539706, + "per_source@raw": 0.833444326075437, + "per_source@logit": 0.8337659337218213, + "uniform": 0.33138683445274353, + "uniform@raw": 0.18373960615593887, + "uniform@logit": 0.2195153840155939 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "multiclass", + "weight_stats": { + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.2581988897471611, + "top_source": "multiclass", + "top_source_is_best_teacher": true + }, + "init_teacher": "multiclass", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 297.4, + "round0_val": { + "val_dice": 0.8828835329074521, + "val_vi": 0.06287297304223494, + "val_ece": 0.017977323639087096 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 67, + "selection_seconds": 26.5, + "round_seconds": 188.4, + "teacher_dice_labeled": [ + 0.28764289107382435, + 2.8393119779215102e-06, + 0.02286523788150808, + 0.931847537434372 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 352197, + "evidence_nonempty_patches": 62, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9405833832935072, + "restart_teacher_dice": 0.931847537434372, + "weights": [ + 5.639154488319541e-05, + 3.102040970640405e-06, + 1.9079057817406194e-06, + 0.9999385985083644 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.931847537434372, + "uniform": 0.35405039348013917, + "top1": 0.931847537434372, + "top2": 0.931847537434372, + "top3": 0.931847537434372 + }, + "val_dice": 0.8396091755576145, + "val_vi": 0.07736927737850084, + "val_ece": 0.004640707095079419 + }, + { + "round": 2, + "n_labeled": 134, + "selection_seconds": 14.7, + "round_seconds": 33.6, + "teacher_dice_labeled": [ + 0.3011068765921574, + 1.3266153531848055e-06, + 0.01569504071448025, + 0.9067196874507328 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 753797, + "evidence_nonempty_patches": 126, + "first_correction": false, + "restart": "", + "weights": [ + 8.294765692777686e-05, + 3.988023823667651e-06, + 2.283147332119484e-06, + 0.9999107811719165 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9067203009655527, + "uniform": 0.25551139294976644, + "top1": 0.9067196874507328, + "top2": 0.9067203009655527, + "top3": 0.9067203009655527 + }, + "pseudo_student_dice": 0.8752639476687243, + "pseudo_ensemble_dice": 0.8837822445994745, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8533680473829939, + "val_vi": 0.07419593523986495, + "val_ece": 0.006158910996843013 + }, + { + "round": 3, + "n_labeled": 201, + "selection_seconds": 14.5, + "round_seconds": 44.0, + "teacher_dice_labeled": [ + 0.3087750673805181, + 8.285875978665526e-07, + 0.013835628894829068, + 0.8648362461079152 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1206872, + "evidence_nonempty_patches": 193, + "first_correction": false, + "restart": "", + "weights": [ + 0.0001361380842502468, + 6.062176320072765e-06, + 3.406682011317713e-06, + 0.9998543930574184 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8648392611184557, + "uniform": 0.21312755144915677, + "top1": 0.8648362461079152, + "top2": 0.8648393762131779, + "top3": 0.8648393762131779 + }, + "pseudo_student_dice": 0.8269207228528156, + "pseudo_ensemble_dice": 0.7937579957233646, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8261597506474494, + "val_vi": 0.07990777521122358, + "val_ece": 0.004743392772214335 + }, + { + "round": 4, + "n_labeled": 268, + "selection_seconds": 14.3, + "round_seconds": 97.5, + "teacher_dice_labeled": [ + 0.2892263085845723, + 7.075814994432409e-05, + 0.016203675526485186, + 0.8992552224594974 + ], + "teacher_dice_pool": [ + 0.2803271103745624, + 0.00014087862205624266, + 0.020335466602570428, + 0.9531202058815045 + ], + "evidence_fg_pixels": 1568189, + "evidence_nonempty_patches": 258, + "first_correction": false, + "restart": "", + "weights": [ + 7.936455206912762e-05, + 4.300098257328851e-06, + 2.4726463645173816e-06, + 0.9999138627033091 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8992559365537377, + "uniform": 0.2746789860548836, + "top1": 0.8992552224594974, + "top2": 0.8992563121458366, + "top3": 0.8992562592548089 + }, + "pseudo_student_dice": 0.8541598889899547, + "pseudo_ensemble_dice": 0.7856628171412207, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8594037441389399, + "val_vi": 0.06757612107032326, + "val_ece": 0.005420278845385046 + }, + { + "round": 5, + "n_labeled": 335, + "selection_seconds": 14.3, + "round_seconds": 104.4, + "teacher_dice_labeled": [ + 0.29474300397585645, + 0.00028153880854306037, + 0.01675766866521816, + 0.9068035993400686 + ], + "teacher_dice_pool": [ + 0.29016305965059463, + 0.0005625711133245705, + 0.023050799231682424, + 0.9554395198623801 + ], + "evidence_fg_pixels": 1973080, + "evidence_nonempty_patches": 320, + "first_correction": false, + "restart": "", + "weights": [ + 7.77685473402855e-05, + 3.99588059093161e-06, + 2.305614441136434e-06, + 0.9999159299576277 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9068045377909055, + "uniform": 0.2707635372794016, + "top1": 0.9068035993400686, + "top2": 0.9068046621977184, + "top3": 0.906804514211268 + }, + "pseudo_student_dice": 0.8929506153219638, + "pseudo_ensemble_dice": 0.9077268055685321, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8670737659208252, + "val_vi": 0.06528445468735829, + "val_ece": 0.003970009200344247 + }, + { + "round": 6, + "n_labeled": 402, + "selection_seconds": 14.1, + "round_seconds": 111.5, + "teacher_dice_labeled": [ + 0.2881840665547969, + 0.00022079237833331483, + 0.01406424505050275, + 0.9049557140833979 + ], + "teacher_dice_pool": [ + 0.2784077529876977, + 0.0004411604867291246, + 0.019176387238284822, + 0.9542683999177805 + ], + "evidence_fg_pixels": 2355751, + "evidence_nonempty_patches": 382, + "first_correction": false, + "restart": "", + "weights": [ + 7.419005251816658e-05, + 4.067948802061995e-06, + 2.2862098345211875e-06, + 0.9999194557888453 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.904956811049235, + "uniform": 0.268674041517003, + "top1": 0.9049557140833979, + "top2": 0.9049568444205593, + "top3": 0.9049568686724057 + }, + "pseudo_student_dice": 0.8816519623345359, + "pseudo_ensemble_dice": 0.8421670436279903, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8858488334790857, + "val_vi": 0.060419605472612346, + "val_ece": 0.0033178283843365874 + }, + { + "round": 7, + "n_labeled": 402, + "selection_seconds": 0.0, + "round_seconds": 75.2, + "final_consolidation": true, + "val_dice": 0.8828107213324201, + "val_vi": 0.06019669163854541, + "val_ece": 0.0042202416444402325 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.08059150136378555, + "auroc_top20_error": 0.460871337890625 + }, + "al_seconds": 667.9, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 7.419005251816658e-05, + 4.067948802061995e-06, + 2.2862098345211875e-06, + 0.9999194557888453 + ], + "weight_stats_final": { + "max_weight": 0.9999194557888453, + "entropy": 0.0008661929018405332, + "effective_sources": 1.0008665681552515, + "min_weight": 2.2862098345211875e-06 + }, + "top_source_final": "multiclass", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.9157855515031588, + "patch_iou": 0.8874205055772556, + "patch_precision": 0.9405785973865597, + "patch_recall": 0.9415064399571657, + "patch_vi": 0.038743517732617, + "n_patches": 1750, + "cal_ece": 0.0024675705473814194, + "cal_nll": 0.01878510760898028, + "cal_brier": 0.0033792735670207666, + "cal_mean_confidence": 0.9983581955473814, + "cal_accuracy": 0.995890625, + "cal_predicted_prevalence": 0.04443526785714286, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04435463535560046 + }, + "test_notta": { + "patch_dice": 0.9157855515031588, + "patch_iou": 0.8874205055772556, + "patch_precision": 0.9405785973865597, + "patch_recall": 0.9415064399571657, + "patch_vi": 0.038743517732617, + "n_patches": 1750, + "cal_ece": 0.0024675705473814194, + "cal_nll": 0.01878510760898028, + "cal_brier": 0.0033792735670207666, + "cal_mean_confidence": 0.9983581955473814, + "cal_accuracy": 0.995890625, + "cal_predicted_prevalence": 0.04443526785714286, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04435463535560046 + }, + "test": { + "patch_dice": 0.9198535244380187, + "patch_iou": 0.892963054722157, + "patch_precision": 0.9501963726165985, + "patch_recall": 0.937402500992288, + "patch_vi": 0.03784304238065405, + "n_patches": 1750, + "image_dice": 0.9594996689103868, + "image_iou": 0.9225252007267537, + "image_precision": 0.9728874469845783, + "image_recall": 0.9470512163713781, + "image_vi": 0.04205999326081306, + "n_images": 50, + "cal_ece": 0.00173885814855935, + "cal_nll": 0.01574090217454858, + "cal_brier": 0.0031900130058693603, + "cal_mean_confidence": 0.9977119275408037, + "cal_accuracy": 0.9959827008928571, + "cal_predicted_prevalence": 0.04404743303571428, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04428765661220907 + }, + "eval_seconds": 26.7, + "total_seconds": 1020.4, + "resources": { + "phases": { + "setup": { + "seconds": 4.44, + "peak_cpu_rss_mb": 1026.8, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 19.41, + "peak_cpu_rss_mb": 1791.9, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 297.42, + "peak_cpu_rss_mb": 3684.1, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 667.93, + "peak_cpu_rss_mb": 3684.1, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4176.0, + "rounds": 6 + }, + "inference": { + "seconds": 26.74, + "peak_cpu_rss_mb": 3684.1, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4176.0, + "patches": 1750, + "images": 50, + "patches_per_second": 65.4, + "seconds_per_image": 0.535 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3684.1, + "total_seconds": 1015.9 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-mcdropout_b5_s1.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/rounds.csv b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..156c1fa4f445bec288d524879b4508bc8e814616 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,67,26.5,188.4,"[0.28764289107382435, 2.8393119779215102e-06, 0.02286523788150808, 0.931847537434372]",,352197,62,True,kept,0.9405833832935072,0.931847537434372,"[5.639154488319541e-05, 3.102040970640405e-06, 1.9079057817406194e-06, 0.9999385985083644]",mixture,"{'mixture': 0.931847537434372, 'uniform': 0.35405039348013917, 'top1': 0.931847537434372, 'top2': 0.931847537434372, 'top3': 0.931847537434372}",0.8396091755576145,0.07736927737850084,0.004640707095079419,,,,, +2,134,14.7,33.6,"[0.3011068765921574, 1.3266153531848055e-06, 0.01569504071448025, 0.9067196874507328]",,753797,126,False,,,,"[8.294765692777686e-05, 3.988023823667651e-06, 2.283147332119484e-06, 0.9999107811719165]",mixture,"{'mixture': 0.9067203009655527, 'uniform': 0.25551139294976644, 'top1': 0.9067196874507328, 'top2': 0.9067203009655527, 'top3': 0.9067203009655527}",0.8533680473829939,0.07419593523986495,0.006158910996843013,0.8752639476687243,0.8837822445994745,ensemble,1.0, +3,201,14.5,44.0,"[0.3087750673805181, 8.285875978665526e-07, 0.013835628894829068, 0.8648362461079152]",,1206872,193,False,,,,"[0.0001361380842502468, 6.062176320072765e-06, 3.406682011317713e-06, 0.9998543930574184]",mixture,"{'mixture': 0.8648392611184557, 'uniform': 0.21312755144915677, 'top1': 0.8648362461079152, 'top2': 0.8648393762131779, 'top3': 0.8648393762131779}",0.8261597506474494,0.07990777521122358,0.004743392772214335,0.8269207228528156,0.7937579957233646,student,1.0, +4,268,14.3,97.5,"[0.2892263085845723, 7.075814994432409e-05, 0.016203675526485186, 0.8992552224594974]","[0.2803271103745624, 0.00014087862205624266, 0.020335466602570428, 0.9531202058815045]",1568189,258,False,,,,"[7.936455206912762e-05, 4.300098257328851e-06, 2.4726463645173816e-06, 0.9999138627033091]",mixture,"{'mixture': 0.8992559365537377, 'uniform': 0.2746789860548836, 'top1': 0.8992552224594974, 'top2': 0.8992563121458366, 'top3': 0.8992562592548089}",0.8594037441389399,0.06757612107032326,0.005420278845385046,0.8541598889899547,0.7856628171412207,student,1.0, +5,335,14.3,104.4,"[0.29474300397585645, 0.00028153880854306037, 0.01675766866521816, 0.9068035993400686]","[0.29016305965059463, 0.0005625711133245705, 0.023050799231682424, 0.9554395198623801]",1973080,320,False,,,,"[7.77685473402855e-05, 3.99588059093161e-06, 2.305614441136434e-06, 0.9999159299576277]",mixture,"{'mixture': 0.9068045377909055, 'uniform': 0.2707635372794016, 'top1': 0.9068035993400686, 'top2': 0.9068046621977184, 'top3': 0.906804514211268}",0.8670737659208252,0.06528445468735829,0.003970009200344247,0.8929506153219638,0.9077268055685321,student,1.0, +6,402,14.1,111.5,"[0.2881840665547969, 0.00022079237833331483, 0.01406424505050275, 0.9049557140833979]","[0.2784077529876977, 0.0004411604867291246, 0.019176387238284822, 0.9542683999177805]",2355751,382,False,,,,"[7.419005251816658e-05, 4.067948802061995e-06, 2.2862098345211875e-06, 0.9999194557888453]",mixture,"{'mixture': 0.904956811049235, 'uniform': 0.268674041517003, 'top1': 0.9049557140833979, 'top2': 0.9049568444205593, 'top3': 0.9049568686724057}",0.8858488334790857,0.060419605472612346,0.0033178283843365874,0.8816519623345359,0.8421670436279903,student,1.0, +7,402,0.0,75.2,,,,,,,,,,,,0.8828107213324201,0.06019669163854541,0.0042202416444402325,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/log.txt b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..a8063160c9ef754d8b6be98c79df07aa9a4d9201 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/log.txt @@ -0,0 +1,103 @@ +[07:54:02][ 0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-mcdropout budget=5.0% seed=2 | {"task": "mitochondria", "target": "lucchi-pp", "variant": "macs-v11-norestart-acq-mcdropout", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "mcdropout", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[07:54:02][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[07:54:02][ 0s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches +[07:54:02][ 1s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60 +[07:54:03][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[07:54:04][ 2s][WEIGHTS] mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 | eff.sources=1.44 | MMD2 mix=0.0050 uniform=0.2092 +[07:54:04][ 2s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass +[07:54:17][ 15s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833 +[07:54:17][ 15s][TEACHERS] ensemble target-val Dice: mixture=0.834, mixture@raw=0.833, mixture@logit=0.835, per_source=0.834, per_source@raw=0.833, per_source@logit=0.834, uniform=0.331, uniform@raw=0.184, uniform@logit=0.220 +[07:54:17][ 15s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 +[07:54:17][ 15s][INIT] student initialised from teacher multiclass (weight 0.920) +[07:54:18][ 16s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[07:54:18][ 16s][KD] distilling the mixture ensemble (prob space) into the student +[07:54:52][ 51s][KD] soft labels for 6000 patches from 4 teachers in 34s +[07:55:46][ 104s][KD] epoch 1/5 | patches=6000 | loss=0.1187 | 53s +[07:56:38][ 156s][KD] epoch 2/5 | patches=6000 | loss=0.1067 | 52s +[07:57:31][ 209s][KD] epoch 3/5 | patches=6000 | loss=0.1048 | 53s +[07:58:24][ 263s][KD] epoch 4/5 | patches=6000 | loss=0.1039 | 53s +[07:59:17][ 315s][KD] epoch 5/5 | patches=6000 | loss=0.1030 | 53s +[07:59:19][ 317s][ROUND0] before any target label: val_dice=0.8779 val_vi=0.0683 +[07:59:19][ 317s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=mcdropout | first_round=stratified +[07:59:34][ 332s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.0970013558496726, 'auroc_top20_error': 0.47497646484375} +[07:59:45][ 344s][AL] round 1: stratified cold start: 34 patches from the membrane-rich half (ensemble prevalence >= 0.072) + 33 from the membrane-poor half of 2000 candidates +[07:59:47][ 346s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.946, uniform=0.320, top1=0.946, top2=0.946, top3=0.946 -> using mixture +[07:59:48][ 346s][WEIGHTS] round 1: student kept (labeled Dice 0.947 vs best teacher multiclass 0.946); distilling the corrected ensemble into it +[07:59:48][ 346s][WEIGHTS] round 1: teacher Dice on 67 labeled patches + 0 blank (313095 fg px): fly-mito=0.265, kasthuri=0.000, dros-vnc=0.016, multiclass=0.946 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:00:22][ 381s][KD] soft labels for 6000 patches from 4 teachers in 34s +[08:01:15][ 434s][KD] epoch 1/2 | patches=6000 | loss=0.0470 | 53s +[08:02:08][ 487s][KD] epoch 2/2 | patches=6000 | loss=0.0461 | 53s +[08:02:20][ 498s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 12s +[08:02:22][ 500s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3228 +[08:02:23][ 501s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3542 +[08:02:25][ 503s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3475 +[08:02:26][ 505s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.3536 +[08:02:28][ 506s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.3081 +[08:02:30][ 508s][AL] round 1/6 | labeled=67 | selection_seconds=26.4000 round_seconds=188.8000 restart_student_dice=0.9466 restart_teacher_dice=0.9463 val_dice=0.8414 val_vi=0.0867 val_ece=0.0070 +[08:02:44][ 522s][AL] round 2: 1353/7983 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:02:48][ 526s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.838, uniform=0.243, top1=0.838, top2=0.838, top3=0.838 -> using mixture +[08:02:48][ 526s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (803768 fg px): fly-mito=0.299, kasthuri=0.000, dros-vnc=0.020, multiclass=0.838 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:02:50][ 528s][CONS] round 2: on the 67 new patches student Dice=0.712 vs ensemble 0.762 -> pseudo-labels from the ensemble | consistency x1.00 +[08:02:52][ 531s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.5025 +[08:02:55][ 533s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.4224 +[08:02:58][ 536s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.3544 +[08:03:00][ 539s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.3758 +[08:03:03][ 541s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.3122 +[08:03:05][ 543s][AL] round 2/6 | labeled=134 | selection_seconds=14.5000 round_seconds=33.4000 pseudo_student_dice=0.7120 pseudo_ensemble_dice=0.7619 consistency_scale=1.0000 val_dice=0.8598 val_vi=0.0659 val_ece=0.0028 +[08:03:19][ 558s][AL] round 3: 1280/7916 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:03:24][ 563s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.830, uniform=0.234, top1=0.830, top2=0.830, top3=0.830 -> using mixture +[08:03:24][ 563s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (1174472 fg px): fly-mito=0.289, kasthuri=0.000, dros-vnc=0.016, multiclass=0.830 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:03:26][ 565s][CONS] round 3: on the 67 new patches student Dice=0.822 vs ensemble 0.813 -> pseudo-labels from the ensemble | consistency x1.00 +[08:03:30][ 569s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.3336 +[08:03:34][ 573s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.3384 +[08:03:38][ 576s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.3453 +[08:03:42][ 580s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.2909 +[08:03:46][ 584s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2735 +[08:03:48][ 586s][AL] round 3/6 | labeled=201 | selection_seconds=14.5000 round_seconds=40.8000 pseudo_student_dice=0.8225 pseudo_ensemble_dice=0.8134 consistency_scale=1.0000 val_dice=0.8554 val_vi=0.0688 val_ece=0.0038 +[08:04:02][ 600s][AL] round 4: 1338/7849 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:04:08][ 607s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.840, uniform=0.246, top1=0.840, top2=0.840, top3=0.840 -> using mixture +[08:04:08][ 607s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1668205 fg px): fly-mito=0.312, kasthuri=0.000, dros-vnc=0.019, multiclass=0.840 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:04:10][ 609s][CONS] round 4: on the 67 new patches student Dice=0.828 vs ensemble 0.867 -> pseudo-labels from the ensemble | consistency x1.00 +[08:04:15][ 614s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.2599 +[08:04:20][ 619s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.2425 +[08:04:25][ 624s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.2168 +[08:04:30][ 629s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.2005 +[08:04:35][ 634s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.2388 +[08:04:37][ 635s][AL] round 4/6 | labeled=268 | selection_seconds=14.4000 round_seconds=47.6000 pseudo_student_dice=0.8280 pseudo_ensemble_dice=0.8674 consistency_scale=1.0000 val_dice=0.8802 val_vi=0.0636 val_ece=0.0035 +[08:04:51][ 650s][AL] round 5: 1296/7782 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:04:59][ 657s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.848, uniform=0.235, top1=0.848, top2=0.848, top3=0.848 -> using mixture +[08:04:59][ 657s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (2074752 fg px): fly-mito=0.312, kasthuri=0.000, dros-vnc=0.016, multiclass=0.848 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:05:01][ 659s][CONS] round 5: on the 67 new patches student Dice=0.877 vs ensemble 0.880 -> pseudo-labels from the ensemble | consistency x1.00 +[08:05:07][ 666s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2388 +[08:05:13][ 672s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.2234 +[08:05:19][ 678s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.2142 +[08:05:25][ 684s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.2186 +[08:05:32][ 690s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.2137 +[08:05:33][ 692s][AL] round 5/6 | labeled=335 | selection_seconds=14.2000 round_seconds=54.5000 pseudo_student_dice=0.8766 pseudo_ensemble_dice=0.8802 consistency_scale=1.0000 val_dice=0.8495 val_vi=0.0638 val_ece=0.0033 +[08:05:47][ 706s][AL] round 6: 1263/7715 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:05:57][ 715s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.858, uniform=0.236, top1=0.858, top2=0.858, top3=0.858 -> using mixture +[08:05:57][ 715s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (2466068 fg px): fly-mito=0.318, kasthuri=0.001, dros-vnc=0.022, multiclass=0.858 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:05:59][ 717s][CONS] round 6: on the 67 new patches student Dice=0.900 vs ensemble 0.906 -> pseudo-labels from the ensemble | consistency x1.00 +[08:06:06][ 725s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2219 +[08:06:13][ 732s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.1939 +[08:06:21][ 739s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.2065 +[08:06:28][ 746s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.2008 +[08:06:35][ 753s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.1909 +[08:06:37][ 755s][AL] round 6/6 | labeled=402 | selection_seconds=14.1000 round_seconds=61.5000 pseudo_student_dice=0.8996 pseudo_ensemble_dice=0.9064 consistency_scale=1.0000 val_dice=0.8659 val_vi=0.0671 val_ece=0.0041 +[08:06:44][ 762s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.1832 +[08:06:51][ 770s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.1824 +[08:06:59][ 777s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.1715 +[08:07:06][ 784s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.1809 +[08:07:13][ 791s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.1634 +[08:07:20][ 799s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1468 +[08:07:27][ 806s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1542 +[08:07:35][ 813s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1487 +[08:07:42][ 820s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1484 +[08:07:49][ 828s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1348 +[08:07:51][ 830s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=72.3000 val_dice=0.8838 val_vi=0.0574 val_ece=0.0035 +[08:07:51][ 830s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:08:15][ 853s][TEST] raw student (no EMA, no TTA): patch_dice=0.9160 patch_vi=0.0407 +[08:08:18][ 856s][TEST] final model without TTA: patch_dice=0.9160 patch_vi=0.0407 +[08:08:18][ 856s][RESOURCES] {"phases": {"setup": {"seconds": 0.57, "peak_cpu_rss_mb": 1027.4, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 15.38, "peak_cpu_rss_mb": 1800.5, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 299.26, "peak_cpu_rss_mb": 3684.9, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 512.34, "peak_cpu_rss_mb": 3684.9, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 26.73, "peak_cpu_rss_mb": 3684.9, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 65.5, "seconds_per_image": 0.535}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3684.9, "total_seconds": 854.3} +[08:08:18][ 856s][TEST] patch_dice=0.9242 patch_iou=0.8965 patch_precision=0.9617 patch_recall=0.9303 patch_vi=0.0388 image_dice=0.9577 image_iou=0.9193 image_precision=0.9739 image_recall=0.9427 image_vi=0.0432 cal_ece=0.0015 cal_nll=0.0160 cal_brier=0.0032 cal_mean_confidence=0.9974 cal_accuracy=0.9958 cal_predicted_prevalence=0.0438 cal_true_prevalence=0.0452 cal_mean_probability=0.0441 +[08:08:18][ 856s][DONE] lucchi-pp macs-v11-norestart-acq-mcdropout b5 s2 | test patch Dice=0.9242 VI=0.0388 | image Dice=0.9577 VI=0.0432 | 856s diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/result.json b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..ae161287b27a3280be156ba4e1f330554e11425c --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/result.json @@ -0,0 +1,647 @@ +{ + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 2, + "config": { + "task": "mitochondria", + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "mcdropout", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "lucchi-pp", + "--variants", + "macs-v11-norestart-acq-mcdropout", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "per_source": [ + 0.01302141100636151, + 0.021594387588085767, + 0.004953657914530629, + 0.960430543491022 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.001413550071584186, + "mmd2_mixture": 0.005039751415091165, + "mmd2_uniform": 0.2092225964739336, + "mmd2_single_source": [ + 0.5680514029229977, + 0.4644134839747518, + 0.6002496450435827, + 0.004698779437425582 + ], + "median_squared_distance": 0.18927357450912807, + "gammas": [ + 1.3208394285803657, + 2.6416788571607315, + 5.283357714321463, + 10.566715428642926, + 21.133430857285852 + ], + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6675364398352843, + 0.5899254221485877, + 0.8158221669373606, + 0.007664827003586616 + ], + "standardized": [ + 0.48001954758469056, + 0.22710032562990826, + 0.9632539109975666, + -1.6703737842121655 + ], + "tau": 0.5, + "max_weight": 0.960430543491022, + "entropy": 0.20441768713797392, + "effective_sources": 1.226810469408674, + "min_weight": 0.004953657914530629 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "multiclass", + "seconds": 1.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.243175358344919, + 0.16070093323689605, + 0.16243110642819403, + 0.8333930692567719 + ], + "ensemble_val_dice": { + "mixture": 0.8341160757031219, + "mixture@raw": 0.8334634835118209, + "mixture@logit": 0.8348662448964319, + "per_source": 0.833502984539706, + "per_source@raw": 0.833444326075437, + "per_source@logit": 0.8337659337218213, + "uniform": 0.33138683445274353, + "uniform@raw": 0.18373960615593887, + "uniform@logit": 0.2195153840155939 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "multiclass", + "weight_stats": { + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.2581988897471611, + "top_source": "multiclass", + "top_source_is_best_teacher": true + }, + "init_teacher": "multiclass", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 299.3, + "round0_val": { + "val_dice": 0.8779491435263782, + "val_vi": 0.06825049918780253, + "val_ece": 0.016595805602442455 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 67, + "selection_seconds": 26.4, + "round_seconds": 188.8, + "teacher_dice_labeled": [ + 0.2650592536077415, + 3.1912075848621877e-06, + 0.015534355861264863, + 0.9463032217749737 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 313095, + "evidence_nonempty_patches": 59, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9465589072209675, + "restart_teacher_dice": 0.9463032217749737, + "weights": [ + 3.893697301149704e-05, + 2.684582425110558e-06, + 1.5344296891974475e-06, + 0.9999568440148742 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9463032217749737, + "uniform": 0.3202334692828928, + "top1": 0.9463032217749737, + "top2": 0.9463032217749737, + "top3": 0.9463032217749737 + }, + "val_dice": 0.8414022702035394, + "val_vi": 0.08666243972725031, + "val_ece": 0.007021713555085711 + }, + { + "round": 2, + "n_labeled": 134, + "selection_seconds": 14.5, + "round_seconds": 33.4, + "teacher_dice_labeled": [ + 0.2991416012543912, + 1.2437284990435728e-06, + 0.020372806747900502, + 0.8375735217572351 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 803768, + "evidence_nonempty_patches": 122, + "first_correction": false, + "restart": "", + "weights": [ + 0.00016237894382911382, + 7.96192451671591e-06, + 4.776501177634436e-06, + 0.9998248826304765 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8375731881639311, + "uniform": 0.24281656424767087, + "top1": 0.8375735217572351, + "top2": 0.8375726432109015, + "top3": 0.8375731881639311 + }, + "pseudo_student_dice": 0.7120064965637388, + "pseudo_ensemble_dice": 0.7619411679044706, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8597648015177228, + "val_vi": 0.0658964051897875, + "val_ece": 0.002754811178931108 + }, + { + "round": 3, + "n_labeled": 201, + "selection_seconds": 14.5, + "round_seconds": 40.8, + "teacher_dice_labeled": [ + 0.2891765445671489, + 8.506113769271664e-07, + 0.015998550871162337, + 0.8295804278721998 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1174472, + "evidence_nonempty_patches": 187, + "first_correction": false, + "restart": "", + "weights": [ + 0.000159208627892964, + 8.624440437008878e-06, + 4.952532290120253e-06, + 0.9998272143993799 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8295808091398001, + "uniform": 0.2343069632818933, + "top1": 0.8295804278721998, + "top2": 0.8295808091398001, + "top3": 0.8295808091398001 + }, + "pseudo_student_dice": 0.8224789860540216, + "pseudo_ensemble_dice": 0.813354522180919, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8553512752213046, + "val_vi": 0.06884337157122646, + "val_ece": 0.003812648781613399 + }, + { + "round": 4, + "n_labeled": 268, + "selection_seconds": 14.4, + "round_seconds": 47.6, + "teacher_dice_labeled": [ + 0.31161893544105346, + 5.990326820250659e-07, + 0.019232521726163354, + 0.8404083974629101 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1668205, + "evidence_nonempty_patches": 254, + "first_correction": false, + "restart": "", + "weights": [ + 0.00017881313459032123, + 7.739209200375557e-06, + 4.590278381572685e-06, + 0.9998088573778278 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8404082089621864, + "uniform": 0.2457004173271107, + "top1": 0.8404083974629101, + "top2": 0.840407686240947, + "top3": 0.840407686240947 + }, + "pseudo_student_dice": 0.8279592216197965, + "pseudo_ensemble_dice": 0.8673621638397173, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8801558349146971, + "val_vi": 0.06363229266456624, + "val_ece": 0.0034598149472670455 + }, + { + "round": 5, + "n_labeled": 335, + "selection_seconds": 14.2, + "round_seconds": 54.5, + "teacher_dice_labeled": [ + 0.3115826019402249, + 4.813431398975702e-07, + 0.016249049498497304, + 0.8483931630389719 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 2074752, + "evidence_nonempty_patches": 318, + "first_correction": false, + "restart": "", + "weights": [ + 0.0001650329552051454, + 7.145376704143948e-06, + 4.113495473807198e-06, + 0.9998237081726169 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8483934043542636, + "uniform": 0.2347708701382939, + "top1": 0.8483931630389719, + "top2": 0.8483932688493395, + "top3": 0.8483931934853834 + }, + "pseudo_student_dice": 0.876590868042218, + "pseudo_ensemble_dice": 0.880176624817099, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8494713601133347, + "val_vi": 0.06376301465827551, + "val_ece": 0.0033383623519546305 + }, + { + "round": 6, + "n_labeled": 402, + "selection_seconds": 14.1, + "round_seconds": 61.5, + "teacher_dice_labeled": [ + 0.3184675597053298, + 0.0009056687495293519, + 0.022052554111966854, + 0.8580474733941708 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 2466068, + "evidence_nonempty_patches": 382, + "first_correction": false, + "restart": "", + "weights": [ + 0.00016052615972160096, + 6.546820846704229e-06, + 3.9581246961340295e-06, + 0.9998289688947355 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8580478851737139, + "uniform": 0.23598459748237136, + "top1": 0.8580474733941708, + "top2": 0.8580477073846513, + "top3": 0.8580477073846513 + }, + "pseudo_student_dice": 0.8995743836225631, + "pseudo_ensemble_dice": 0.9064270163359456, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.865923287490848, + "val_vi": 0.0670894976457988, + "val_ece": 0.0041427219033347016 + }, + { + "round": 7, + "n_labeled": 402, + "selection_seconds": 0.0, + "round_seconds": 72.3, + "final_consolidation": true, + "val_dice": 0.8838258335603796, + "val_vi": 0.05742677148427432, + "val_ece": 0.003537881529310446 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.0970013558496726, + "auroc_top20_error": 0.47497646484375 + }, + "al_seconds": 512.3, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 0.00016052615972160096, + 6.546820846704229e-06, + 3.9581246961340295e-06, + 0.9998289688947355 + ], + "weight_stats_final": { + "max_weight": 0.9998289688947355, + "entropy": 0.0017009265195500958, + "effective_sources": 1.0017023739155844, + "min_weight": 3.9581246961340295e-06 + }, + "top_source_final": "multiclass", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.9159543672949144, + "patch_iou": 0.8859094915316439, + "patch_precision": 0.9479850456236946, + "patch_recall": 0.9313137274179656, + "patch_vi": 0.040666403131386315, + "n_patches": 1750, + "cal_ece": 0.0023215560856687115, + "cal_nll": 0.019756787038157805, + "cal_brier": 0.0036058551469715385, + "cal_mean_confidence": 0.9979198458061832, + "cal_accuracy": 0.9956099330357143, + "cal_predicted_prevalence": 0.04404966517857143, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.044163917420721545 + }, + "test_notta": { + "patch_dice": 0.9159543672949144, + "patch_iou": 0.8859094915316439, + "patch_precision": 0.9479850456236946, + "patch_recall": 0.9313137274179656, + "patch_vi": 0.040666403131386315, + "n_patches": 1750, + "cal_ece": 0.0023215560856687115, + "cal_nll": 0.019756787038157805, + "cal_brier": 0.0036058551469715385, + "cal_mean_confidence": 0.9979198458061832, + "cal_accuracy": 0.9956099330357143, + "cal_predicted_prevalence": 0.04404966517857143, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.044163917420721545 + }, + "test": { + "patch_dice": 0.9241743149650631, + "patch_iou": 0.8964802182390087, + "patch_precision": 0.9617294751198411, + "patch_recall": 0.93032266511686, + "patch_vi": 0.03876481848693228, + "n_patches": 1750, + "image_dice": 0.9577395782882179, + "image_iou": 0.9192747648238735, + "image_precision": 0.973896041335226, + "image_recall": 0.9427052516535608, + "image_vi": 0.043222378415404016, + "n_images": 50, + "cal_ece": 0.001525377707148676, + "cal_nll": 0.01597899841833155, + "cal_brier": 0.0032463206934083876, + "cal_mean_confidence": 0.9973747080642915, + "cal_accuracy": 0.9958493303571428, + "cal_predicted_prevalence": 0.04377901785714286, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.044063124169567786 + }, + "eval_seconds": 26.7, + "total_seconds": 856.4, + "resources": { + "phases": { + "setup": { + "seconds": 0.57, + "peak_cpu_rss_mb": 1027.4, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 15.38, + "peak_cpu_rss_mb": 1800.5, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 299.26, + "peak_cpu_rss_mb": 3684.9, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 512.34, + "peak_cpu_rss_mb": 3684.9, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4176.0, + "rounds": 6 + }, + "inference": { + "seconds": 26.73, + "peak_cpu_rss_mb": 3684.9, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4176.0, + "patches": 1750, + "images": 50, + "patches_per_second": 65.5, + "seconds_per_image": 0.535 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3684.9, + "total_seconds": 854.3 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-mcdropout_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..f37f959a2adb435e180e7bc576b3a19c15a5eeb6 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,67,26.4,188.8,"[0.2650592536077415, 3.1912075848621877e-06, 0.015534355861264863, 0.9463032217749737]",,313095,59,True,kept,0.9465589072209675,0.9463032217749737,"[3.893697301149704e-05, 2.684582425110558e-06, 1.5344296891974475e-06, 0.9999568440148742]",mixture,"{'mixture': 0.9463032217749737, 'uniform': 0.3202334692828928, 'top1': 0.9463032217749737, 'top2': 0.9463032217749737, 'top3': 0.9463032217749737}",0.8414022702035394,0.08666243972725031,0.007021713555085711,,,,, +2,134,14.5,33.4,"[0.2991416012543912, 1.2437284990435728e-06, 0.020372806747900502, 0.8375735217572351]",,803768,122,False,,,,"[0.00016237894382911382, 7.96192451671591e-06, 4.776501177634436e-06, 0.9998248826304765]",mixture,"{'mixture': 0.8375731881639311, 'uniform': 0.24281656424767087, 'top1': 0.8375735217572351, 'top2': 0.8375726432109015, 'top3': 0.8375731881639311}",0.8597648015177228,0.0658964051897875,0.002754811178931108,0.7120064965637388,0.7619411679044706,ensemble,1.0, +3,201,14.5,40.8,"[0.2891765445671489, 8.506113769271664e-07, 0.015998550871162337, 0.8295804278721998]",,1174472,187,False,,,,"[0.000159208627892964, 8.624440437008878e-06, 4.952532290120253e-06, 0.9998272143993799]",mixture,"{'mixture': 0.8295808091398001, 'uniform': 0.2343069632818933, 'top1': 0.8295804278721998, 'top2': 0.8295808091398001, 'top3': 0.8295808091398001}",0.8553512752213046,0.06884337157122646,0.003812648781613399,0.8224789860540216,0.813354522180919,ensemble,1.0, +4,268,14.4,47.6,"[0.31161893544105346, 5.990326820250659e-07, 0.019232521726163354, 0.8404083974629101]",,1668205,254,False,,,,"[0.00017881313459032123, 7.739209200375557e-06, 4.590278381572685e-06, 0.9998088573778278]",mixture,"{'mixture': 0.8404082089621864, 'uniform': 0.2457004173271107, 'top1': 0.8404083974629101, 'top2': 0.840407686240947, 'top3': 0.840407686240947}",0.8801558349146971,0.06363229266456624,0.0034598149472670455,0.8279592216197965,0.8673621638397173,ensemble,1.0, +5,335,14.2,54.5,"[0.3115826019402249, 4.813431398975702e-07, 0.016249049498497304, 0.8483931630389719]",,2074752,318,False,,,,"[0.0001650329552051454, 7.145376704143948e-06, 4.113495473807198e-06, 0.9998237081726169]",mixture,"{'mixture': 0.8483934043542636, 'uniform': 0.2347708701382939, 'top1': 0.8483931630389719, 'top2': 0.8483932688493395, 'top3': 0.8483931934853834}",0.8494713601133347,0.06376301465827551,0.0033383623519546305,0.876590868042218,0.880176624817099,ensemble,1.0, +6,402,14.1,61.5,"[0.3184675597053298, 0.0009056687495293519, 0.022052554111966854, 0.8580474733941708]",,2466068,382,False,,,,"[0.00016052615972160096, 6.546820846704229e-06, 3.9581246961340295e-06, 0.9998289688947355]",mixture,"{'mixture': 0.8580478851737139, 'uniform': 0.23598459748237136, 'top1': 0.8580474733941708, 'top2': 0.8580477073846513, 'top3': 0.8580477073846513}",0.865923287490848,0.0670894976457988,0.0041427219033347016,0.8995743836225631,0.9064270163359456,ensemble,1.0, +7,402,0.0,72.3,,,,,,,,,,,,0.8838258335603796,0.05742677148427432,0.003537881529310446,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/log.txt b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..3f8ce251cc8e9e400372298bc144116ae5744fd8 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/log.txt @@ -0,0 +1,105 @@ +[12:14:09][ 0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-mcdropout budget=5.0% seed=3 | {"task": "mitochondria", "target": "lucchi-pp", "variant": "macs-v11-norestart-acq-mcdropout", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "mcdropout", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[12:14:09][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[12:14:09][ 0s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches +[12:14:11][ 2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60 +[12:14:11][ 2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[12:14:12][ 4s][WEIGHTS] mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 | eff.sources=1.44 | MMD2 mix=0.0050 uniform=0.2092 +[12:14:12][ 4s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass +[12:14:36][ 27s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833 +[12:14:36][ 27s][TEACHERS] ensemble target-val Dice: mixture=0.834, mixture@raw=0.833, mixture@logit=0.835, per_source=0.834, per_source@raw=0.833, per_source@logit=0.834, uniform=0.331, uniform@raw=0.184, uniform@logit=0.220 +[12:14:36][ 27s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 +[12:14:36][ 27s][INIT] student initialised from teacher multiclass (weight 0.920) +[12:14:38][ 29s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[12:14:38][ 29s][KD] distilling the mixture ensemble (prob space) into the student +[12:15:42][ 93s][KD] soft labels for 6000 patches from 4 teachers in 64s +[12:17:30][ 201s][KD] epoch 1/5 | patches=6000 | loss=0.1181 | 108s +[12:19:18][ 309s][KD] epoch 2/5 | patches=6000 | loss=0.1061 | 108s +[12:21:05][ 416s][KD] epoch 3/5 | patches=6000 | loss=0.1045 | 107s +[12:22:52][ 524s][KD] epoch 4/5 | patches=6000 | loss=0.1039 | 107s +[12:24:41][ 632s][KD] epoch 5/5 | patches=6000 | loss=0.1027 | 108s +[12:24:44][ 635s][ROUND0] before any target label: val_dice=0.8788 val_vi=0.0705 +[12:24:44][ 635s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=mcdropout | first_round=stratified +[12:25:16][ 667s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.06667865844265394, 'auroc_top20_error': 0.46323427734375} +[12:25:40][ 691s][AL] round 1: stratified cold start: 34 patches from the membrane-rich half (ensemble prevalence >= 0.073) + 33 from the membrane-poor half of 2000 candidates +[12:25:44][ 696s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.931, uniform=0.391, top1=0.931, top2=0.931, top3=0.931 -> using mixture +[12:25:45][ 696s][WEIGHTS] round 1: student kept (labeled Dice 0.930 vs best teacher multiclass 0.931); distilling the corrected ensemble into it +[12:25:45][ 696s][WEIGHTS] round 1: teacher Dice on 67 labeled patches + 0 blank (327320 fg px): fly-mito=0.283, kasthuri=0.000, dros-vnc=0.014, multiclass=0.931 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[12:26:56][ 767s][KD] soft labels for 6000 patches from 4 teachers in 71s +[12:28:43][ 874s][KD] epoch 1/2 | patches=6000 | loss=0.0482 | 107s +[12:30:30][ 981s][KD] epoch 2/2 | patches=6000 | loss=0.0457 | 107s +[12:30:54][ 1006s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 25s +[12:30:57][ 1008s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3079 +[12:31:00][ 1011s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3405 +[12:31:03][ 1014s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3642 +[12:31:05][ 1016s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.3217 +[12:31:08][ 1019s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.2970 +[12:31:12][ 1023s][AL] round 1/6 | labeled=67 | selection_seconds=56.2000 round_seconds=383.9000 restart_student_dice=0.9295 restart_teacher_dice=0.9312 val_dice=0.8475 val_vi=0.0802 val_ece=0.0046 +[12:31:43][ 1054s][AL] round 2: 1299/7983 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[12:31:51][ 1062s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.858, uniform=0.214, top1=0.858, top2=0.858, top3=0.858 -> using mixture +[12:31:51][ 1062s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (824224 fg px): fly-mito=0.319, kasthuri=0.000, dros-vnc=0.008, multiclass=0.858 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[12:31:54][ 1065s][CONS] round 2: on the 67 new patches student Dice=0.732 vs ensemble 0.805 -> pseudo-labels from the ensemble | consistency x1.00 +[12:31:59][ 1070s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.4353 +[12:32:04][ 1075s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.3688 +[12:32:09][ 1080s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.3547 +[12:32:14][ 1085s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.2975 +[12:32:19][ 1090s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.3078 +[12:32:22][ 1093s][AL] round 2/6 | labeled=134 | selection_seconds=31.6000 round_seconds=67.4000 pseudo_student_dice=0.7319 pseudo_ensemble_dice=0.8052 consistency_scale=1.0000 val_dice=0.8462 val_vi=0.0704 val_ece=0.0027 +[12:32:52][ 1124s][AL] round 3: 1195/7916 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[12:33:04][ 1135s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.854, uniform=0.192, top1=0.854, top2=0.854, top3=0.854 -> using mixture +[12:33:04][ 1135s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (1176437 fg px): fly-mito=0.293, kasthuri=0.000, dros-vnc=0.006, multiclass=0.854 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[12:33:07][ 1138s][CONS] round 3: on the 67 new patches student Dice=0.840 vs ensemble 0.846 -> pseudo-labels from the ensemble | consistency x1.00 +[12:33:14][ 1146s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.3243 +[12:33:22][ 1153s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.3314 +[12:33:29][ 1161s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.3061 +[12:33:37][ 1168s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.2799 +[12:33:45][ 1176s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2583 +[12:33:48][ 1179s][AL] round 3/6 | labeled=201 | selection_seconds=30.3000 round_seconds=82.2000 pseudo_student_dice=0.8403 pseudo_ensemble_dice=0.8462 consistency_scale=1.0000 val_dice=0.7817 val_vi=0.0878 val_ece=0.0063 +[12:34:19][ 1210s][AL] round 4: 729/7849 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[12:34:34][ 1225s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.850, uniform=0.207, top1=0.850, top2=0.850, top3=0.850 -> using mixture +[12:34:34][ 1225s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1485887 fg px): fly-mito=0.292, kasthuri=0.001, dros-vnc=0.037, multiclass=0.850 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[12:34:37][ 1228s][CONS] round 4: on the 67 new patches student Dice=0.812 vs ensemble 0.832 -> pseudo-labels from the ensemble | consistency x1.00 +[12:34:47][ 1238s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.4857 +[12:34:57][ 1248s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.3265 +[12:35:06][ 1257s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.3067 +[12:35:16][ 1267s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.2821 +[12:35:26][ 1277s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.2654 +[12:35:29][ 1280s][AL] round 4/6 | labeled=268 | selection_seconds=31.1000 round_seconds=97.5000 pseudo_student_dice=0.8125 pseudo_ensemble_dice=0.8323 consistency_scale=1.0000 val_dice=0.8644 val_vi=0.0646 val_ece=0.0035 +[12:36:00][ 1311s][AL] round 5: 1253/7782 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[12:36:19][ 1330s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.849, uniform=0.199, top1=0.849, top2=0.849, top3=0.849 -> using mixture +[12:36:19][ 1330s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (1885968 fg px): fly-mito=0.292, kasthuri=0.001, dros-vnc=0.031, multiclass=0.849 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[12:36:27][ 1339s][CONS] round 5: on the 67 new patches student Dice=0.864 vs ensemble 0.847 -> pseudo-labels from the student | consistency x1.00 +[12:36:40][ 1351s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2631 +[12:36:52][ 1363s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.2587 +[12:37:04][ 1375s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.2519 +[12:37:16][ 1387s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.2552 +[12:37:27][ 1398s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.2394 +[12:37:29][ 1400s][AL] round 5/6 | labeled=335 | selection_seconds=30.9000 round_seconds=117.8000 pseudo_student_dice=0.8637 pseudo_ensemble_dice=0.8470 consistency_scale=1.0000 val_dice=0.8809 val_vi=0.0607 val_ece=0.0033 +[12:37:59][ 1430s][AL] round 6: 1283/7715 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[12:38:57][ 1488s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.277, kasthuri=0.001, dros-vnc=0.021, multiclass=0.953 | blended with labeled Dice at 0.50 +[12:40:07][ 1558s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.905, uniform=0.273, top1=0.905, top2=0.905, top3=0.905 -> using mixture +[12:40:07][ 1558s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (2293597 fg px): fly-mito=0.286, kasthuri=0.001, dros-vnc=0.024, multiclass=0.905 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[12:40:16][ 1567s][CONS] round 6: on the 67 new patches student Dice=0.894 vs ensemble 0.890 -> pseudo-labels from the student | consistency x1.00 +[12:40:29][ 1581s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2251 +[12:40:44][ 1595s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.2193 +[12:40:59][ 1610s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.2200 +[12:41:14][ 1625s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.2212 +[12:41:28][ 1639s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.2214 +[12:41:32][ 1643s][AL] round 6/6 | labeled=402 | selection_seconds=29.5000 round_seconds=238.9000 pseudo_student_dice=0.8943 pseudo_ensemble_dice=0.8899 consistency_scale=1.0000 val_dice=0.8668 val_vi=0.0639 val_ece=0.0037 +[12:41:38][ 1649s][CONS] final consolidation: pseudo-labels refreshed from the student +[12:41:53][ 1664s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.2495 +[12:42:07][ 1678s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.2027 +[12:42:22][ 1693s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.2137 +[12:42:37][ 1708s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.2033 +[12:42:51][ 1722s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.1573 +[12:43:05][ 1736s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1593 +[12:43:20][ 1751s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1504 +[12:43:34][ 1765s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1424 +[12:43:49][ 1780s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1397 +[12:44:04][ 1795s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1352 +[12:44:07][ 1798s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=152.0000 val_dice=0.8730 val_vi=0.0597 val_ece=0.0042 +[12:44:07][ 1798s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[12:44:56][ 1847s][TEST] raw student (no EMA, no TTA): patch_dice=0.9228 patch_vi=0.0406 +[12:45:02][ 1853s][TEST] final model without TTA: patch_dice=0.9228 patch_vi=0.0406 +[12:45:02][ 1853s][RESOURCES] {"phases": {"setup": {"seconds": 1.71, "peak_cpu_rss_mb": 1027.4, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 27.44, "peak_cpu_rss_mb": 1801.0, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 602.38, "peak_cpu_rss_mb": 3684.8, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 1163.12, "peak_cpu_rss_mb": 3684.8, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 54.56, "peak_cpu_rss_mb": 3684.8, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 32.1, "seconds_per_image": 1.091}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3684.8, "total_seconds": 1849.2} +[12:45:02][ 1853s][TEST] patch_dice=0.9331 patch_iou=0.9060 patch_precision=0.9689 patch_recall=0.9335 patch_vi=0.0388 image_dice=0.9581 image_iou=0.9202 image_precision=0.9711 image_recall=0.9462 image_vi=0.0432 cal_ece=0.0018 cal_nll=0.0158 cal_brier=0.0032 cal_mean_confidence=0.9977 cal_accuracy=0.9958 cal_predicted_prevalence=0.0438 cal_true_prevalence=0.0452 cal_mean_probability=0.0440 +[12:45:02][ 1853s][DONE] lucchi-pp macs-v11-norestart-acq-mcdropout b5 s3 | test patch Dice=0.9331 VI=0.0388 | image Dice=0.9581 VI=0.0432 | 1853s diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/result.json b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..7e8f9255efb4de525f8c7b720c8b520bcdfc2db2 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/result.json @@ -0,0 +1,652 @@ +{ + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 3, + "config": { + "task": "mitochondria", + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-mcdropout", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "mcdropout", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "lucchi-pp", + "--variants", + "macs-v11-norestart-acq-mcdropout", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "per_source": [ + 0.01302141100636151, + 0.021594387588085767, + 0.004953657914530629, + 0.960430543491022 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.001413550071584186, + "mmd2_mixture": 0.005039751415091165, + "mmd2_uniform": 0.2092225964739336, + "mmd2_single_source": [ + 0.5680514029229977, + 0.4644134839747518, + 0.6002496450435827, + 0.004698779437425582 + ], + "median_squared_distance": 0.18927357450912807, + "gammas": [ + 1.3208394285803657, + 2.6416788571607315, + 5.283357714321463, + 10.566715428642926, + 21.133430857285852 + ], + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6675364398352843, + 0.5899254221485877, + 0.8158221669373606, + 0.007664827003586616 + ], + "standardized": [ + 0.48001954758469056, + 0.22710032562990826, + 0.9632539109975666, + -1.6703737842121655 + ], + "tau": 0.5, + "max_weight": 0.960430543491022, + "entropy": 0.20441768713797392, + "effective_sources": 1.226810469408674, + "min_weight": 0.004953657914530629 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "multiclass", + "seconds": 1.6 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.243175358344919, + 0.16070093323689605, + 0.16243110642819403, + 0.8333930692567719 + ], + "ensemble_val_dice": { + "mixture": 0.8341160757031219, + "mixture@raw": 0.8334634835118209, + "mixture@logit": 0.8348662448964319, + "per_source": 0.833502984539706, + "per_source@raw": 0.833444326075437, + "per_source@logit": 0.8337659337218213, + "uniform": 0.33138683445274353, + "uniform@raw": 0.18373960615593887, + "uniform@logit": 0.2195153840155939 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "multiclass", + "weight_stats": { + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.2581988897471611, + "top_source": "multiclass", + "top_source_is_best_teacher": true + }, + "init_teacher": "multiclass", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 602.4, + "round0_val": { + "val_dice": 0.8788253265325645, + "val_vi": 0.07051344125492959, + "val_ece": 0.017125493497017343 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 67, + "selection_seconds": 56.2, + "round_seconds": 383.9, + "teacher_dice_labeled": [ + 0.2829946467808237, + 3.0551049275787378e-06, + 0.014406917734236871, + 0.931227563092392 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 327320, + "evidence_nonempty_patches": 60, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9295241436152791, + "restart_teacher_dice": 0.931227563092392, + "weights": [ + 5.4165213515608284e-05, + 3.121346642323861e-06, + 1.7640725486920737e-06, + 0.9999409493672934 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9312292062554963, + "uniform": 0.39082248145471704, + "top1": 0.931227563092392, + "top2": 0.9312292062554963, + "top3": 0.9312292062554963 + }, + "val_dice": 0.847523785507917, + "val_vi": 0.08016155524257361, + "val_ece": 0.004602505492721512 + }, + { + "round": 2, + "n_labeled": 134, + "selection_seconds": 31.6, + "round_seconds": 67.4, + "teacher_dice_labeled": [ + 0.3187275607286123, + 1.2132609420971215e-06, + 0.008457843219879845, + 0.85773634808294 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 824224, + "evidence_nonempty_patches": 124, + "first_correction": false, + "restart": "", + "weights": [ + 0.0001614455272518443, + 6.5080890653730986e-06, + 3.4657674949020854e-06, + 0.9998285806161878 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8577374450649478, + "uniform": 0.21395039355170267, + "top1": 0.85773634808294, + "top2": 0.8577374450649478, + "top3": 0.8577374450649478 + }, + "pseudo_student_dice": 0.731878279118573, + "pseudo_ensemble_dice": 0.8051827689587958, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8462201174958578, + "val_vi": 0.07035304668908086, + "val_ece": 0.002708612694885825 + }, + { + "round": 3, + "n_labeled": 201, + "selection_seconds": 30.3, + "round_seconds": 82.2, + "teacher_dice_labeled": [ + 0.2929894860082066, + 8.500235456522146e-07, + 0.006009219138776959, + 0.8541647416464438 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1176437, + "evidence_nonempty_patches": 187, + "first_correction": false, + "restart": "", + "weights": [ + 0.00012935168472274415, + 6.744923376034194e-06, + 3.505018342197696e-06, + 0.9998603983735591 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8541656350800747, + "uniform": 0.1918509200066622, + "top1": 0.8541647416464438, + "top2": 0.8541657640291969, + "top3": 0.8541657640291969 + }, + "pseudo_student_dice": 0.8403092826286882, + "pseudo_ensemble_dice": 0.8462284823598807, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.7817143038202493, + "val_vi": 0.08781561432411808, + "val_ece": 0.006329736730954452 + }, + { + "round": 4, + "n_labeled": 268, + "selection_seconds": 31.1, + "round_seconds": 97.5, + "teacher_dice_labeled": [ + 0.2915549596385615, + 0.0009082413999421845, + 0.03713796814407696, + 0.8496914471202142 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1485887, + "evidence_nonempty_patches": 245, + "first_correction": false, + "restart": "", + "weights": [ + 0.0001333419327372865, + 7.117746167166703e-06, + 5.003863560713537e-06, + 0.9998545364575349 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.8496925731279343, + "uniform": 0.20705462550715772, + "top1": 0.8496914471202142, + "top2": 0.8496923800690132, + "top3": 0.8496922743708712 + }, + "pseudo_student_dice": 0.812457729519864, + "pseudo_ensemble_dice": 0.8323162196003799, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8643962604052019, + "val_vi": 0.06456911124642525, + "val_ece": 0.003466241159430552 + }, + { + "round": 5, + "n_labeled": 335, + "selection_seconds": 30.9, + "round_seconds": 117.8, + "teacher_dice_labeled": [ + 0.2924712263073467, + 0.0007157346664046015, + 0.030554611538117355, + 0.8491022307207292 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1885968, + "evidence_nonempty_patches": 309, + "first_correction": false, + "restart": "", + "weights": [ + 0.00013536432169718426, + 7.146026461845644e-06, + 4.712729087531835e-06, + 0.9998527769227535 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.849103111847891, + "uniform": 0.19850537132181223, + "top1": 0.8491022307207292, + "top2": 0.8491032776747063, + "top3": 0.8491031947613442 + }, + "pseudo_student_dice": 0.8636739526045385, + "pseudo_ensemble_dice": 0.8469980170249343, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8809494576309032, + "val_vi": 0.060736778674735564, + "val_ece": 0.003287415663818985 + }, + { + "round": 6, + "n_labeled": 402, + "selection_seconds": 29.5, + "round_seconds": 238.9, + "teacher_dice_labeled": [ + 0.2859606321045594, + 0.000561505519584998, + 0.023832370589920458, + 0.9046321418384846 + ], + "teacher_dice_pool": [ + 0.2771804874200709, + 0.0005343913110417282, + 0.020549285203589812, + 0.9527536435042196 + ], + "evidence_fg_pixels": 2293597, + "evidence_nonempty_patches": 376, + "first_correction": false, + "restart": "", + "weights": [ + 7.279393071052499e-05, + 4.095066157160319e-06, + 2.5289734172532974e-06, + 0.999920582029715 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9046330175892203, + "uniform": 0.2725112860333749, + "top1": 0.9046321418384846, + "top2": 0.9046329047608885, + "top3": 0.9046330254372459 + }, + "pseudo_student_dice": 0.8942960845899264, + "pseudo_ensemble_dice": 0.8898612770230331, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8668335966062999, + "val_vi": 0.06392912241937598, + "val_ece": 0.0036721664900155205 + }, + { + "round": 7, + "n_labeled": 402, + "selection_seconds": 0.0, + "round_seconds": 152.0, + "final_consolidation": true, + "val_dice": 0.8730342771833822, + "val_vi": 0.059741696319020904, + "val_ece": 0.00423163155309635 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.06667865844265394, + "auroc_top20_error": 0.46323427734375 + }, + "al_seconds": 1163.1, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 7.279393071052499e-05, + 4.095066157160319e-06, + 2.5289734172532974e-06, + 0.999920582029715 + ], + "weight_stats_final": { + "max_weight": 0.999920582029715, + "entropy": 0.0008563814247464127, + "effective_sources": 1.000856748224018, + "min_weight": 2.5289734172532974e-06 + }, + "top_source_final": "multiclass", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.9227699873094142, + "patch_iou": 0.8939099422884038, + "patch_precision": 0.9530975840573026, + "patch_recall": 0.9338041354157988, + "patch_vi": 0.04062597837287667, + "n_patches": 1750, + "cal_ece": 0.0026727696070865917, + "cal_nll": 0.019863818664560946, + "cal_brier": 0.0036270226855356896, + "cal_mean_confidence": 0.9982468530745331, + "cal_accuracy": 0.9955881696428571, + "cal_predicted_prevalence": 0.04405691964285714, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.044002080964898924 + }, + "test_notta": { + "patch_dice": 0.9227699873094142, + "patch_iou": 0.8939099422884038, + "patch_precision": 0.9530975840573026, + "patch_recall": 0.9338041354157988, + "patch_vi": 0.04062597837287667, + "n_patches": 1750, + "cal_ece": 0.0026727696070865917, + "cal_nll": 0.019863818664560946, + "cal_brier": 0.0036270226855356896, + "cal_mean_confidence": 0.9982468530745331, + "cal_accuracy": 0.9955881696428571, + "cal_predicted_prevalence": 0.04405691964285714, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.044002080964898924 + }, + "test": { + "patch_dice": 0.9331409498369507, + "patch_iou": 0.9059920299864583, + "patch_precision": 0.9688578427053718, + "patch_recall": 0.9334537124749891, + "patch_vi": 0.038759490047539105, + "n_patches": 1750, + "image_dice": 0.9581429518610457, + "image_iou": 0.9201689422484445, + "image_precision": 0.971103163202821, + "image_recall": 0.9461607759226698, + "image_vi": 0.04315136295420089, + "n_images": 50, + "cal_ece": 0.0018331411229110803, + "cal_nll": 0.015842480371836876, + "cal_brier": 0.0032433324089357154, + "cal_mean_confidence": 0.9976729848729111, + "cal_accuracy": 0.99583984375, + "cal_predicted_prevalence": 0.04379520089285714, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04400162847343526 + }, + "eval_seconds": 54.6, + "total_seconds": 1852.9, + "resources": { + "phases": { + "setup": { + "seconds": 1.71, + "peak_cpu_rss_mb": 1027.4, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 27.44, + "peak_cpu_rss_mb": 1801.0, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 602.38, + "peak_cpu_rss_mb": 3684.8, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 1163.12, + "peak_cpu_rss_mb": 3684.8, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4176.0, + "rounds": 6 + }, + "inference": { + "seconds": 54.56, + "peak_cpu_rss_mb": 3684.8, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4176.0, + "patches": 1750, + "images": 50, + "patches_per_second": 32.1, + "seconds_per_image": 1.091 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3684.8, + "total_seconds": 1849.2 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-mcdropout_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..076b02f0e8c23bedef2fe13f71647b36874c758e --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-mcdropout/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,67,56.2,383.9,"[0.2829946467808237, 3.0551049275787378e-06, 0.014406917734236871, 0.931227563092392]",,327320,60,True,kept,0.9295241436152791,0.931227563092392,"[5.4165213515608284e-05, 3.121346642323861e-06, 1.7640725486920737e-06, 0.9999409493672934]",mixture,"{'mixture': 0.9312292062554963, 'uniform': 0.39082248145471704, 'top1': 0.931227563092392, 'top2': 0.9312292062554963, 'top3': 0.9312292062554963}",0.847523785507917,0.08016155524257361,0.004602505492721512,,,,, +2,134,31.6,67.4,"[0.3187275607286123, 1.2132609420971215e-06, 0.008457843219879845, 0.85773634808294]",,824224,124,False,,,,"[0.0001614455272518443, 6.5080890653730986e-06, 3.4657674949020854e-06, 0.9998285806161878]",mixture,"{'mixture': 0.8577374450649478, 'uniform': 0.21395039355170267, 'top1': 0.85773634808294, 'top2': 0.8577374450649478, 'top3': 0.8577374450649478}",0.8462201174958578,0.07035304668908086,0.002708612694885825,0.731878279118573,0.8051827689587958,ensemble,1.0, +3,201,30.3,82.2,"[0.2929894860082066, 8.500235456522146e-07, 0.006009219138776959, 0.8541647416464438]",,1176437,187,False,,,,"[0.00012935168472274415, 6.744923376034194e-06, 3.505018342197696e-06, 0.9998603983735591]",mixture,"{'mixture': 0.8541656350800747, 'uniform': 0.1918509200066622, 'top1': 0.8541647416464438, 'top2': 0.8541657640291969, 'top3': 0.8541657640291969}",0.7817143038202493,0.08781561432411808,0.006329736730954452,0.8403092826286882,0.8462284823598807,ensemble,1.0, +4,268,31.1,97.5,"[0.2915549596385615, 0.0009082413999421845, 0.03713796814407696, 0.8496914471202142]",,1485887,245,False,,,,"[0.0001333419327372865, 7.117746167166703e-06, 5.003863560713537e-06, 0.9998545364575349]",mixture,"{'mixture': 0.8496925731279343, 'uniform': 0.20705462550715772, 'top1': 0.8496914471202142, 'top2': 0.8496923800690132, 'top3': 0.8496922743708712}",0.8643962604052019,0.06456911124642525,0.003466241159430552,0.812457729519864,0.8323162196003799,ensemble,1.0, +5,335,30.9,117.8,"[0.2924712263073467, 0.0007157346664046015, 0.030554611538117355, 0.8491022307207292]",,1885968,309,False,,,,"[0.00013536432169718426, 7.146026461845644e-06, 4.712729087531835e-06, 0.9998527769227535]",mixture,"{'mixture': 0.849103111847891, 'uniform': 0.19850537132181223, 'top1': 0.8491022307207292, 'top2': 0.8491032776747063, 'top3': 0.8491031947613442}",0.8809494576309032,0.060736778674735564,0.003287415663818985,0.8636739526045385,0.8469980170249343,student,1.0, +6,402,29.5,238.9,"[0.2859606321045594, 0.000561505519584998, 0.023832370589920458, 0.9046321418384846]","[0.2771804874200709, 0.0005343913110417282, 0.020549285203589812, 0.9527536435042196]",2293597,376,False,,,,"[7.279393071052499e-05, 4.095066157160319e-06, 2.5289734172532974e-06, 0.999920582029715]",mixture,"{'mixture': 0.9046330175892203, 'uniform': 0.2725112860333749, 'top1': 0.9046321418384846, 'top2': 0.9046329047608885, 'top3': 0.9046330254372459}",0.8668335966062999,0.06392912241937598,0.0036721664900155205,0.8942960845899264,0.8898612770230331,student,1.0, +7,402,0.0,152.0,,,,,,,,,,,,0.8730342771833822,0.059741696319020904,0.00423163155309635,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..90d60f0dae57875b989c0b975863ba69c80d8f8a --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/log.txt @@ -0,0 +1,98 @@ +[08:08:20][ 0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-random budget=5.0% seed=2 | {"task": "mitochondria", "target": "lucchi-pp", "variant": "macs-v11-norestart-acq-random", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "random", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 0, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[08:08:20][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[08:08:20][ 0s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches +[08:08:21][ 1s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60 +[08:08:21][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[08:08:22][ 2s][WEIGHTS] mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 | eff.sources=1.44 | MMD2 mix=0.0050 uniform=0.2092 +[08:08:22][ 2s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass +[08:08:35][ 15s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833 +[08:08:35][ 15s][TEACHERS] ensemble target-val Dice: mixture=0.834, mixture@raw=0.833, mixture@logit=0.835, per_source=0.834, per_source@raw=0.833, per_source@logit=0.834, uniform=0.331, uniform@raw=0.184, uniform@logit=0.220 +[08:08:35][ 15s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 +[08:08:35][ 15s][INIT] student initialised from teacher multiclass (weight 0.920) +[08:08:37][ 16s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[08:08:37][ 16s][KD] distilling the mixture ensemble (prob space) into the student +[08:09:11][ 51s][KD] soft labels for 6000 patches from 4 teachers in 34s +[08:10:05][ 104s][KD] epoch 1/5 | patches=6000 | loss=0.1187 | 53s +[08:10:58][ 157s][KD] epoch 2/5 | patches=6000 | loss=0.1065 | 53s +[08:11:51][ 211s][KD] epoch 3/5 | patches=6000 | loss=0.1048 | 53s +[08:12:45][ 264s][KD] epoch 4/5 | patches=6000 | loss=0.1038 | 53s +[08:13:38][ 318s][KD] epoch 5/5 | patches=6000 | loss=0.1031 | 53s +[08:13:40][ 319s][ROUND0] before any target label: val_dice=0.8791 val_vi=0.0677 +[08:13:40][ 319s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=random | first_round=stratified +[08:13:52][ 331s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': -0.011803147059276488, 'auroc_top20_error': 0.48493623046875} +[08:14:04][ 343s][AL] round 1: stratified cold start: 34 patches from the membrane-rich half (ensemble prevalence >= 0.072) + 33 from the membrane-poor half of 2000 candidates +[08:14:06][ 345s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.946, uniform=0.320, top1=0.946, top2=0.946, top3=0.946 -> using mixture +[08:14:06][ 346s][WEIGHTS] round 1: student kept (labeled Dice 0.949 vs best teacher multiclass 0.946); distilling the corrected ensemble into it +[08:14:06][ 346s][WEIGHTS] round 1: teacher Dice on 67 labeled patches + 0 blank (313095 fg px): fly-mito=0.265, kasthuri=0.000, dros-vnc=0.016, multiclass=0.946 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:14:40][ 380s][KD] soft labels for 6000 patches from 4 teachers in 34s +[08:15:34][ 433s][KD] epoch 1/2 | patches=6000 | loss=0.0471 | 53s +[08:16:26][ 486s][KD] epoch 2/2 | patches=6000 | loss=0.0459 | 53s +[08:16:38][ 498s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 12s +[08:16:40][ 499s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3219 +[08:16:41][ 501s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3298 +[08:16:43][ 502s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3194 +[08:16:44][ 504s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.3680 +[08:16:46][ 505s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.3518 +[08:16:48][ 507s][AL] round 1/6 | labeled=67 | selection_seconds=23.7000 round_seconds=186.0000 restart_student_dice=0.9486 restart_teacher_dice=0.9463 val_dice=0.8429 val_vi=0.0898 val_ece=0.0069 +[08:16:51][ 511s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.950, uniform=0.301, top1=0.950, top2=0.950, top3=0.950 -> using mixture +[08:16:51][ 511s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (650938 fg px): fly-mito=0.278, kasthuri=0.000, dros-vnc=0.017, multiclass=0.950 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:16:53][ 513s][CONS] round 2: on the 67 new patches student Dice=0.889 vs ensemble 0.953 -> pseudo-labels from the ensemble | consistency x1.00 +[08:16:56][ 516s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.3047 +[08:16:59][ 518s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.3133 +[08:17:02][ 521s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.2729 +[08:17:04][ 524s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.4005 +[08:17:07][ 526s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.3654 +[08:17:09][ 528s][AL] round 2/6 | labeled=134 | selection_seconds=0.2000 round_seconds=19.0000 pseudo_student_dice=0.8890 pseudo_ensemble_dice=0.9535 consistency_scale=1.0000 val_dice=0.8087 val_vi=0.0907 val_ece=0.0081 +[08:17:14][ 534s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.945, uniform=0.303, top1=0.945, top2=0.945, top3=0.945 -> using mixture +[08:17:14][ 534s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (962422 fg px): fly-mito=0.276, kasthuri=0.000, dros-vnc=0.016, multiclass=0.945 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:17:16][ 536s][CONS] round 3: on the 67 new patches student Dice=0.940 vs ensemble 0.936 -> pseudo-labels from the ensemble | consistency x1.00 +[08:17:20][ 539s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.3110 +[08:17:24][ 543s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.3018 +[08:17:27][ 547s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.2696 +[08:17:31][ 551s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.2823 +[08:17:35][ 555s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2941 +[08:17:37][ 556s][AL] round 3/6 | labeled=201 | selection_seconds=0.2000 round_seconds=26.2000 pseudo_student_dice=0.9404 pseudo_ensemble_dice=0.9355 consistency_scale=1.0000 val_dice=0.8584 val_vi=0.0792 val_ece=0.0058 +[08:17:43][ 563s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.945, uniform=0.322, top1=0.945, top2=0.945, top3=0.945 -> using mixture +[08:17:43][ 563s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1317762 fg px): fly-mito=0.279, kasthuri=0.000, dros-vnc=0.016, multiclass=0.945 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:17:45][ 565s][CONS] round 4: on the 67 new patches student Dice=0.937 vs ensemble 0.943 -> pseudo-labels from the ensemble | consistency x1.00 +[08:17:50][ 570s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.2564 +[08:17:55][ 575s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.2162 +[08:18:00][ 580s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.2193 +[08:18:05][ 585s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.2031 +[08:18:10][ 590s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.2179 +[08:18:12][ 592s][AL] round 4/6 | labeled=268 | selection_seconds=0.2000 round_seconds=33.2000 pseudo_student_dice=0.9368 pseudo_ensemble_dice=0.9434 consistency_scale=1.0000 val_dice=0.8729 val_vi=0.0697 val_ece=0.0049 +[08:18:20][ 600s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.943, uniform=0.327, top1=0.943, top2=0.943, top3=0.943 -> using mixture +[08:18:20][ 600s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (1625408 fg px): fly-mito=0.276, kasthuri=0.000, dros-vnc=0.018, multiclass=0.943 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:18:22][ 602s][CONS] round 5: on the 67 new patches student Dice=0.932 vs ensemble 0.934 -> pseudo-labels from the ensemble | consistency x1.00 +[08:18:28][ 608s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2262 +[08:18:34][ 614s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.2403 +[08:18:40][ 620s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.2328 +[08:18:46][ 626s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.2202 +[08:18:52][ 632s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.2706 +[08:18:54][ 634s][AL] round 5/6 | labeled=335 | selection_seconds=0.2000 round_seconds=40.3000 pseudo_student_dice=0.9324 pseudo_ensemble_dice=0.9340 consistency_scale=1.0000 val_dice=0.8867 val_vi=0.0682 val_ece=0.0043 +[08:19:04][ 643s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.942, uniform=0.331, top1=0.942, top2=0.942, top3=0.942 -> using mixture +[08:19:04][ 643s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (2039142 fg px): fly-mito=0.286, kasthuri=0.000, dros-vnc=0.020, multiclass=0.942 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:19:06][ 645s][CONS] round 6: on the 67 new patches student Dice=0.939 vs ensemble 0.938 -> pseudo-labels from the ensemble | consistency x1.00 +[08:19:13][ 653s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2480 +[08:19:20][ 660s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.2425 +[08:19:28][ 667s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.2190 +[08:19:35][ 675s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.1968 +[08:19:42][ 682s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.1928 +[08:19:44][ 684s][AL] round 6/6 | labeled=402 | selection_seconds=0.2000 round_seconds=47.9000 pseudo_student_dice=0.9386 pseudo_ensemble_dice=0.9381 consistency_scale=1.0000 val_dice=0.8726 val_vi=0.0680 val_ece=0.0051 +[08:19:51][ 691s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.1898 +[08:19:59][ 698s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.1983 +[08:20:06][ 705s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.1853 +[08:20:13][ 713s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.1910 +[08:20:20][ 720s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.1857 +[08:20:28][ 727s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1814 +[08:20:35][ 734s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1699 +[08:20:42][ 741s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1783 +[08:20:49][ 749s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1681 +[08:20:56][ 756s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1632 +[08:20:59][ 758s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=72.4000 val_dice=0.8927 val_vi=0.0571 val_ece=0.0032 +[08:20:59][ 758s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[08:21:22][ 782s][TEST] raw student (no EMA, no TTA): patch_dice=0.9198 patch_vi=0.0413 +[08:21:26][ 785s][TEST] final model without TTA: patch_dice=0.9198 patch_vi=0.0413 +[08:21:26][ 785s][RESOURCES] {"phases": {"setup": {"seconds": 0.65, "peak_cpu_rss_mb": 1027.5, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 15.39, "peak_cpu_rss_mb": 1800.9, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 301.3, "peak_cpu_rss_mb": 3685.1, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 438.7, "peak_cpu_rss_mb": 3685.1, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 27.15, "peak_cpu_rss_mb": 3685.1, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 64.5, "seconds_per_image": 0.543}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3685.1, "total_seconds": 783.2} +[08:21:26][ 785s][TEST] patch_dice=0.9221 patch_iou=0.8940 patch_precision=0.9587 patch_recall=0.9318 patch_vi=0.0421 image_dice=0.9517 image_iou=0.9085 image_precision=0.9621 image_recall=0.9430 image_vi=0.0477 cal_ece=0.0019 cal_nll=0.0174 cal_brier=0.0037 cal_mean_confidence=0.9971 cal_accuracy=0.9952 cal_predicted_prevalence=0.0443 cal_true_prevalence=0.0452 cal_mean_probability=0.0445 +[08:21:26][ 785s][DONE] lucchi-pp macs-v11-norestart-acq-random b5 s2 | test patch Dice=0.9221 VI=0.0421 | image Dice=0.9517 VI=0.0477 | 785s diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/result.json b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..bd1e629f1dbbf5a3c9e6ae455751909291eb0226 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/result.json @@ -0,0 +1,647 @@ +{ + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 2, + "config": { + "task": "mitochondria", + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "random", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 0, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "lucchi-pp", + "--variants", + "macs-v11-norestart-acq-random", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "per_source": [ + 0.01302141100636151, + 0.021594387588085767, + 0.004953657914530629, + 0.960430543491022 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.001413550071584186, + "mmd2_mixture": 0.005039751415091165, + "mmd2_uniform": 0.2092225964739336, + "mmd2_single_source": [ + 0.5680514029229977, + 0.4644134839747518, + 0.6002496450435827, + 0.004698779437425582 + ], + "median_squared_distance": 0.18927357450912807, + "gammas": [ + 1.3208394285803657, + 2.6416788571607315, + 5.283357714321463, + 10.566715428642926, + 21.133430857285852 + ], + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6675364398352843, + 0.5899254221485877, + 0.8158221669373606, + 0.007664827003586616 + ], + "standardized": [ + 0.48001954758469056, + 0.22710032562990826, + 0.9632539109975666, + -1.6703737842121655 + ], + "tau": 0.5, + "max_weight": 0.960430543491022, + "entropy": 0.20441768713797392, + "effective_sources": 1.226810469408674, + "min_weight": 0.004953657914530629 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "multiclass", + "seconds": 1.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.243175358344919, + 0.16070093323689605, + 0.16243110642819403, + 0.8333930692567719 + ], + "ensemble_val_dice": { + "mixture": 0.8341160757031219, + "mixture@raw": 0.8334634835118209, + "mixture@logit": 0.8348662448964319, + "per_source": 0.833502984539706, + "per_source@raw": 0.833444326075437, + "per_source@logit": 0.8337659337218213, + "uniform": 0.33138683445274353, + "uniform@raw": 0.18373960615593887, + "uniform@logit": 0.2195153840155939 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "multiclass", + "weight_stats": { + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.2581988897471611, + "top_source": "multiclass", + "top_source_is_best_teacher": true + }, + "init_teacher": "multiclass", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 301.3, + "round0_val": { + "val_dice": 0.8790766237455594, + "val_vi": 0.06768383910512613, + "val_ece": 0.016723816126540798 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 67, + "selection_seconds": 23.7, + "round_seconds": 186.0, + "teacher_dice_labeled": [ + 0.2650592536077415, + 3.1912075848621877e-06, + 0.015534355861264863, + 0.9463032217749737 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 313095, + "evidence_nonempty_patches": 59, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9485712368876301, + "restart_teacher_dice": 0.9463032217749737, + "weights": [ + 3.893697301149704e-05, + 2.684582425110558e-06, + 1.5344296891974475e-06, + 0.9999568440148742 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9463032217749737, + "uniform": 0.3202334692828928, + "top1": 0.9463032217749737, + "top2": 0.9463032217749737, + "top3": 0.9463032217749737 + }, + "val_dice": 0.8429434228228141, + "val_vi": 0.08978167705203295, + "val_ece": 0.0069131388770909555 + }, + { + "round": 2, + "n_labeled": 134, + "selection_seconds": 0.2, + "round_seconds": 19.0, + "teacher_dice_labeled": [ + 0.2778697125764392, + 1.5356171030890474e-06, + 0.017278548656721873, + 0.9499988503544741 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 650938, + "evidence_nonempty_patches": 119, + "first_correction": false, + "restart": "", + "weights": [ + 4.2652631734919196e-05, + 2.5871289933130467e-06, + 1.5047711509529618e-06, + 0.9999532554681207 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9499988503544741, + "uniform": 0.301466056320226, + "top1": 0.9499988503544741, + "top2": 0.9499988503544741, + "top3": 0.9499988503544741 + }, + "pseudo_student_dice": 0.8889955357858202, + "pseudo_ensemble_dice": 0.9534549833958492, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8087181498107298, + "val_vi": 0.09067042658345541, + "val_ece": 0.008121887125095512 + }, + { + "round": 3, + "n_labeled": 201, + "selection_seconds": 0.2, + "round_seconds": 26.2, + "teacher_dice_labeled": [ + 0.2757759204124115, + 1.0379430459891806e-06, + 0.015819885588971527, + 0.9453266366089216 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 962422, + "evidence_nonempty_patches": 177, + "first_correction": false, + "restart": "", + "weights": [ + 4.376664100460177e-05, + 2.710856503369744e-06, + 1.5539111065414257e-06, + 0.9999519685913855 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9453276179956315, + "uniform": 0.3032102710353386, + "top1": 0.9453266366089216, + "top2": 0.9453276179956315, + "top3": 0.9453276179956315 + }, + "pseudo_student_dice": 0.9404276455069537, + "pseudo_ensemble_dice": 0.9355267771269392, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.858410385569507, + "val_vi": 0.07916697967487266, + "val_ece": 0.005806245949674594 + }, + { + "round": 4, + "n_labeled": 268, + "selection_seconds": 0.2, + "round_seconds": 33.2, + "teacher_dice_labeled": [ + 0.2791229530424369, + 7.578622524727151e-07, + 0.01607136272060965, + 0.9448081007080705 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1317762, + "evidence_nonempty_patches": 240, + "first_correction": false, + "restart": "", + "weights": [ + 4.549151570561618e-05, + 2.7249373717634123e-06, + 1.565919868292126e-06, + 0.9999502176270543 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9448084581132512, + "uniform": 0.3215367123223174, + "top1": 0.9448081007080705, + "top2": 0.9448084581132512, + "top3": 0.9448084581132512 + }, + "pseudo_student_dice": 0.9368319283309685, + "pseudo_ensemble_dice": 0.9434137782910188, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8729324795061693, + "val_vi": 0.06965555312520026, + "val_ece": 0.004948527027931386 + }, + { + "round": 5, + "n_labeled": 335, + "selection_seconds": 0.2, + "round_seconds": 40.3, + "teacher_dice_labeled": [ + 0.2763350649493667, + 6.145726231250157e-07, + 0.017712564490377423, + 0.942749806972975 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1625408, + "evidence_nonempty_patches": 295, + "first_correction": false, + "restart": "", + "weights": [ + 4.5160829719439484e-05, + 2.7816023998830874e-06, + 1.624936270161193e-06, + 0.9999504326316105 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9427494832966515, + "uniform": 0.32672321753673544, + "top1": 0.942749806972975, + "top2": 0.9427494832966515, + "top3": 0.9427494832966515 + }, + "pseudo_student_dice": 0.9324190293785697, + "pseudo_ensemble_dice": 0.9340109591064459, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8866854122957751, + "val_vi": 0.0681891458401869, + "val_ece": 0.004261731675815121 + }, + { + "round": 6, + "n_labeled": 402, + "selection_seconds": 0.2, + "round_seconds": 47.9, + "teacher_dice_labeled": [ + 0.28619399196414325, + 4.899844723920699e-07, + 0.019757140589865215, + 0.9418241910231305 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 2039142, + "evidence_nonempty_patches": 354, + "first_correction": false, + "restart": "", + "weights": [ + 5.03032820304342e-05, + 2.8074507369954927e-06, + 1.6739151764252291e-06, + 0.9999452153520562 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9418239320761237, + "uniform": 0.33109172380045615, + "top1": 0.9418241910231305, + "top2": 0.9418239320761237, + "top3": 0.9418239320761237 + }, + "pseudo_student_dice": 0.9386225586393795, + "pseudo_ensemble_dice": 0.9381375163551715, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8726083221884441, + "val_vi": 0.06800429169002674, + "val_ece": 0.0050527751706473355 + }, + { + "round": 7, + "n_labeled": 402, + "selection_seconds": 0.0, + "round_seconds": 72.4, + "final_consolidation": true, + "val_dice": 0.8927018882980704, + "val_vi": 0.05713700060625672, + "val_ece": 0.003212405176374467 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.011803147059276488, + "auroc_top20_error": 0.48493623046875 + }, + "al_seconds": 438.7, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 5.03032820304342e-05, + 2.8074507369954927e-06, + 1.6739151764252291e-06, + 0.9999452153520562 + ], + "weight_stats_final": { + "max_weight": 0.9999452153520562, + "entropy": 0.0006108088233461844, + "effective_sources": 1.0006109954050422, + "min_weight": 1.6739151764252291e-06 + }, + "top_source_final": "multiclass", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.9197620784957001, + "patch_iou": 0.8918172997253122, + "patch_precision": 0.9546867048598593, + "patch_recall": 0.9328226002610653, + "patch_vi": 0.041326195687250676, + "n_patches": 1750, + "cal_ece": 0.002478844528513137, + "cal_nll": 0.020452993414020015, + "cal_brier": 0.0037876075702497074, + "cal_mean_confidence": 0.9977890003084348, + "cal_accuracy": 0.9953565848214285, + "cal_predicted_prevalence": 0.044365513392857145, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04431518173538892 + }, + "test_notta": { + "patch_dice": 0.9197620784957001, + "patch_iou": 0.8918172997253122, + "patch_precision": 0.9546867048598593, + "patch_recall": 0.9328226002610653, + "patch_vi": 0.041326195687250676, + "n_patches": 1750, + "cal_ece": 0.002478844528513137, + "cal_nll": 0.020452993414020015, + "cal_brier": 0.0037876075702497074, + "cal_mean_confidence": 0.9977890003084348, + "cal_accuracy": 0.9953565848214285, + "cal_predicted_prevalence": 0.044365513392857145, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04431518173538892 + }, + "test": { + "patch_dice": 0.9220849208098438, + "patch_iou": 0.8940324922962406, + "patch_precision": 0.9587098470186034, + "patch_recall": 0.9317706018808314, + "patch_vi": 0.04206489978571124, + "n_patches": 1750, + "image_dice": 0.9516731990364949, + "image_iou": 0.9085469684565187, + "image_precision": 0.962119792734, + "image_recall": 0.9430433863720262, + "image_vi": 0.047732963082754765, + "n_images": 50, + "cal_ece": 0.0018714171281055654, + "cal_nll": 0.017363254705145607, + "cal_brier": 0.003678611359157221, + "cal_mean_confidence": 0.99708681891382, + "cal_accuracy": 0.9952154017857143, + "cal_predicted_prevalence": 0.04433370535714286, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.04445614350744048 + }, + "eval_seconds": 27.2, + "total_seconds": 785.3, + "resources": { + "phases": { + "setup": { + "seconds": 0.65, + "peak_cpu_rss_mb": 1027.5, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 15.39, + "peak_cpu_rss_mb": 1800.9, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 301.3, + "peak_cpu_rss_mb": 3685.1, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 438.7, + "peak_cpu_rss_mb": 3685.1, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4176.0, + "rounds": 6 + }, + "inference": { + "seconds": 27.15, + "peak_cpu_rss_mb": 3685.1, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4176.0, + "patches": 1750, + "images": 50, + "patches_per_second": 64.5, + "seconds_per_image": 0.543 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3685.1, + "total_seconds": 783.2 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-random_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..df05b28596adf4ebe59ebcee5b033e209c7290d8 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,67,23.7,186.0,"[0.2650592536077415, 3.1912075848621877e-06, 0.015534355861264863, 0.9463032217749737]",,313095,59,True,kept,0.9485712368876301,0.9463032217749737,"[3.893697301149704e-05, 2.684582425110558e-06, 1.5344296891974475e-06, 0.9999568440148742]",mixture,"{'mixture': 0.9463032217749737, 'uniform': 0.3202334692828928, 'top1': 0.9463032217749737, 'top2': 0.9463032217749737, 'top3': 0.9463032217749737}",0.8429434228228141,0.08978167705203295,0.0069131388770909555,,,,, +2,134,0.2,19.0,"[0.2778697125764392, 1.5356171030890474e-06, 0.017278548656721873, 0.9499988503544741]",,650938,119,False,,,,"[4.2652631734919196e-05, 2.5871289933130467e-06, 1.5047711509529618e-06, 0.9999532554681207]",mixture,"{'mixture': 0.9499988503544741, 'uniform': 0.301466056320226, 'top1': 0.9499988503544741, 'top2': 0.9499988503544741, 'top3': 0.9499988503544741}",0.8087181498107298,0.09067042658345541,0.008121887125095512,0.8889955357858202,0.9534549833958492,ensemble,1.0, +3,201,0.2,26.2,"[0.2757759204124115, 1.0379430459891806e-06, 0.015819885588971527, 0.9453266366089216]",,962422,177,False,,,,"[4.376664100460177e-05, 2.710856503369744e-06, 1.5539111065414257e-06, 0.9999519685913855]",mixture,"{'mixture': 0.9453276179956315, 'uniform': 0.3032102710353386, 'top1': 0.9453266366089216, 'top2': 0.9453276179956315, 'top3': 0.9453276179956315}",0.858410385569507,0.07916697967487266,0.005806245949674594,0.9404276455069537,0.9355267771269392,ensemble,1.0, +4,268,0.2,33.2,"[0.2791229530424369, 7.578622524727151e-07, 0.01607136272060965, 0.9448081007080705]",,1317762,240,False,,,,"[4.549151570561618e-05, 2.7249373717634123e-06, 1.565919868292126e-06, 0.9999502176270543]",mixture,"{'mixture': 0.9448084581132512, 'uniform': 0.3215367123223174, 'top1': 0.9448081007080705, 'top2': 0.9448084581132512, 'top3': 0.9448084581132512}",0.8729324795061693,0.06965555312520026,0.004948527027931386,0.9368319283309685,0.9434137782910188,ensemble,1.0, +5,335,0.2,40.3,"[0.2763350649493667, 6.145726231250157e-07, 0.017712564490377423, 0.942749806972975]",,1625408,295,False,,,,"[4.5160829719439484e-05, 2.7816023998830874e-06, 1.624936270161193e-06, 0.9999504326316105]",mixture,"{'mixture': 0.9427494832966515, 'uniform': 0.32672321753673544, 'top1': 0.942749806972975, 'top2': 0.9427494832966515, 'top3': 0.9427494832966515}",0.8866854122957751,0.0681891458401869,0.004261731675815121,0.9324190293785697,0.9340109591064459,ensemble,1.0, +6,402,0.2,47.9,"[0.28619399196414325, 4.899844723920699e-07, 0.019757140589865215, 0.9418241910231305]",,2039142,354,False,,,,"[5.03032820304342e-05, 2.8074507369954927e-06, 1.6739151764252291e-06, 0.9999452153520562]",mixture,"{'mixture': 0.9418239320761237, 'uniform': 0.33109172380045615, 'top1': 0.9418241910231305, 'top2': 0.9418239320761237, 'top3': 0.9418239320761237}",0.8726083221884441,0.06800429169002674,0.0050527751706473355,0.9386225586393795,0.9381375163551715,ensemble,1.0, +7,402,0.0,72.4,,,,,,,,,,,,0.8927018882980704,0.05713700060625672,0.003212405176374467,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/log.txt b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..d74e64596f79dfd18ced32774932540773435798 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/log.txt @@ -0,0 +1,100 @@ +[12:45:04][ 0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-random budget=5.0% seed=3 | {"task": "mitochondria", "target": "lucchi-pp", "variant": "macs-v11-norestart-acq-random", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "random", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 0, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[12:45:04][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[12:45:04][ 0s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches +[12:45:07][ 2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60 +[12:45:07][ 2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[12:45:08][ 4s][WEIGHTS] mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 | eff.sources=1.44 | MMD2 mix=0.0050 uniform=0.2092 +[12:45:08][ 4s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass +[12:45:34][ 30s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833 +[12:45:34][ 30s][TEACHERS] ensemble target-val Dice: mixture=0.834, mixture@raw=0.833, mixture@logit=0.835, per_source=0.834, per_source@raw=0.833, per_source@logit=0.834, uniform=0.331, uniform@raw=0.184, uniform@logit=0.220 +[12:45:34][ 30s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920 +[12:45:34][ 30s][INIT] student initialised from teacher multiclass (weight 0.920) +[12:45:37][ 33s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[12:45:37][ 33s][KD] distilling the mixture ensemble (prob space) into the student +[12:46:51][ 106s][KD] soft labels for 6000 patches from 4 teachers in 73s +[12:48:39][ 214s][KD] epoch 1/5 | patches=6000 | loss=0.1181 | 107s +[12:50:25][ 320s][KD] epoch 2/5 | patches=6000 | loss=0.1061 | 106s +[12:52:02][ 418s][KD] epoch 3/5 | patches=6000 | loss=0.1044 | 98s +[12:53:50][ 526s][KD] epoch 4/5 | patches=6000 | loss=0.1036 | 107s +[12:55:37][ 633s][KD] epoch 5/5 | patches=6000 | loss=0.1027 | 107s +[12:55:41][ 636s][ROUND0] before any target label: val_dice=0.8747 val_vi=0.0699 +[12:55:41][ 636s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=random | first_round=stratified +[12:56:07][ 663s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.010049961259445711, 'auroc_top20_error': 0.50328623046875} +[12:56:32][ 687s][AL] round 1: stratified cold start: 34 patches from the membrane-rich half (ensemble prevalence >= 0.073) + 33 from the membrane-poor half of 2000 candidates +[12:56:36][ 691s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.931, uniform=0.391, top1=0.931, top2=0.931, top3=0.931 -> using mixture +[12:56:36][ 692s][WEIGHTS] round 1: student kept (labeled Dice 0.931 vs best teacher multiclass 0.931); distilling the corrected ensemble into it +[12:56:36][ 692s][WEIGHTS] round 1: teacher Dice on 67 labeled patches + 0 blank (327320 fg px): fly-mito=0.283, kasthuri=0.000, dros-vnc=0.014, multiclass=0.931 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[12:57:49][ 764s][KD] soft labels for 6000 patches from 4 teachers in 72s +[12:59:36][ 872s][KD] epoch 1/2 | patches=6000 | loss=0.0492 | 107s +[13:01:20][ 976s][KD] epoch 2/2 | patches=6000 | loss=0.0460 | 104s +[13:01:45][ 1001s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 25s +[13:01:48][ 1003s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3050 +[13:01:50][ 1006s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3252 +[13:01:53][ 1009s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3397 +[13:01:56][ 1011s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.3087 +[13:01:59][ 1014s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.3171 +[13:02:02][ 1018s][AL] round 1/6 | labeled=67 | selection_seconds=51.2000 round_seconds=378.0000 restart_student_dice=0.9314 restart_teacher_dice=0.9312 val_dice=0.8066 val_vi=0.0936 val_ece=0.0032 +[13:02:10][ 1026s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.937, uniform=0.385, top1=0.937, top2=0.937, top3=0.937 -> using mixture +[13:02:10][ 1026s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (638484 fg px): fly-mito=0.276, kasthuri=0.000, dros-vnc=0.017, multiclass=0.937 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[13:02:13][ 1028s][CONS] round 2: on the 67 new patches student Dice=0.927 vs ensemble 0.942 -> pseudo-labels from the ensemble | consistency x1.00 +[13:02:18][ 1034s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.2894 +[13:02:23][ 1039s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.3272 +[13:02:28][ 1044s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.3135 +[13:02:33][ 1049s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.2434 +[13:02:38][ 1054s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.2568 +[13:02:42][ 1057s][AL] round 2/6 | labeled=134 | selection_seconds=0.2000 round_seconds=36.2000 pseudo_student_dice=0.9273 pseudo_ensemble_dice=0.9421 consistency_scale=1.0000 val_dice=0.8526 val_vi=0.0773 val_ece=0.0061 +[13:02:52][ 1067s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.933, uniform=0.357, top1=0.933, top2=0.933, top3=0.933 -> using mixture +[13:02:52][ 1067s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (954156 fg px): fly-mito=0.277, kasthuri=0.000, dros-vnc=0.021, multiclass=0.933 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[13:02:55][ 1070s][CONS] round 3: on the 67 new patches student Dice=0.931 vs ensemble 0.925 -> pseudo-labels from the ensemble | consistency x1.00 +[13:03:02][ 1078s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.3464 +[13:03:10][ 1085s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.2821 +[13:03:17][ 1093s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.2818 +[13:03:25][ 1101s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.2607 +[13:03:32][ 1108s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2396 +[13:03:36][ 1112s][AL] round 3/6 | labeled=201 | selection_seconds=0.2000 round_seconds=50.6000 pseudo_student_dice=0.9312 pseudo_ensemble_dice=0.9249 consistency_scale=1.0000 val_dice=0.8575 val_vi=0.0771 val_ece=0.0055 +[13:03:51][ 1127s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.937, uniform=0.356, top1=0.937, top2=0.937, top3=0.937 -> using mixture +[13:03:51][ 1127s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1247725 fg px): fly-mito=0.271, kasthuri=0.000, dros-vnc=0.022, multiclass=0.937 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[13:03:54][ 1130s][CONS] round 4: on the 67 new patches student Dice=0.943 vs ensemble 0.949 -> pseudo-labels from the ensemble | consistency x1.00 +[13:04:04][ 1140s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.2505 +[13:04:14][ 1149s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.2478 +[13:04:24][ 1159s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.2475 +[13:04:34][ 1169s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.2093 +[13:04:42][ 1178s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.2052 +[13:04:46][ 1182s][AL] round 4/6 | labeled=268 | selection_seconds=0.2000 round_seconds=66.6000 pseudo_student_dice=0.9433 pseudo_ensemble_dice=0.9492 consistency_scale=1.0000 val_dice=0.8743 val_vi=0.0662 val_ece=0.0043 +[13:05:05][ 1200s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.937, uniform=0.364, top1=0.937, top2=0.937, top3=0.937 -> using mixture +[13:05:05][ 1200s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (1552623 fg px): fly-mito=0.272, kasthuri=0.000, dros-vnc=0.024, multiclass=0.937 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[13:05:14][ 1209s][CONS] round 5: on the 67 new patches student Dice=0.956 vs ensemble 0.941 -> pseudo-labels from the student | consistency x1.00 +[13:05:26][ 1222s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2184 +[13:05:38][ 1234s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.1802 +[13:05:51][ 1246s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.2077 +[13:06:03][ 1258s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.1895 +[13:06:15][ 1271s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.2392 +[13:06:19][ 1274s][AL] round 5/6 | labeled=335 | selection_seconds=0.2000 round_seconds=89.0000 pseudo_student_dice=0.9558 pseudo_ensemble_dice=0.9410 consistency_scale=1.0000 val_dice=0.7626 val_vi=0.1118 val_ece=0.0159 +[13:07:23][ 1338s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.268, kasthuri=0.000, dros-vnc=0.022, multiclass=0.907 | blended with labeled Dice at 0.50 +[13:08:31][ 1407s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.922, uniform=0.347, top1=0.922, top2=0.922, top3=0.922 -> using mixture +[13:08:31][ 1407s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (1889558 fg px): fly-mito=0.271, kasthuri=0.000, dros-vnc=0.022, multiclass=0.922 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[13:08:40][ 1416s][CONS] round 6: on the 67 new patches student Dice=0.894 vs ensemble 0.939 -> pseudo-labels from the student | consistency x1.00 +[13:08:54][ 1429s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2493 +[13:09:06][ 1442s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.2115 +[13:09:19][ 1455s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.2075 +[13:09:33][ 1468s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.2330 +[13:09:47][ 1483s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.1925 +[13:09:51][ 1487s][AL] round 6/6 | labeled=402 | selection_seconds=0.2000 round_seconds=208.9000 pseudo_student_dice=0.8941 pseudo_ensemble_dice=0.9391 consistency_scale=1.0000 val_dice=0.8450 val_vi=0.0784 val_ece=0.0090 +[13:09:57][ 1493s][CONS] final consolidation: pseudo-labels refreshed from the student +[13:10:12][ 1508s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.2034 +[13:10:27][ 1522s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.2859 +[13:10:41][ 1537s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.2076 +[13:10:55][ 1551s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.1690 +[13:11:10][ 1566s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.1735 +[13:11:25][ 1580s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1633 +[13:11:39][ 1595s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1612 +[13:11:54][ 1610s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1619 +[13:12:09][ 1625s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1562 +[13:12:24][ 1639s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1475 +[13:12:27][ 1643s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=152.7000 val_dice=0.8820 val_vi=0.0631 val_ece=0.0047 +[13:12:27][ 1643s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000 +[13:13:15][ 1691s][TEST] raw student (no EMA, no TTA): patch_dice=0.9058 patch_vi=0.0451 +[13:13:21][ 1697s][TEST] final model without TTA: patch_dice=0.9058 patch_vi=0.0451 +[13:13:21][ 1697s][RESOURCES] {"phases": {"setup": {"seconds": 2.16, "peak_cpu_rss_mb": 1027.5, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 30.8, "peak_cpu_rss_mb": 1800.3, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 599.57, "peak_cpu_rss_mb": 3684.7, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 1006.62, "peak_cpu_rss_mb": 3684.7, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 53.74, "peak_cpu_rss_mb": 3684.7, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 32.6, "seconds_per_image": 1.075}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 3684.7, "total_seconds": 1692.9} +[13:13:21][ 1697s][TEST] patch_dice=0.9245 patch_iou=0.8965 patch_precision=0.9529 patch_recall=0.9401 patch_vi=0.0412 image_dice=0.9525 image_iou=0.9096 image_precision=0.9574 image_recall=0.9486 image_vi=0.0481 cal_ece=0.0019 cal_nll=0.0169 cal_brier=0.0036 cal_mean_confidence=0.9973 cal_accuracy=0.9954 cal_predicted_prevalence=0.0451 cal_true_prevalence=0.0452 cal_mean_probability=0.0452 +[13:13:21][ 1697s][DONE] lucchi-pp macs-v11-norestart-acq-random b5 s3 | test patch Dice=0.9245 VI=0.0412 | image Dice=0.9525 VI=0.0481 | 1697s diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/result.json b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..c61a0161f33a0b9035d60d84aba9917f7cd1f6a1 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/result.json @@ -0,0 +1,652 @@ +{ + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 3, + "config": { + "task": "mitochondria", + "target": "lucchi-pp", + "variant": "macs-v11-norestart-acq-random", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "random", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 0, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "lucchi-pp", + "--variants", + "macs-v11-norestart-acq-random", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "alphas": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "drosophila-vnc", + "multiclass" + ], + "mixture": [ + 0.03256909508222815, + 0.03180124345477557, + 0.015561926534604847, + 0.9200677349283914 + ], + "per_source": [ + 0.01302141100636151, + 0.021594387588085767, + 0.004953657914530629, + 0.960430543491022 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.001413550071584186, + "mmd2_mixture": 0.005039751415091165, + "mmd2_uniform": 0.2092225964739336, + "mmd2_single_source": [ + 0.5680514029229977, + 0.4644134839747518, + 0.6002496450435827, + 0.004698779437425582 + ], + "median_squared_distance": 0.18927357450912807, + "gammas": [ + 1.3208394285803657, + 2.6416788571607315, + 5.283357714321463, + 10.566715428642926, + 21.133430857285852 + ], + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6675364398352843, + 0.5899254221485877, + 0.8158221669373606, + 0.007664827003586616 + ], + "standardized": [ + 0.48001954758469056, + 0.22710032562990826, + 0.9632539109975666, + -1.6703737842121655 + ], + "tau": 0.5, + "max_weight": 0.960430543491022, + "entropy": 0.20441768713797392, + "effective_sources": 1.226810469408674, + "min_weight": 0.004953657914530629 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "multiclass", + "seconds": 1.3 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.243175358344919, + 0.16070093323689605, + 0.16243110642819403, + 0.8333930692567719 + ], + "ensemble_val_dice": { + "mixture": 0.8341160757031219, + "mixture@raw": 0.8334634835118209, + "mixture@logit": 0.8348662448964319, + "per_source": 0.833502984539706, + "per_source@raw": 0.833444326075437, + "per_source@logit": 0.8337659337218213, + "uniform": 0.33138683445274353, + "uniform@raw": 0.18373960615593887, + "uniform@logit": 0.2195153840155939 + }, + "val_patches": 1024 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 2.767761468887329, + 2.597255229949951 + ], + "top_source": "multiclass", + "weight_stats": { + "max_weight": 0.9200677349283914, + "entropy": 0.3626201343506979, + "effective_sources": 1.4370898544728725, + "min_weight": 0.015561926534604847 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.7999999999999999, + "spearman_weight_taxonomy": -0.2581988897471611, + "top_source": "multiclass", + "top_source_is_best_teacher": true + }, + "init_teacher": "multiclass", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 599.6, + "round0_val": { + "val_dice": 0.8746797680525892, + "val_vi": 0.06990791441807019, + "val_ece": 0.017578699866858123 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 67, + "selection_seconds": 51.2, + "round_seconds": 378.0, + "teacher_dice_labeled": [ + 0.2829946467808237, + 3.0551049275787378e-06, + 0.014406917734236871, + 0.931227563092392 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 327320, + "evidence_nonempty_patches": 60, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9313719093513289, + "restart_teacher_dice": 0.931227563092392, + "weights": [ + 5.4165213515608284e-05, + 3.121346642323861e-06, + 1.7640725486920737e-06, + 0.9999409493672934 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9312292062554963, + "uniform": 0.39082248145471704, + "top1": 0.931227563092392, + "top2": 0.9312292062554963, + "top3": 0.9312292062554963 + }, + "val_dice": 0.8066373040592636, + "val_vi": 0.0935794564465943, + "val_ece": 0.0032048020791910145 + }, + { + "round": 2, + "n_labeled": 134, + "selection_seconds": 0.2, + "round_seconds": 36.2, + "teacher_dice_labeled": [ + 0.27641629711751664, + 1.5662075068325803e-06, + 0.016847013828506712, + 0.9365173424315543 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 638484, + "evidence_nonempty_patches": 121, + "first_correction": false, + "restart": "", + "weights": [ + 4.8103929460706e-05, + 2.960499826833363e-06, + 1.7145223414133562e-06, + 0.999947221048371 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9365180813869538, + "uniform": 0.38509771307669166, + "top1": 0.9365173424315543, + "top2": 0.9365180813869538, + "top3": 0.9365180813869538 + }, + "pseudo_student_dice": 0.9273011704438608, + "pseudo_ensemble_dice": 0.9420944112507659, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8525544422758828, + "val_vi": 0.07727239261727706, + "val_ece": 0.006057534209917306 + }, + { + "round": 3, + "n_labeled": 201, + "selection_seconds": 0.2, + "round_seconds": 50.6, + "teacher_dice_labeled": [ + 0.2768622697323119, + 1.0479115646472363e-06, + 0.020822716969196436, + 0.9326735920000169 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 954156, + "evidence_nonempty_patches": 179, + "first_correction": false, + "restart": "", + "weights": [ + 5.021223559471877e-05, + 3.0764861089940886e-06, + 1.8539651978061727e-06, + 0.9999448573130985 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9326735210106274, + "uniform": 0.35685311497406697, + "top1": 0.9326735920000169, + "top2": 0.9326735210106274, + "top3": 0.9326735210106274 + }, + "pseudo_student_dice": 0.9312279767591186, + "pseudo_ensemble_dice": 0.9249314084813591, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8574854553614087, + "val_vi": 0.07705768906264865, + "val_ece": 0.00547243533428279 + }, + { + "round": 4, + "n_labeled": 268, + "selection_seconds": 0.2, + "round_seconds": 66.6, + "teacher_dice_labeled": [ + 0.27132866269357364, + 8.013796552144171e-07, + 0.021680446086307453, + 0.9365732745697064 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1247725, + "evidence_nonempty_patches": 236, + "first_correction": false, + "restart": "", + "weights": [ + 4.5692333705226276e-05, + 2.9588286815306353e-06, + 1.7984260296201381e-06, + 0.9999495504115836 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9365732235170067, + "uniform": 0.35573698396375353, + "top1": 0.9365732745697064, + "top2": 0.9365736514973704, + "top3": 0.9365736514973704 + }, + "pseudo_student_dice": 0.9433402395509605, + "pseudo_ensemble_dice": 0.9491557998683569, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.8742530911049364, + "val_vi": 0.06621405912392658, + "val_ece": 0.0042668670209311255 + }, + { + "round": 5, + "n_labeled": 335, + "selection_seconds": 0.2, + "round_seconds": 89.0, + "teacher_dice_labeled": [ + 0.2724814741345257, + 6.44019920824191e-07, + 0.02375011352067791, + 0.9374371989266431 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 1552623, + "evidence_nonempty_patches": 294, + "first_correction": false, + "restart": "", + "weights": [ + 4.582451791310805e-05, + 2.933371747898655e-06, + 1.8202414490509616e-06, + 0.9999494218688899 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9374371584734787, + "uniform": 0.3639959586365716, + "top1": 0.9374371989266431, + "top2": 0.937437502000421, + "top3": 0.937437502000421 + }, + "pseudo_student_dice": 0.955806172217447, + "pseudo_ensemble_dice": 0.9409659666241469, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.7625931027313826, + "val_vi": 0.11178115282419912, + "val_ece": 0.015894579311064826 + }, + { + "round": 6, + "n_labeled": 402, + "selection_seconds": 0.2, + "round_seconds": 208.9, + "teacher_dice_labeled": [ + 0.27084226294250596, + 3.18202249863172e-07, + 0.021507052351710243, + 0.9224539870041173 + ], + "teacher_dice_pool": [ + 0.26759511037933337, + 1.0721493238115038e-07, + 0.021850080774008702, + 0.9071754901205471 + ], + "evidence_fg_pixels": 1889558, + "evidence_nonempty_patches": 349, + "first_correction": false, + "restart": "", + "weights": [ + 5.236571157041346e-05, + 3.4074837873539075e-06, + 2.067548076585342e-06, + 0.9999421592565656 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9224541596783467, + "uniform": 0.3468398621263498, + "top1": 0.9224539870041173, + "top2": 0.9224541693554108, + "top3": 0.9224541929990463 + }, + "pseudo_student_dice": 0.8940731225781136, + "pseudo_ensemble_dice": 0.9390615371840871, + "pseudo_source": "student", + "consistency_scale": 1.0, + "val_dice": 0.8450380225144468, + "val_vi": 0.07840620810051646, + "val_ece": 0.008959131773761333 + }, + { + "round": 7, + "n_labeled": 402, + "selection_seconds": 0.0, + "round_seconds": 152.7, + "final_consolidation": true, + "val_dice": 0.8819887160319484, + "val_vi": 0.06313731607063136, + "val_ece": 0.004743810323116236 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": 0.010049961259445711, + "auroc_top20_error": 0.50328623046875 + }, + "al_seconds": 1006.6, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 5.236571157041346e-05, + 3.4074837873539075e-06, + 2.067548076585342e-06, + 0.9999421592565656 + ], + "weight_stats_final": { + "max_weight": 0.9999421592565656, + "entropy": 0.0006439825105239936, + "effective_sources": 1.0006441899117795, + "min_weight": 2.067548076585342e-06 + }, + "top_source_final": "multiclass", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.905763957898399, + "patch_iou": 0.8752892878782716, + "patch_precision": 0.9282110290668417, + "patch_recall": 0.9420744904236191, + "patch_vi": 0.045114815269431625, + "n_patches": 1750, + "cal_ece": 0.0032511465621953587, + "cal_nll": 0.021478469374991947, + "cal_brier": 0.0042622075831601864, + "cal_mean_confidence": 0.9980351185288723, + "cal_accuracy": 0.9947918526785714, + "cal_predicted_prevalence": 0.045818638392857144, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.045613436778392386 + }, + "test_notta": { + "patch_dice": 0.905763957898399, + "patch_iou": 0.8752892878782716, + "patch_precision": 0.9282110290668417, + "patch_recall": 0.9420744904236191, + "patch_vi": 0.045114815269431625, + "n_patches": 1750, + "cal_ece": 0.0032511465621953587, + "cal_nll": 0.021478469374991947, + "cal_brier": 0.0042622075831601864, + "cal_mean_confidence": 0.9980351185288723, + "cal_accuracy": 0.9947918526785714, + "cal_predicted_prevalence": 0.045818638392857144, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.045613436778392386 + }, + "test": { + "patch_dice": 0.9244646070531019, + "patch_iou": 0.8964544528796147, + "patch_precision": 0.9529173667417955, + "patch_recall": 0.9400653331135798, + "patch_vi": 0.041198004191915875, + "n_patches": 1750, + "image_dice": 0.9524724217216665, + "image_iou": 0.9096158255492567, + "image_precision": 0.9574444131373425, + "image_recall": 0.9485733345926947, + "image_vi": 0.04807926525787828, + "n_images": 50, + "cal_ece": 0.00194445714194496, + "cal_nll": 0.016850147530774795, + "cal_brier": 0.0036057341277805476, + "cal_mean_confidence": 0.9972990189657643, + "cal_accuracy": 0.9953772321428571, + "cal_predicted_prevalence": 0.045103794642857145, + "cal_true_prevalence": 0.04515513392857143, + "cal_mean_probability": 0.0451731278423339 + }, + "eval_seconds": 53.7, + "total_seconds": 1696.6, + "resources": { + "phases": { + "setup": { + "seconds": 2.16, + "peak_cpu_rss_mb": 1027.5, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 30.8, + "peak_cpu_rss_mb": 1800.3, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 599.57, + "peak_cpu_rss_mb": 3684.7, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4128.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 1006.62, + "peak_cpu_rss_mb": 3684.7, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4176.0, + "rounds": 6 + }, + "inference": { + "seconds": 53.74, + "peak_cpu_rss_mb": 3684.7, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4176.0, + "patches": 1750, + "images": 50, + "patches_per_second": 32.6, + "seconds_per_image": 1.075 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 3684.7, + "total_seconds": 1692.9 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-random_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..61173305482a105fe7d4e71dc7af754c76ddc6cf --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,67,51.2,378.0,"[0.2829946467808237, 3.0551049275787378e-06, 0.014406917734236871, 0.931227563092392]",,327320,60,True,kept,0.9313719093513289,0.931227563092392,"[5.4165213515608284e-05, 3.121346642323861e-06, 1.7640725486920737e-06, 0.9999409493672934]",mixture,"{'mixture': 0.9312292062554963, 'uniform': 0.39082248145471704, 'top1': 0.931227563092392, 'top2': 0.9312292062554963, 'top3': 0.9312292062554963}",0.8066373040592636,0.0935794564465943,0.0032048020791910145,,,,, +2,134,0.2,36.2,"[0.27641629711751664, 1.5662075068325803e-06, 0.016847013828506712, 0.9365173424315543]",,638484,121,False,,,,"[4.8103929460706e-05, 2.960499826833363e-06, 1.7145223414133562e-06, 0.999947221048371]",mixture,"{'mixture': 0.9365180813869538, 'uniform': 0.38509771307669166, 'top1': 0.9365173424315543, 'top2': 0.9365180813869538, 'top3': 0.9365180813869538}",0.8525544422758828,0.07727239261727706,0.006057534209917306,0.9273011704438608,0.9420944112507659,ensemble,1.0, +3,201,0.2,50.6,"[0.2768622697323119, 1.0479115646472363e-06, 0.020822716969196436, 0.9326735920000169]",,954156,179,False,,,,"[5.021223559471877e-05, 3.0764861089940886e-06, 1.8539651978061727e-06, 0.9999448573130985]",mixture,"{'mixture': 0.9326735210106274, 'uniform': 0.35685311497406697, 'top1': 0.9326735920000169, 'top2': 0.9326735210106274, 'top3': 0.9326735210106274}",0.8574854553614087,0.07705768906264865,0.00547243533428279,0.9312279767591186,0.9249314084813591,ensemble,1.0, +4,268,0.2,66.6,"[0.27132866269357364, 8.013796552144171e-07, 0.021680446086307453, 0.9365732745697064]",,1247725,236,False,,,,"[4.5692333705226276e-05, 2.9588286815306353e-06, 1.7984260296201381e-06, 0.9999495504115836]",mixture,"{'mixture': 0.9365732235170067, 'uniform': 0.35573698396375353, 'top1': 0.9365732745697064, 'top2': 0.9365736514973704, 'top3': 0.9365736514973704}",0.8742530911049364,0.06621405912392658,0.0042668670209311255,0.9433402395509605,0.9491557998683569,ensemble,1.0, +5,335,0.2,89.0,"[0.2724814741345257, 6.44019920824191e-07, 0.02375011352067791, 0.9374371989266431]",,1552623,294,False,,,,"[4.582451791310805e-05, 2.933371747898655e-06, 1.8202414490509616e-06, 0.9999494218688899]",mixture,"{'mixture': 0.9374371584734787, 'uniform': 0.3639959586365716, 'top1': 0.9374371989266431, 'top2': 0.937437502000421, 'top3': 0.937437502000421}",0.7625931027313826,0.11178115282419912,0.015894579311064826,0.955806172217447,0.9409659666241469,student,1.0, +6,402,0.2,208.9,"[0.27084226294250596, 3.18202249863172e-07, 0.021507052351710243, 0.9224539870041173]","[0.26759511037933337, 1.0721493238115038e-07, 0.021850080774008702, 0.9071754901205471]",1889558,349,False,,,,"[5.236571157041346e-05, 3.4074837873539075e-06, 2.067548076585342e-06, 0.9999421592565656]",mixture,"{'mixture': 0.9224541596783467, 'uniform': 0.3468398621263498, 'top1': 0.9224539870041173, 'top2': 0.9224541693554108, 'top3': 0.9224541929990463}",0.8450380225144468,0.07840620810051646,0.008959131773761333,0.8940731225781136,0.9390615371840871,student,1.0, +7,402,0.0,152.7,,,,,,,,,,,,0.8819887160319484,0.06313731607063136,0.004743810323116236,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/log.txt b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..7f760ebeaeb7764e98cf6ff56f8f4079532e086f --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/log.txt @@ -0,0 +1,103 @@ +[04:52:32][ 0s][START] target=multiclass variant=macs-v11-norestart-acq-entropy budget=5.0% seed=1 | {"task": "mitochondria", "target": "multiclass", "variant": "macs-v11-norestart-acq-entropy", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "entropy", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[04:52:32][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[04:52:32][ 0s][DATA] multiclass: train=1855 val=350 test=175 patches +[04:52:33][ 1s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, dros-vnc=2.77 +[04:52:33][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[04:52:34][ 2s][WEIGHTS] mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 | eff.sources=1.19 | MMD2 mix=0.0059 uniform=0.2484 +[04:52:34][ 2s][WEIGHTS] per-source: fly-mito=0.014, kasthuri=0.019, lucchi=0.962, dros-vnc=0.005 | min-MMD source (NeuroADDA rule) = lucchi +[04:52:40][ 8s][TEACHERS] target-val Dice: fly-mito=0.258, kasthuri=0.175, lucchi=0.968, dros-vnc=0.176 +[04:52:40][ 8s][TEACHERS] ensemble target-val Dice: mixture=0.967, mixture@raw=0.968, mixture@logit=0.967, per_source=0.967, per_source@raw=0.967, per_source@logit=0.966, uniform=0.248, uniform@raw=0.208, uniform@logit=0.195 +[04:52:40][ 8s][WEIGHTS] using mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 +[04:52:40][ 8s][INIT] student initialised from teacher lucchi (weight 0.968) +[04:52:41][ 9s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[04:52:42][ 9s][KD] distilling the mixture ensemble (prob space) into the student +[04:52:52][ 20s][KD] soft labels for 1855 patches from 4 teachers in 11s +[04:53:09][ 37s][KD] epoch 1/5 | patches=1855 | loss=0.1029 | 17s +[04:53:26][ 53s][KD] epoch 2/5 | patches=1855 | loss=0.0735 | 17s +[04:53:42][ 70s][KD] epoch 3/5 | patches=1855 | loss=0.0709 | 16s +[04:53:58][ 86s][KD] epoch 4/5 | patches=1855 | loss=0.0699 | 16s +[04:54:15][ 103s][KD] epoch 5/5 | patches=1855 | loss=0.0686 | 17s +[04:54:16][ 104s][ROUND0] before any target label: val_dice=0.9624 val_vi=0.0433 +[04:54:16][ 104s][AL] patches=1855 (pool 1855 after dropping blank; free blank negatives: True) | budget=92 (5.0%) | rounds=6 | per_round=15 | acquisition=entropy | first_round=stratified +[04:54:19][ 106s][AL] round-1 uncertainty quality on 1855 patches: {'spearman_uncertainty_error': -0.15377629786596558, 'auroc_top20_error': 0.26876984328797376} +[04:54:30][ 117s][AL] round 1: stratified cold start: 8 patches from the membrane-rich half (ensemble prevalence >= 0.071) + 7 from the membrane-poor half of 1855 candidates +[04:54:31][ 118s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.967, uniform=0.174, top1=0.967, top2=0.967, top3=0.967 -> using mixture +[04:54:31][ 118s][WEIGHTS] round 1: student kept (labeled Dice 0.966 vs best teacher lucchi 0.967); distilling the corrected ensemble into it +[04:54:31][ 118s][WEIGHTS] round 1: teacher Dice on 15 labeled patches + 0 blank (72996 fg px): fly-mito=0.291, kasthuri=0.000, lucchi=0.967, dros-vnc=0.018 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[04:54:41][ 129s][KD] soft labels for 1840 patches from 4 teachers in 11s +[04:54:58][ 146s][KD] epoch 1/2 | patches=1840 | loss=0.0456 | 17s +[04:55:15][ 163s][KD] epoch 2/2 | patches=1840 | loss=0.0444 | 17s +[04:55:26][ 173s][CONS] pseudo-labels for 1840 unlabeled patches from the current ensemble in 11s +[04:55:26][ 174s][FT] round 1/6 epoch 1/5 | patches=15 | loss=0.2187 +[04:55:27][ 174s][FT] round 1/6 epoch 2/5 | patches=15 | loss=0.1802 +[04:55:27][ 175s][FT] round 1/6 epoch 3/5 | patches=15 | loss=0.1665 +[04:55:28][ 176s][FT] round 1/6 epoch 4/5 | patches=15 | loss=0.1811 +[04:55:29][ 176s][FT] round 1/6 epoch 5/5 | patches=15 | loss=0.1581 +[04:55:29][ 177s][AL] round 1/6 | labeled=15 | selection_seconds=13.7000 round_seconds=72.7000 restart_student_dice=0.9664 restart_teacher_dice=0.9671 val_dice=0.9320 val_vi=0.0494 val_ece=0.0024 +[04:55:32][ 180s][AL] round 2: 278/1840 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:55:33][ 181s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.960, uniform=0.084, top1=0.960, top2=0.960, top3=0.960 -> using mixture +[04:55:33][ 181s][WEIGHTS] round 2: teacher Dice on 30 labeled patches + 0 blank (162318 fg px): fly-mito=0.330, kasthuri=0.000, lucchi=0.960, dros-vnc=0.008 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[04:55:34][ 182s][CONS] round 2: on the 15 new patches student Dice=0.933 vs ensemble 0.953 -> pseudo-labels from the ensemble | consistency x1.00 +[04:55:35][ 183s][FT] round 2/6 epoch 1/5 | patches=30 | loss=0.2058 +[04:55:36][ 184s][FT] round 2/6 epoch 2/5 | patches=30 | loss=0.2472 +[04:55:37][ 185s][FT] round 2/6 epoch 3/5 | patches=30 | loss=0.1966 +[04:55:38][ 186s][FT] round 2/6 epoch 4/5 | patches=30 | loss=0.2155 +[04:55:39][ 186s][FT] round 2/6 epoch 5/5 | patches=30 | loss=0.1822 +[04:55:40][ 187s][AL] round 2/6 | labeled=30 | selection_seconds=3.0000 round_seconds=9.4000 pseudo_student_dice=0.9326 pseudo_ensemble_dice=0.9535 consistency_scale=1.0000 val_dice=0.9256 val_vi=0.0527 val_ece=0.0029 +[04:55:42][ 190s][AL] round 3: 277/1825 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:55:44][ 192s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.956, uniform=0.072, top1=0.956, top2=0.956, top3=0.956 -> using mixture +[04:55:44][ 192s][WEIGHTS] round 3: teacher Dice on 45 labeled patches + 0 blank (253609 fg px): fly-mito=0.331, kasthuri=0.000, lucchi=0.956, dros-vnc=0.008 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[04:55:45][ 193s][CONS] round 3: on the 15 new patches student Dice=0.914 vs ensemble 0.950 -> pseudo-labels from the ensemble | consistency x1.00 +[04:55:46][ 194s][FT] round 3/6 epoch 1/5 | patches=45 | loss=0.2431 +[04:55:47][ 195s][FT] round 3/6 epoch 2/5 | patches=45 | loss=0.2112 +[04:55:49][ 196s][FT] round 3/6 epoch 3/5 | patches=45 | loss=0.2139 +[04:55:50][ 197s][FT] round 3/6 epoch 4/5 | patches=45 | loss=0.2073 +[04:55:51][ 198s][FT] round 3/6 epoch 5/5 | patches=45 | loss=0.2152 +[04:55:52][ 199s][AL] round 3/6 | labeled=45 | selection_seconds=3.1000 round_seconds=11.1000 pseudo_student_dice=0.9141 pseudo_ensemble_dice=0.9502 consistency_scale=1.0000 val_dice=0.9326 val_vi=0.0508 val_ece=0.0018 +[04:55:54][ 202s][AL] round 4: 278/1810 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:55:56][ 204s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.954, uniform=0.078, top1=0.954, top2=0.954, top3=0.954 -> using mixture +[04:55:56][ 204s][WEIGHTS] round 4: teacher Dice on 60 labeled patches + 0 blank (345564 fg px): fly-mito=0.338, kasthuri=0.000, lucchi=0.954, dros-vnc=0.011 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[04:55:57][ 205s][CONS] round 4: on the 15 new patches student Dice=0.925 vs ensemble 0.949 -> pseudo-labels from the ensemble | consistency x1.00 +[04:55:59][ 206s][FT] round 4/6 epoch 1/5 | patches=60 | loss=0.2037 +[04:56:00][ 208s][FT] round 4/6 epoch 2/5 | patches=60 | loss=0.2460 +[04:56:01][ 209s][FT] round 4/6 epoch 3/5 | patches=60 | loss=0.2196 +[04:56:03][ 210s][FT] round 4/6 epoch 4/5 | patches=60 | loss=0.1920 +[04:56:04][ 212s][FT] round 4/6 epoch 5/5 | patches=60 | loss=0.1549 +[04:56:05][ 213s][AL] round 4/6 | labeled=60 | selection_seconds=3.0000 round_seconds=12.6000 pseudo_student_dice=0.9252 pseudo_ensemble_dice=0.9490 consistency_scale=1.0000 val_dice=0.9398 val_vi=0.0454 val_ece=0.0015 +[04:56:08][ 215s][AL] round 5: 277/1795 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:56:10][ 218s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.955, uniform=0.091, top1=0.955, top2=0.955, top3=0.955 -> using mixture +[04:56:10][ 218s][WEIGHTS] round 5: teacher Dice on 75 labeled patches + 0 blank (438284 fg px): fly-mito=0.342, kasthuri=0.000, lucchi=0.955, dros-vnc=0.014 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[04:56:11][ 219s][CONS] round 5: on the 15 new patches student Dice=0.953 vs ensemble 0.956 -> pseudo-labels from the ensemble | consistency x1.00 +[04:56:13][ 220s][FT] round 5/6 epoch 1/5 | patches=75 | loss=0.1611 +[04:56:14][ 222s][FT] round 5/6 epoch 2/5 | patches=75 | loss=0.1646 +[04:56:16][ 224s][FT] round 5/6 epoch 3/5 | patches=75 | loss=0.1491 +[04:56:18][ 225s][FT] round 5/6 epoch 4/5 | patches=75 | loss=0.1619 +[04:56:19][ 227s][FT] round 5/6 epoch 5/5 | patches=75 | loss=0.1655 +[04:56:20][ 228s][AL] round 5/6 | labeled=75 | selection_seconds=3.0000 round_seconds=14.3000 pseudo_student_dice=0.9526 pseudo_ensemble_dice=0.9564 consistency_scale=1.0000 val_dice=0.9578 val_vi=0.0420 val_ece=0.0013 +[04:56:23][ 230s][AL] round 6: 279/1780 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[04:56:26][ 233s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.952, uniform=0.083, top1=0.952, top2=0.952, top3=0.952 -> using mixture +[04:56:26][ 233s][WEIGHTS] round 6: teacher Dice on 90 labeled patches + 0 blank (528615 fg px): fly-mito=0.343, kasthuri=0.000, lucchi=0.952, dros-vnc=0.012 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[04:56:26][ 234s][CONS] round 6: on the 15 new patches student Dice=0.934 vs ensemble 0.938 -> pseudo-labels from the ensemble | consistency x1.00 +[04:56:28][ 236s][FT] round 6/6 epoch 1/5 | patches=90 | loss=0.1728 +[04:56:30][ 238s][FT] round 6/6 epoch 2/5 | patches=90 | loss=0.1560 +[04:56:32][ 240s][FT] round 6/6 epoch 3/5 | patches=90 | loss=0.1586 +[04:56:34][ 242s][FT] round 6/6 epoch 4/5 | patches=90 | loss=0.1581 +[04:56:36][ 243s][FT] round 6/6 epoch 5/5 | patches=90 | loss=0.1437 +[04:56:37][ 244s][AL] round 6/6 | labeled=90 | selection_seconds=3.1000 round_seconds=15.8000 pseudo_student_dice=0.9344 pseudo_ensemble_dice=0.9375 consistency_scale=1.0000 val_dice=0.9595 val_vi=0.0418 val_ece=0.0013 +[04:56:38][ 246s][FT] final 1/10 epoch 1/1 | patches=90 | loss=0.1452 +[04:56:40][ 248s][FT] final 2/10 epoch 1/1 | patches=90 | loss=0.1429 +[04:56:42][ 250s][FT] final 3/10 epoch 1/1 | patches=90 | loss=0.1381 +[04:56:44][ 252s][FT] final 4/10 epoch 1/1 | patches=90 | loss=0.1253 +[04:56:46][ 254s][FT] final 5/10 epoch 1/1 | patches=90 | loss=0.1300 +[04:56:48][ 255s][FT] final 6/10 epoch 1/1 | patches=90 | loss=0.1288 +[04:56:50][ 257s][FT] final 7/10 epoch 1/1 | patches=90 | loss=0.1239 +[04:56:52][ 259s][FT] final 8/10 epoch 1/1 | patches=90 | loss=0.1226 +[04:56:53][ 261s][FT] final 9/10 epoch 1/1 | patches=90 | loss=0.1298 +[04:56:55][ 263s][FT] final 10/10 epoch 1/1 | patches=90 | loss=0.1247 +[04:56:56][ 264s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=18.8000 val_dice=0.9618 val_vi=0.0397 val_ece=0.0014 +[04:56:56][ 264s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[04:56:59][ 266s][TEST] raw student (no EMA, no TTA): patch_dice=0.9069 patch_vi=0.0742 +[04:56:59][ 267s][TEST] final model without TTA: patch_dice=0.9069 patch_vi=0.0742 +[04:56:59][ 267s][RESOURCES] {"phases": {"setup": {"seconds": 0.65, "peak_cpu_rss_mb": 1025.7, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 8.17, "peak_cpu_rss_mb": 1803.6, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 93.58, "peak_cpu_rss_mb": 2393.2, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4148.0, "kd_epochs": 5}, "active_learning": {"seconds": 160.27, "peak_cpu_rss_mb": 2666.5, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4156.0, "rounds": 6}, "inference": {"seconds": 3.07, "peak_cpu_rss_mb": 2666.5, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4156.0, "patches": 175, "images": 5, "patches_per_second": 57.0, "seconds_per_image": 0.614}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 2666.5, "total_seconds": 265.7} +[04:56:59][ 267s][TEST] patch_dice=0.8927 patch_iou=0.8473 patch_precision=0.9653 patch_recall=0.8726 patch_vi=0.0891 image_dice=0.9065 image_iou=0.8326 image_precision=0.9662 image_recall=0.8623 image_vi=0.1107 cal_ece=0.0079 cal_nll=0.0701 cal_brier=0.0108 cal_mean_confidence=0.9952 cal_accuracy=0.9876 cal_predicted_prevalence=0.0819 cal_true_prevalence=0.0886 cal_mean_probability=0.0811 +[04:56:59][ 267s][DONE] multiclass macs-v11-norestart-acq-entropy b5 s1 | test patch Dice=0.8927 VI=0.0891 | image Dice=0.9065 VI=0.1107 | 267s diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/result.json b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/result.json new file mode 100644 index 0000000000000000000000000000000000000000..e48c61cb2552f62c21a332c1a7ebd43dfc04ddc2 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/result.json @@ -0,0 +1,647 @@ +{ + "target": "multiclass", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 1, + "config": { + "task": "mitochondria", + "target": "multiclass", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 1, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "entropy", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "multiclass", + "--variants", + "macs-v11-norestart-acq-entropy", + "--budgets", + "5", + "--seeds", + "1", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "alphas": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "mixture": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "per_source": [ + 0.014321466605441749, + 0.018762803993963635, + 0.9619825923792787, + 0.004933137021315701 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.004128299797828663, + "mmd2_mixture": 0.005874306045373645, + "mmd2_uniform": 0.24840696635360943, + "mmd2_single_source": [ + 0.6122003026141286, + 0.5667974971994383, + 0.004845544051820361, + 0.668933003056658 + ], + "median_squared_distance": 0.20473960170977523, + "gammas": [ + 1.221063233064128, + 2.442126466128256, + 4.884252932256512, + 9.768505864513024, + 19.537011729026048 + ], + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6529830716388123, + 0.6112535306587584, + 0.0030196383578795505, + 0.8176329290240223 + ], + "standardized": [ + 0.4264461417532283, + 0.2913877286688084, + -1.6771722494431436, + 0.9593383790211064 + ], + "tau": 0.5, + "max_weight": 0.9619825923792787, + "entropy": 0.19889667121630067, + "effective_sources": 1.2200558924950622, + "min_weight": 0.004933137021315701 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "lucchi-pp", + "seconds": 1.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.25842773603204217, + 0.17499716869513837, + 0.9675530639716557, + 0.17625312705123047 + ], + "ensemble_val_dice": { + "mixture": 0.9673733701024737, + "mixture@raw": 0.9675071736744472, + "mixture@logit": 0.9667425331047603, + "per_source": 0.9673222165448325, + "per_source@raw": 0.967496030160359, + "per_source@logit": 0.9664733615943364, + "uniform": 0.24779489094285445, + "uniform@raw": 0.20821287741695416, + "uniform@logit": 0.19508316544090382 + }, + "val_patches": 350 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 3.7969236373901367, + 2.767761468887329 + ], + "top_source": "lucchi-pp", + "weight_stats": { + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.39999999999999997, + "spearman_weight_taxonomy": -0.8944271909999159, + "top_source": "lucchi-pp", + "top_source_is_best_teacher": true + }, + "init_teacher": "lucchi-pp", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 93.6, + "round0_val": { + "val_dice": 0.9623668424391527, + "val_vi": 0.04334010872659934, + "val_ece": 0.009792624342429557 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 15, + "selection_seconds": 13.7, + "round_seconds": 72.7, + "teacher_dice_labeled": [ + 0.2913624549969472, + 1.3699193117525378e-05, + 0.9671381891880713, + 0.017978534308470944 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 72996, + "evidence_nonempty_patches": 14, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9663568811994745, + "restart_teacher_dice": 0.9671381891880713, + "weights": [ + 1.3822181833298221e-05, + 7.71346360076235e-07, + 0.9999847642529057, + 6.422189008040626e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9671381891880713, + "uniform": 0.17358948382174527, + "top1": 0.9671381891880713, + "top2": 0.9671381891880713, + "top3": 0.9671381891880713 + }, + "val_dice": 0.9319506625786227, + "val_vi": 0.04937648161721857, + "val_ece": 0.0023662851884014673 + }, + { + "round": 2, + "n_labeled": 30, + "selection_seconds": 3.0, + "round_seconds": 9.4, + "teacher_dice_labeled": [ + 0.32987697574801783, + 6.160708235018698e-06, + 0.9595826814659841, + 0.008118922297317999 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 162318, + "evidence_nonempty_patches": 29, + "first_correction": false, + "restart": "", + "weights": [ + 2.191051878525161e-05, + 8.31814172563099e-07, + 0.9999766300814678, + 6.275855744537276e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9595826814659841, + "uniform": 0.08355321020228672, + "top1": 0.9595826814659841, + "top2": 0.9595826814659841, + "top3": 0.9595826814659841 + }, + "pseudo_student_dice": 0.9325759730457042, + "pseudo_ensemble_dice": 0.9534601805505899, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9256289309396039, + "val_vi": 0.052683010793089916, + "val_ece": 0.0028720350159565197 + }, + { + "round": 3, + "n_labeled": 45, + "selection_seconds": 3.1, + "round_seconds": 11.1, + "teacher_dice_labeled": [ + 0.3308303446901773, + 3.943062182090611e-06, + 0.9562169377564641, + 0.007531692766592096 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 253609, + "evidence_nonempty_patches": 44, + "first_correction": false, + "restart": "", + "weights": [ + 2.2877569772798032e-05, + 8.602674360138689e-07, + 0.9999756168958046, + 6.452669866354808e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9562190018905387, + "uniform": 0.07236612096614833, + "top1": 0.9562169377564641, + "top2": 0.9562190018905387, + "top3": 0.9562190018905387 + }, + "pseudo_student_dice": 0.9141349146508081, + "pseudo_ensemble_dice": 0.9502169507951191, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.932625614525691, + "val_vi": 0.050775190582086224, + "val_ece": 0.0018120487925275953 + }, + { + "round": 4, + "n_labeled": 60, + "selection_seconds": 3.0, + "round_seconds": 12.6, + "teacher_dice_labeled": [ + 0.3379793039672823, + 2.8938115839277703e-06, + 0.9543185502640407, + 0.011366903992362636 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 345564, + "evidence_nonempty_patches": 59, + "first_correction": false, + "restart": "", + "weights": [ + 2.5043847571946893e-05, + 8.767434993837445e-07, + 0.9999733960652449, + 6.83343683689413e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9543200761113186, + "uniform": 0.07789769930503605, + "top1": 0.9543185502640407, + "top2": 0.9543200761113186, + "top3": 0.9543200761113186 + }, + "pseudo_student_dice": 0.9252031888434891, + "pseudo_ensemble_dice": 0.9489748771132895, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9398434771253794, + "val_vi": 0.04541818295524556, + "val_ece": 0.0015108771491320813 + }, + { + "round": 5, + "n_labeled": 75, + "selection_seconds": 3.0, + "round_seconds": 14.3, + "teacher_dice_labeled": [ + 0.3417825878532159, + 2.2816204068129187e-06, + 0.9547577491192346, + 0.014399384274088578 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 438284, + "evidence_nonempty_patches": 73, + "first_correction": false, + "restart": "", + "weights": [ + 2.5900652743231587e-05, + 8.72895191563596e-07, + 0.9999725251561902, + 7.012958749669605e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9547589556882025, + "uniform": 0.09115317112259874, + "top1": 0.9547577491192346, + "top2": 0.9547589556882025, + "top3": 0.9547589556882025 + }, + "pseudo_student_dice": 0.9526084339635942, + "pseudo_ensemble_dice": 0.9564268467204102, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9577642455185371, + "val_vi": 0.04203404017915722, + "val_ece": 0.001263884740347842 + }, + { + "round": 6, + "n_labeled": 90, + "selection_seconds": 3.1, + "round_seconds": 15.8, + "teacher_dice_labeled": [ + 0.34315345671338743, + 1.8917323728377498e-06, + 0.9518143535011094, + 0.012029558020394432 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 528615, + "evidence_nonempty_patches": 88, + "first_correction": false, + "restart": "", + "weights": [ + 2.7042498407418326e-05, + 8.989652489251487e-07, + 0.9999713532073192, + 7.053290243927761e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9518153564689537, + "uniform": 0.08274670235414014, + "top1": 0.9518143535011094, + "top2": 0.9518153564689537, + "top3": 0.9518153564689537 + }, + "pseudo_student_dice": 0.9344471319594173, + "pseudo_ensemble_dice": 0.9375471919727503, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9594637922995282, + "val_vi": 0.0418189900151767, + "val_ece": 0.001297739111801087 + }, + { + "round": 7, + "n_labeled": 90, + "selection_seconds": 0.0, + "round_seconds": 18.8, + "final_consolidation": true, + "val_dice": 0.9617829917476898, + "val_vi": 0.03969062607765164, + "val_ece": 0.0014369260176021873 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.15377629786596558, + "auroc_top20_error": 0.26876984328797376 + }, + "al_seconds": 160.3, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 2.7042498407418326e-05, + 8.989652489251487e-07, + 0.9999713532073192, + 7.053290243927761e-07 + ], + "weight_stats_final": { + "max_weight": 0.9999713532073192, + "entropy": 0.000335588227584204, + "effective_sources": 1.000335644543613, + "min_weight": 7.053290243927761e-07 + }, + "top_source_final": "lucchi-pp", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.906926864701326, + "patch_iou": 0.8708886953441158, + "patch_precision": 0.9746803112520882, + "patch_recall": 0.8886945926356378, + "patch_vi": 0.07415238886546492, + "n_patches": 175, + "cal_ece": 0.0075636897846711295, + "cal_nll": 0.07369072944435516, + "cal_brier": 0.009159593145892403, + "cal_mean_confidence": 0.9972980662031419, + "cal_accuracy": 0.9898549107142857, + "cal_predicted_prevalence": 0.08227120535714286, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.082003879415212 + }, + "test_notta": { + "patch_dice": 0.906926864701326, + "patch_iou": 0.8708886953441158, + "patch_precision": 0.9746803112520882, + "patch_recall": 0.8886945926356378, + "patch_vi": 0.07415238886546492, + "n_patches": 175, + "cal_ece": 0.0075636897846711295, + "cal_nll": 0.07369072944435516, + "cal_brier": 0.009159593145892403, + "cal_mean_confidence": 0.9972980662031419, + "cal_accuracy": 0.9898549107142857, + "cal_predicted_prevalence": 0.08227120535714286, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.082003879415212 + }, + "test": { + "patch_dice": 0.8927127447434549, + "patch_iou": 0.8473055184797504, + "patch_precision": 0.9653497122409346, + "patch_recall": 0.8725511742573188, + "patch_vi": 0.08912678515855041, + "n_patches": 175, + "image_dice": 0.9065200041570909, + "image_iou": 0.8326441154650153, + "image_precision": 0.9661788429840165, + "image_recall": 0.8622996239232489, + "image_vi": 0.11070523460868953, + "n_images": 5, + "cal_ece": 0.007945259294557274, + "cal_nll": 0.07014566486453372, + "cal_brier": 0.010790662233450677, + "cal_mean_confidence": 0.9951933837287379, + "cal_accuracy": 0.9876339285714286, + "cal_predicted_prevalence": 0.08190290178571429, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08109014216565821 + }, + "eval_seconds": 3.1, + "total_seconds": 266.9, + "resources": { + "phases": { + "setup": { + "seconds": 0.65, + "peak_cpu_rss_mb": 1025.7, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 8.17, + "peak_cpu_rss_mb": 1803.6, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 93.58, + "peak_cpu_rss_mb": 2393.2, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4148.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 160.27, + "peak_cpu_rss_mb": 2666.5, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4156.0, + "rounds": 6 + }, + "inference": { + "seconds": 3.07, + "peak_cpu_rss_mb": 2666.5, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4156.0, + "patches": 175, + "images": 5, + "patches_per_second": 57.0, + "seconds_per_image": 0.614 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 2666.5, + "total_seconds": 265.7 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/multiclass/macs-v11-norestart-acq-entropy_b5_s1.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/rounds.csv b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..28882b8c7360e1a553eeb1b29919246867d1fd9f --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-1/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,15,13.7,72.7,"[0.2913624549969472, 1.3699193117525378e-05, 0.9671381891880713, 0.017978534308470944]",,72996,14,True,kept,0.9663568811994745,0.9671381891880713,"[1.3822181833298221e-05, 7.71346360076235e-07, 0.9999847642529057, 6.422189008040626e-07]",mixture,"{'mixture': 0.9671381891880713, 'uniform': 0.17358948382174527, 'top1': 0.9671381891880713, 'top2': 0.9671381891880713, 'top3': 0.9671381891880713}",0.9319506625786227,0.04937648161721857,0.0023662851884014673,,,,, +2,30,3.0,9.4,"[0.32987697574801783, 6.160708235018698e-06, 0.9595826814659841, 0.008118922297317999]",,162318,29,False,,,,"[2.191051878525161e-05, 8.31814172563099e-07, 0.9999766300814678, 6.275855744537276e-07]",mixture,"{'mixture': 0.9595826814659841, 'uniform': 0.08355321020228672, 'top1': 0.9595826814659841, 'top2': 0.9595826814659841, 'top3': 0.9595826814659841}",0.9256289309396039,0.052683010793089916,0.0028720350159565197,0.9325759730457042,0.9534601805505899,ensemble,1.0, +3,45,3.1,11.1,"[0.3308303446901773, 3.943062182090611e-06, 0.9562169377564641, 0.007531692766592096]",,253609,44,False,,,,"[2.2877569772798032e-05, 8.602674360138689e-07, 0.9999756168958046, 6.452669866354808e-07]",mixture,"{'mixture': 0.9562190018905387, 'uniform': 0.07236612096614833, 'top1': 0.9562169377564641, 'top2': 0.9562190018905387, 'top3': 0.9562190018905387}",0.932625614525691,0.050775190582086224,0.0018120487925275953,0.9141349146508081,0.9502169507951191,ensemble,1.0, +4,60,3.0,12.6,"[0.3379793039672823, 2.8938115839277703e-06, 0.9543185502640407, 0.011366903992362636]",,345564,59,False,,,,"[2.5043847571946893e-05, 8.767434993837445e-07, 0.9999733960652449, 6.83343683689413e-07]",mixture,"{'mixture': 0.9543200761113186, 'uniform': 0.07789769930503605, 'top1': 0.9543185502640407, 'top2': 0.9543200761113186, 'top3': 0.9543200761113186}",0.9398434771253794,0.04541818295524556,0.0015108771491320813,0.9252031888434891,0.9489748771132895,ensemble,1.0, +5,75,3.0,14.3,"[0.3417825878532159, 2.2816204068129187e-06, 0.9547577491192346, 0.014399384274088578]",,438284,73,False,,,,"[2.5900652743231587e-05, 8.72895191563596e-07, 0.9999725251561902, 7.012958749669605e-07]",mixture,"{'mixture': 0.9547589556882025, 'uniform': 0.09115317112259874, 'top1': 0.9547577491192346, 'top2': 0.9547589556882025, 'top3': 0.9547589556882025}",0.9577642455185371,0.04203404017915722,0.001263884740347842,0.9526084339635942,0.9564268467204102,ensemble,1.0, +6,90,3.1,15.8,"[0.34315345671338743, 1.8917323728377498e-06, 0.9518143535011094, 0.012029558020394432]",,528615,88,False,,,,"[2.7042498407418326e-05, 8.989652489251487e-07, 0.9999713532073192, 7.053290243927761e-07]",mixture,"{'mixture': 0.9518153564689537, 'uniform': 0.08274670235414014, 'top1': 0.9518143535011094, 'top2': 0.9518153564689537, 'top3': 0.9518153564689537}",0.9594637922995282,0.0418189900151767,0.001297739111801087,0.9344471319594173,0.9375471919727503,ensemble,1.0, +7,90,0.0,18.8,,,,,,,,,,,,0.9617829917476898,0.03969062607765164,0.0014369260176021873,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/log.txt b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..e91a9e46f98e73b27667f0ea798f24f1199748a8 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/log.txt @@ -0,0 +1,103 @@ +[08:21:28][ 0s][START] target=multiclass variant=macs-v11-norestart-acq-entropy budget=5.0% seed=2 | {"task": "mitochondria", "target": "multiclass", "variant": "macs-v11-norestart-acq-entropy", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "entropy", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[08:21:28][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[08:21:28][ 0s][DATA] multiclass: train=1855 val=350 test=175 patches +[08:21:29][ 1s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, dros-vnc=2.77 +[08:21:29][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[08:21:30][ 2s][WEIGHTS] mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 | eff.sources=1.19 | MMD2 mix=0.0059 uniform=0.2484 +[08:21:30][ 2s][WEIGHTS] per-source: fly-mito=0.014, kasthuri=0.019, lucchi=0.962, dros-vnc=0.005 | min-MMD source (NeuroADDA rule) = lucchi +[08:21:36][ 8s][TEACHERS] target-val Dice: fly-mito=0.258, kasthuri=0.175, lucchi=0.968, dros-vnc=0.176 +[08:21:36][ 8s][TEACHERS] ensemble target-val Dice: mixture=0.967, mixture@raw=0.968, mixture@logit=0.967, per_source=0.967, per_source@raw=0.967, per_source@logit=0.966, uniform=0.248, uniform@raw=0.208, uniform@logit=0.195 +[08:21:36][ 8s][WEIGHTS] using mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 +[08:21:36][ 8s][INIT] student initialised from teacher lucchi (weight 0.968) +[08:21:37][ 9s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[08:21:37][ 9s][KD] distilling the mixture ensemble (prob space) into the student +[08:21:48][ 20s][KD] soft labels for 1855 patches from 4 teachers in 11s +[08:22:05][ 37s][KD] epoch 1/5 | patches=1855 | loss=0.1028 | 17s +[08:22:22][ 54s][KD] epoch 2/5 | patches=1855 | loss=0.0735 | 17s +[08:22:39][ 71s][KD] epoch 3/5 | patches=1855 | loss=0.0710 | 17s +[08:22:55][ 87s][KD] epoch 4/5 | patches=1855 | loss=0.0695 | 16s +[08:23:12][ 104s][KD] epoch 5/5 | patches=1855 | loss=0.0687 | 17s +[08:23:13][ 105s][ROUND0] before any target label: val_dice=0.9665 val_vi=0.0425 +[08:23:13][ 105s][AL] patches=1855 (pool 1855 after dropping blank; free blank negatives: True) | budget=92 (5.0%) | rounds=6 | per_round=15 | acquisition=entropy | first_round=stratified +[08:23:16][ 108s][AL] round-1 uncertainty quality on 1855 patches: {'spearman_uncertainty_error': -0.12831241173071842, 'auroc_top20_error': 0.2808193052942074} +[08:23:27][ 119s][AL] round 1: stratified cold start: 8 patches from the membrane-rich half (ensemble prevalence >= 0.071) + 7 from the membrane-poor half of 1855 candidates +[08:23:28][ 119s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.971, uniform=0.135, top1=0.971, top2=0.971, top3=0.971 -> using mixture +[08:23:28][ 120s][WEIGHTS] round 1: student kept (labeled Dice 0.973 vs best teacher lucchi 0.971); distilling the corrected ensemble into it +[08:23:28][ 120s][WEIGHTS] round 1: teacher Dice on 15 labeled patches + 0 blank (92186 fg px): fly-mito=0.353, kasthuri=0.000, lucchi=0.971, dros-vnc=0.006 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:23:39][ 130s][KD] soft labels for 1840 patches from 4 teachers in 11s +[08:23:55][ 147s][KD] epoch 1/2 | patches=1840 | loss=0.0450 | 17s +[08:24:12][ 164s][KD] epoch 2/2 | patches=1840 | loss=0.0441 | 17s +[08:24:23][ 175s][CONS] pseudo-labels for 1840 unlabeled patches from the current ensemble in 11s +[08:24:24][ 176s][FT] round 1/6 epoch 1/5 | patches=15 | loss=0.1525 +[08:24:24][ 176s][FT] round 1/6 epoch 2/5 | patches=15 | loss=0.1169 +[08:24:25][ 177s][FT] round 1/6 epoch 3/5 | patches=15 | loss=0.1214 +[08:24:25][ 177s][FT] round 1/6 epoch 4/5 | patches=15 | loss=0.1068 +[08:24:26][ 178s][FT] round 1/6 epoch 5/5 | patches=15 | loss=0.0952 +[08:24:27][ 179s][AL] round 1/6 | labeled=15 | selection_seconds=13.9000 round_seconds=73.1000 restart_student_dice=0.9733 restart_teacher_dice=0.9708 val_dice=0.9561 val_vi=0.0477 val_ece=0.0025 +[08:24:30][ 181s][AL] round 2: 285/1840 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:24:31][ 183s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.960, uniform=0.086, top1=0.960, top2=0.960, top3=0.960 -> using mixture +[08:24:31][ 183s][WEIGHTS] round 2: teacher Dice on 30 labeled patches + 0 blank (182882 fg px): fly-mito=0.346, kasthuri=0.000, lucchi=0.960, dros-vnc=0.006 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:24:32][ 184s][CONS] round 2: on the 15 new patches student Dice=0.926 vs ensemble 0.949 -> pseudo-labels from the ensemble | consistency x1.00 +[08:24:33][ 185s][FT] round 2/6 epoch 1/5 | patches=30 | loss=0.1766 +[08:24:34][ 185s][FT] round 2/6 epoch 2/5 | patches=30 | loss=0.1573 +[08:24:34][ 186s][FT] round 2/6 epoch 3/5 | patches=30 | loss=0.1571 +[08:24:35][ 187s][FT] round 2/6 epoch 4/5 | patches=30 | loss=0.1519 +[08:24:36][ 188s][FT] round 2/6 epoch 5/5 | patches=30 | loss=0.1661 +[08:24:37][ 189s][AL] round 2/6 | labeled=30 | selection_seconds=3.1000 round_seconds=9.4000 pseudo_student_dice=0.9260 pseudo_ensemble_dice=0.9488 consistency_scale=1.0000 val_dice=0.9485 val_vi=0.0441 val_ece=0.0012 +[08:24:40][ 192s][AL] round 3: 280/1825 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:24:41][ 193s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.957, uniform=0.067, top1=0.957, top2=0.957, top3=0.957 -> using mixture +[08:24:41][ 193s][WEIGHTS] round 3: teacher Dice on 45 labeled patches + 0 blank (267747 fg px): fly-mito=0.345, kasthuri=0.000, lucchi=0.957, dros-vnc=0.004 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:24:42][ 194s][CONS] round 3: on the 15 new patches student Dice=0.946 vs ensemble 0.952 -> pseudo-labels from the ensemble | consistency x1.00 +[08:24:43][ 195s][FT] round 3/6 epoch 1/5 | patches=45 | loss=0.1619 +[08:24:44][ 196s][FT] round 3/6 epoch 2/5 | patches=45 | loss=0.2008 +[08:24:46][ 198s][FT] round 3/6 epoch 3/5 | patches=45 | loss=0.1771 +[08:24:47][ 199s][FT] round 3/6 epoch 4/5 | patches=45 | loss=0.1778 +[08:24:48][ 200s][FT] round 3/6 epoch 5/5 | patches=45 | loss=0.1765 +[08:24:49][ 201s][AL] round 3/6 | labeled=45 | selection_seconds=2.9000 round_seconds=10.9000 pseudo_student_dice=0.9456 pseudo_ensemble_dice=0.9521 consistency_scale=1.0000 val_dice=0.9247 val_vi=0.0477 val_ece=0.0016 +[08:24:52][ 203s][AL] round 4: 281/1810 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:24:53][ 205s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.954, uniform=0.053, top1=0.954, top2=0.954, top3=0.954 -> using mixture +[08:24:53][ 205s][WEIGHTS] round 4: teacher Dice on 60 labeled patches + 0 blank (353875 fg px): fly-mito=0.345, kasthuri=0.000, lucchi=0.954, dros-vnc=0.003 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:24:54][ 206s][CONS] round 4: on the 15 new patches student Dice=0.933 vs ensemble 0.944 -> pseudo-labels from the ensemble | consistency x1.00 +[08:24:56][ 208s][FT] round 4/6 epoch 1/5 | patches=60 | loss=0.1750 +[08:24:57][ 209s][FT] round 4/6 epoch 2/5 | patches=60 | loss=0.1845 +[08:24:59][ 211s][FT] round 4/6 epoch 3/5 | patches=60 | loss=0.1626 +[08:25:00][ 212s][FT] round 4/6 epoch 4/5 | patches=60 | loss=0.1623 +[08:25:01][ 213s][FT] round 4/6 epoch 5/5 | patches=60 | loss=0.1496 +[08:25:02][ 214s][AL] round 4/6 | labeled=60 | selection_seconds=3.0000 round_seconds=12.6000 pseudo_student_dice=0.9328 pseudo_ensemble_dice=0.9444 consistency_scale=1.0000 val_dice=0.9462 val_vi=0.0426 val_ece=0.0012 +[08:25:05][ 217s][AL] round 5: 285/1795 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:25:07][ 219s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.951, uniform=0.047, top1=0.951, top2=0.951, top3=0.951 -> using mixture +[08:25:07][ 219s][WEIGHTS] round 5: teacher Dice on 75 labeled patches + 0 blank (450137 fg px): fly-mito=0.351, kasthuri=0.000, lucchi=0.951, dros-vnc=0.005 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:25:08][ 220s][CONS] round 5: on the 15 new patches student Dice=0.930 vs ensemble 0.939 -> pseudo-labels from the ensemble | consistency x1.00 +[08:25:10][ 222s][FT] round 5/6 epoch 1/5 | patches=75 | loss=0.1713 +[08:25:11][ 223s][FT] round 5/6 epoch 2/5 | patches=75 | loss=0.1599 +[08:25:13][ 225s][FT] round 5/6 epoch 3/5 | patches=75 | loss=0.1535 +[08:25:15][ 227s][FT] round 5/6 epoch 4/5 | patches=75 | loss=0.1368 +[08:25:16][ 228s][FT] round 5/6 epoch 5/5 | patches=75 | loss=0.1359 +[08:25:17][ 229s][AL] round 5/6 | labeled=75 | selection_seconds=3.0000 round_seconds=14.3000 pseudo_student_dice=0.9300 pseudo_ensemble_dice=0.9387 consistency_scale=1.0000 val_dice=0.9540 val_vi=0.0443 val_ece=0.0017 +[08:25:20][ 232s][AL] round 6: 283/1780 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:25:23][ 234s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.951, uniform=0.067, top1=0.951, top2=0.951, top3=0.951 -> using mixture +[08:25:23][ 234s][WEIGHTS] round 6: teacher Dice on 90 labeled patches + 0 blank (543563 fg px): fly-mito=0.352, kasthuri=0.000, lucchi=0.951, dros-vnc=0.008 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:25:24][ 236s][CONS] round 6: on the 15 new patches student Dice=0.951 vs ensemble 0.951 -> pseudo-labels from the ensemble | consistency x1.00 +[08:25:25][ 237s][FT] round 6/6 epoch 1/5 | patches=90 | loss=0.1411 +[08:25:27][ 239s][FT] round 6/6 epoch 2/5 | patches=90 | loss=0.1454 +[08:25:29][ 241s][FT] round 6/6 epoch 3/5 | patches=90 | loss=0.1529 +[08:25:31][ 243s][FT] round 6/6 epoch 4/5 | patches=90 | loss=0.1530 +[08:25:33][ 245s][FT] round 6/6 epoch 5/5 | patches=90 | loss=0.1315 +[08:25:34][ 246s][AL] round 6/6 | labeled=90 | selection_seconds=3.0000 round_seconds=15.8000 pseudo_student_dice=0.9507 pseudo_ensemble_dice=0.9513 consistency_scale=1.0000 val_dice=0.9486 val_vi=0.0431 val_ece=0.0017 +[08:25:36][ 248s][FT] final 1/10 epoch 1/1 | patches=90 | loss=0.1232 +[08:25:38][ 249s][FT] final 2/10 epoch 1/1 | patches=90 | loss=0.1208 +[08:25:39][ 251s][FT] final 3/10 epoch 1/1 | patches=90 | loss=0.1167 +[08:25:41][ 253s][FT] final 4/10 epoch 1/1 | patches=90 | loss=0.1204 +[08:25:43][ 255s][FT] final 5/10 epoch 1/1 | patches=90 | loss=0.1185 +[08:25:45][ 257s][FT] final 6/10 epoch 1/1 | patches=90 | loss=0.1189 +[08:25:47][ 259s][FT] final 7/10 epoch 1/1 | patches=90 | loss=0.1120 +[08:25:49][ 261s][FT] final 8/10 epoch 1/1 | patches=90 | loss=0.1023 +[08:25:51][ 263s][FT] final 9/10 epoch 1/1 | patches=90 | loss=0.1081 +[08:25:53][ 265s][FT] final 10/10 epoch 1/1 | patches=90 | loss=0.1173 +[08:25:54][ 265s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=19.0000 val_dice=0.9496 val_vi=0.0451 val_ece=0.0017 +[08:25:54][ 265s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:25:56][ 268s][TEST] raw student (no EMA, no TTA): patch_dice=0.9057 patch_vi=0.0768 +[08:25:57][ 269s][TEST] final model without TTA: patch_dice=0.9057 patch_vi=0.0768 +[08:25:57][ 269s][RESOURCES] {"phases": {"setup": {"seconds": 0.58, "peak_cpu_rss_mb": 1025.8, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 8.11, "peak_cpu_rss_mb": 1803.7, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 95.2, "peak_cpu_rss_mb": 2392.9, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4148.0, "kd_epochs": 5}, "active_learning": {"seconds": 160.7, "peak_cpu_rss_mb": 2666.4, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4156.0, "rounds": 6}, "inference": {"seconds": 3.12, "peak_cpu_rss_mb": 2666.4, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4156.0, "patches": 175, "images": 5, "patches_per_second": 56.1, "seconds_per_image": 0.624}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 2666.4, "total_seconds": 267.7} +[08:25:57][ 269s][TEST] patch_dice=0.8928 patch_iou=0.8475 patch_precision=0.9709 patch_recall=0.8723 patch_vi=0.0866 image_dice=0.9100 image_iou=0.8390 image_precision=0.9700 image_recall=0.8638 image_vi=0.1057 cal_ece=0.0076 cal_nll=0.0862 cal_brier=0.0104 cal_mean_confidence=0.9953 cal_accuracy=0.9880 cal_predicted_prevalence=0.0816 cal_true_prevalence=0.0886 cal_mean_probability=0.0812 +[08:25:57][ 269s][DONE] multiclass macs-v11-norestart-acq-entropy b5 s2 | test patch Dice=0.8928 VI=0.0866 | image Dice=0.9100 VI=0.1057 | 269s diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/result.json b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..7a3c4d380b7207b312aec3835f554eca7b5acff5 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/result.json @@ -0,0 +1,647 @@ +{ + "target": "multiclass", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 2, + "config": { + "task": "mitochondria", + "target": "multiclass", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "entropy", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "multiclass", + "--variants", + "macs-v11-norestart-acq-entropy", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "alphas": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "mixture": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "per_source": [ + 0.014321466605441749, + 0.018762803993963635, + 0.9619825923792787, + 0.004933137021315701 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.004128299797828663, + "mmd2_mixture": 0.005874306045373645, + "mmd2_uniform": 0.24840696635360943, + "mmd2_single_source": [ + 0.6122003026141286, + 0.5667974971994383, + 0.004845544051820361, + 0.668933003056658 + ], + "median_squared_distance": 0.20473960170977523, + "gammas": [ + 1.221063233064128, + 2.442126466128256, + 4.884252932256512, + 9.768505864513024, + 19.537011729026048 + ], + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6529830716388123, + 0.6112535306587584, + 0.0030196383578795505, + 0.8176329290240223 + ], + "standardized": [ + 0.4264461417532283, + 0.2913877286688084, + -1.6771722494431436, + 0.9593383790211064 + ], + "tau": 0.5, + "max_weight": 0.9619825923792787, + "entropy": 0.19889667121630067, + "effective_sources": 1.2200558924950622, + "min_weight": 0.004933137021315701 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "lucchi-pp", + "seconds": 1.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.25842773603204217, + 0.17499716869513837, + 0.9675530639716557, + 0.17625312705123047 + ], + "ensemble_val_dice": { + "mixture": 0.9673733701024737, + "mixture@raw": 0.9675071736744472, + "mixture@logit": 0.9667425331047603, + "per_source": 0.9673222165448325, + "per_source@raw": 0.967496030160359, + "per_source@logit": 0.9664733615943364, + "uniform": 0.24779489094285445, + "uniform@raw": 0.20821287741695416, + "uniform@logit": 0.19508316544090382 + }, + "val_patches": 350 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 3.7969236373901367, + 2.767761468887329 + ], + "top_source": "lucchi-pp", + "weight_stats": { + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.39999999999999997, + "spearman_weight_taxonomy": -0.8944271909999159, + "top_source": "lucchi-pp", + "top_source_is_best_teacher": true + }, + "init_teacher": "lucchi-pp", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 95.2, + "round0_val": { + "val_dice": 0.9664771075304173, + "val_vi": 0.04252684213583605, + "val_ece": 0.008563918584383703 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 15, + "selection_seconds": 13.9, + "round_seconds": 73.1, + "teacher_dice_labeled": [ + 0.3531319282015097, + 1.084751646110623e-05, + 0.9708165777595404, + 0.005894504591224218 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 92186, + "evidence_nonempty_patches": 15, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9733298226559371, + "restart_teacher_dice": 0.9708165777595404, + "weights": [ + 2.4709178198489446e-05, + 7.43459534993782e-07, + 0.9999739988035825, + 5.485586840192426e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9708165777595404, + "uniform": 0.13523545594303746, + "top1": 0.9708165777595404, + "top2": 0.9708165777595404, + "top3": 0.9708165777595404 + }, + "val_dice": 0.9561205283798695, + "val_vi": 0.04769281726133547, + "val_ece": 0.0024989405631414426 + }, + { + "round": 2, + "n_labeled": 30, + "selection_seconds": 3.1, + "round_seconds": 9.4, + "teacher_dice_labeled": [ + 0.3455985560355696, + 5.467976793906487e-06, + 0.9598918245374148, + 0.0062085393307423535 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 182882, + "evidence_nonempty_patches": 30, + "first_correction": false, + "restart": "", + "weights": [ + 2.5561506532048935e-05, + 8.292378888999875e-07, + 0.9999729954482146, + 6.138073644465558e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9598946918634302, + "uniform": 0.08623242623860028, + "top1": 0.9598918245374148, + "top2": 0.9598946918634302, + "top3": 0.9598946918634302 + }, + "pseudo_student_dice": 0.9260126061934777, + "pseudo_ensemble_dice": 0.9488028626834323, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.948504296422355, + "val_vi": 0.04412232683433521, + "val_ece": 0.0012063830287261372 + }, + { + "round": 3, + "n_labeled": 45, + "selection_seconds": 2.9, + "round_seconds": 10.9, + "teacher_dice_labeled": [ + 0.34475183208939014, + 3.7348551623168054e-06, + 0.9574048023556605, + 0.004348798211291224 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 267747, + "evidence_nonempty_patches": 44, + "first_correction": false, + "restart": "", + "weights": [ + 2.5984237486537435e-05, + 8.50104697956662e-07, + 0.9999725479883287, + 6.176694867959208e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.957406755278498, + "uniform": 0.06704928980092768, + "top1": 0.9574048023556605, + "top2": 0.957406755278498, + "top3": 0.957406755278498 + }, + "pseudo_student_dice": 0.9455882437490298, + "pseudo_ensemble_dice": 0.9521330060776063, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9246563993051506, + "val_vi": 0.04769887359826809, + "val_ece": 0.0015577263385053875 + }, + { + "round": 4, + "n_labeled": 60, + "selection_seconds": 3.0, + "round_seconds": 12.6, + "teacher_dice_labeled": [ + 0.34453338417655527, + 2.825848602335281e-06, + 0.9542391436909891, + 0.003292164816998516 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 353875, + "evidence_nonempty_patches": 59, + "first_correction": false, + "restart": "", + "weights": [ + 2.6761422551568005e-05, + 8.774379098590275e-07, + 0.9999717303054219, + 6.308341166648288e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9542406266884298, + "uniform": 0.053254519248753976, + "top1": 0.9542391436909891, + "top2": 0.9542406266884298, + "top3": 0.9542406266884298 + }, + "pseudo_student_dice": 0.9327660486308477, + "pseudo_ensemble_dice": 0.9443737449213094, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9462075115648259, + "val_vi": 0.04264386962610886, + "val_ece": 0.001244382501987774 + }, + { + "round": 5, + "n_labeled": 75, + "selection_seconds": 3.0, + "round_seconds": 14.3, + "teacher_dice_labeled": [ + 0.3506447158063697, + 2.221540949664325e-06, + 0.9510012595730923, + 0.005349749569694057 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 450137, + "evidence_nonempty_patches": 74, + "first_correction": false, + "restart": "", + "weights": [ + 2.9384016354757164e-05, + 9.063053744092813e-07, + 0.9999690445400026, + 6.651382682639803e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9510024371926146, + "uniform": 0.04746826520808437, + "top1": 0.9510012595730923, + "top2": 0.9510024371926146, + "top3": 0.9510024371926146 + }, + "pseudo_student_dice": 0.9300016521614002, + "pseudo_ensemble_dice": 0.9387111894102245, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9540271268249594, + "val_vi": 0.04425781279722695, + "val_ece": 0.0017131517242046632 + }, + { + "round": 6, + "n_labeled": 90, + "selection_seconds": 3.0, + "round_seconds": 15.8, + "teacher_dice_labeled": [ + 0.35159325856374474, + 1.82955834461561e-06, + 0.9510522100457386, + 0.008159101564789311 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 543563, + "evidence_nonempty_patches": 89, + "first_correction": false, + "restart": "", + "weights": [ + 2.96489438690486e-05, + 9.058399179916786e-07, + 0.9999687614755668, + 6.837406461312083e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9510541615311272, + "uniform": 0.06719139301873447, + "top1": 0.9510522100457386, + "top2": 0.9510541615311272, + "top3": 0.9510541615311272 + }, + "pseudo_student_dice": 0.9506762902800429, + "pseudo_ensemble_dice": 0.951293643858516, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9486477837073507, + "val_vi": 0.04313568054178121, + "val_ece": 0.0016851823814077627 + }, + { + "round": 7, + "n_labeled": 90, + "selection_seconds": 0.0, + "round_seconds": 19.0, + "final_consolidation": true, + "val_dice": 0.949564268705659, + "val_vi": 0.04505678306437529, + "val_ece": 0.0016889943583958036 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.12831241173071842, + "auroc_top20_error": 0.2808193052942074 + }, + "al_seconds": 160.7, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 2.96489438690486e-05, + 9.058399179916786e-07, + 0.9999687614755668, + 6.837406461312083e-07 + ], + "weight_stats_final": { + "max_weight": 0.9999687614755668, + "entropy": 0.00036267080757036686, + "effective_sources": 1.0003627365805787, + "min_weight": 6.837406461312083e-07 + }, + "top_source_final": "lucchi-pp", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.9057062479713137, + "patch_iou": 0.8687375509642382, + "patch_precision": 0.9752218886210253, + "patch_recall": 0.8913436716899186, + "patch_vi": 0.07678005182063208, + "n_patches": 175, + "cal_ece": 0.0077720425807492905, + "cal_nll": 0.08728789471292396, + "cal_brier": 0.00951070060690909, + "cal_mean_confidence": 0.997033995191103, + "cal_accuracy": 0.9892689732142858, + "cal_predicted_prevalence": 0.08261160714285715, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08224475975074996 + }, + "test_notta": { + "patch_dice": 0.9057062479713137, + "patch_iou": 0.8687375509642382, + "patch_precision": 0.9752218886210253, + "patch_recall": 0.8913436716899186, + "patch_vi": 0.07678005182063208, + "n_patches": 175, + "cal_ece": 0.0077720425807492905, + "cal_nll": 0.08728789471292396, + "cal_brier": 0.00951070060690909, + "cal_mean_confidence": 0.997033995191103, + "cal_accuracy": 0.9892689732142858, + "cal_predicted_prevalence": 0.08261160714285715, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08224475975074996 + }, + "test": { + "patch_dice": 0.8927861428615392, + "patch_iou": 0.8475407521973922, + "patch_precision": 0.9709491979900278, + "patch_recall": 0.8723434787635678, + "patch_vi": 0.08655481635308798, + "n_patches": 175, + "image_dice": 0.9100269582049542, + "image_iou": 0.8390254033199754, + "image_precision": 0.9699866655556232, + "image_recall": 0.8637524288959739, + "image_vi": 0.10568350437303162, + "n_images": 5, + "cal_ece": 0.007587271449733971, + "cal_nll": 0.08618297985345229, + "cal_brier": 0.01044600613952142, + "cal_mean_confidence": 0.9952512580664353, + "cal_accuracy": 0.9879743303571429, + "cal_predicted_prevalence": 0.08158482142857143, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08118676514872443 + }, + "eval_seconds": 3.1, + "total_seconds": 268.6, + "resources": { + "phases": { + "setup": { + "seconds": 0.58, + "peak_cpu_rss_mb": 1025.8, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 8.11, + "peak_cpu_rss_mb": 1803.7, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 95.2, + "peak_cpu_rss_mb": 2392.9, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4148.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 160.7, + "peak_cpu_rss_mb": 2666.4, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4156.0, + "rounds": 6 + }, + "inference": { + "seconds": 3.12, + "peak_cpu_rss_mb": 2666.4, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4156.0, + "patches": 175, + "images": 5, + "patches_per_second": 56.1, + "seconds_per_image": 0.624 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 2666.4, + "total_seconds": 267.7 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/multiclass/macs-v11-norestart-acq-entropy_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..7388a22a637dbbf4044948d41d3e410fc306b53f --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,15,13.9,73.1,"[0.3531319282015097, 1.084751646110623e-05, 0.9708165777595404, 0.005894504591224218]",,92186,15,True,kept,0.9733298226559371,0.9708165777595404,"[2.4709178198489446e-05, 7.43459534993782e-07, 0.9999739988035825, 5.485586840192426e-07]",mixture,"{'mixture': 0.9708165777595404, 'uniform': 0.13523545594303746, 'top1': 0.9708165777595404, 'top2': 0.9708165777595404, 'top3': 0.9708165777595404}",0.9561205283798695,0.04769281726133547,0.0024989405631414426,,,,, +2,30,3.1,9.4,"[0.3455985560355696, 5.467976793906487e-06, 0.9598918245374148, 0.0062085393307423535]",,182882,30,False,,,,"[2.5561506532048935e-05, 8.292378888999875e-07, 0.9999729954482146, 6.138073644465558e-07]",mixture,"{'mixture': 0.9598946918634302, 'uniform': 0.08623242623860028, 'top1': 0.9598918245374148, 'top2': 0.9598946918634302, 'top3': 0.9598946918634302}",0.948504296422355,0.04412232683433521,0.0012063830287261372,0.9260126061934777,0.9488028626834323,ensemble,1.0, +3,45,2.9,10.9,"[0.34475183208939014, 3.7348551623168054e-06, 0.9574048023556605, 0.004348798211291224]",,267747,44,False,,,,"[2.5984237486537435e-05, 8.50104697956662e-07, 0.9999725479883287, 6.176694867959208e-07]",mixture,"{'mixture': 0.957406755278498, 'uniform': 0.06704928980092768, 'top1': 0.9574048023556605, 'top2': 0.957406755278498, 'top3': 0.957406755278498}",0.9246563993051506,0.04769887359826809,0.0015577263385053875,0.9455882437490298,0.9521330060776063,ensemble,1.0, +4,60,3.0,12.6,"[0.34453338417655527, 2.825848602335281e-06, 0.9542391436909891, 0.003292164816998516]",,353875,59,False,,,,"[2.6761422551568005e-05, 8.774379098590275e-07, 0.9999717303054219, 6.308341166648288e-07]",mixture,"{'mixture': 0.9542406266884298, 'uniform': 0.053254519248753976, 'top1': 0.9542391436909891, 'top2': 0.9542406266884298, 'top3': 0.9542406266884298}",0.9462075115648259,0.04264386962610886,0.001244382501987774,0.9327660486308477,0.9443737449213094,ensemble,1.0, +5,75,3.0,14.3,"[0.3506447158063697, 2.221540949664325e-06, 0.9510012595730923, 0.005349749569694057]",,450137,74,False,,,,"[2.9384016354757164e-05, 9.063053744092813e-07, 0.9999690445400026, 6.651382682639803e-07]",mixture,"{'mixture': 0.9510024371926146, 'uniform': 0.04746826520808437, 'top1': 0.9510012595730923, 'top2': 0.9510024371926146, 'top3': 0.9510024371926146}",0.9540271268249594,0.04425781279722695,0.0017131517242046632,0.9300016521614002,0.9387111894102245,ensemble,1.0, +6,90,3.0,15.8,"[0.35159325856374474, 1.82955834461561e-06, 0.9510522100457386, 0.008159101564789311]",,543563,89,False,,,,"[2.96489438690486e-05, 9.058399179916786e-07, 0.9999687614755668, 6.837406461312083e-07]",mixture,"{'mixture': 0.9510541615311272, 'uniform': 0.06719139301873447, 'top1': 0.9510522100457386, 'top2': 0.9510541615311272, 'top3': 0.9510541615311272}",0.9486477837073507,0.04313568054178121,0.0016851823814077627,0.9506762902800429,0.951293643858516,ensemble,1.0, +7,90,0.0,19.0,,,,,,,,,,,,0.949564268705659,0.04505678306437529,0.0016889943583958036,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/log.txt b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..ef2246c3585540ac4ddf9916ff5fe2dc668bdb12 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/log.txt @@ -0,0 +1,103 @@ +[13:13:23][ 0s][START] target=multiclass variant=macs-v11-norestart-acq-entropy budget=5.0% seed=3 | {"task": "mitochondria", "target": "multiclass", "variant": "macs-v11-norestart-acq-entropy", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "entropy", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[13:13:23][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[13:13:23][ 0s][DATA] multiclass: train=1855 val=350 test=175 patches +[13:13:26][ 2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, dros-vnc=2.77 +[13:13:26][ 2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[13:13:27][ 4s][WEIGHTS] mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 | eff.sources=1.19 | MMD2 mix=0.0059 uniform=0.2484 +[13:13:27][ 4s][WEIGHTS] per-source: fly-mito=0.014, kasthuri=0.019, lucchi=0.962, dros-vnc=0.005 | min-MMD source (NeuroADDA rule) = lucchi +[13:13:38][ 14s][TEACHERS] target-val Dice: fly-mito=0.258, kasthuri=0.175, lucchi=0.968, dros-vnc=0.176 +[13:13:38][ 14s][TEACHERS] ensemble target-val Dice: mixture=0.967, mixture@raw=0.968, mixture@logit=0.967, per_source=0.967, per_source@raw=0.967, per_source@logit=0.966, uniform=0.248, uniform@raw=0.208, uniform@logit=0.195 +[13:13:38][ 14s][WEIGHTS] using mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 +[13:13:38][ 14s][INIT] student initialised from teacher lucchi (weight 0.968) +[13:13:41][ 17s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[13:13:41][ 17s][KD] distilling the mixture ensemble (prob space) into the student +[13:14:04][ 40s][KD] soft labels for 1855 patches from 4 teachers in 23s +[13:14:36][ 73s][KD] epoch 1/5 | patches=1855 | loss=0.1033 | 33s +[13:15:10][ 107s][KD] epoch 2/5 | patches=1855 | loss=0.0740 | 34s +[13:15:43][ 140s][KD] epoch 3/5 | patches=1855 | loss=0.0711 | 34s +[13:16:17][ 174s][KD] epoch 4/5 | patches=1855 | loss=0.0696 | 33s +[13:16:50][ 206s][KD] epoch 5/5 | patches=1855 | loss=0.0683 | 33s +[13:16:51][ 208s][ROUND0] before any target label: val_dice=0.9645 val_vi=0.0433 +[13:16:51][ 208s][AL] patches=1855 (pool 1855 after dropping blank; free blank negatives: True) | budget=92 (5.0%) | rounds=6 | per_round=15 | acquisition=entropy | first_round=stratified +[13:16:58][ 214s][AL] round-1 uncertainty quality on 1855 patches: {'spearman_uncertainty_error': -0.1498049315012361, 'auroc_top20_error': 0.2730745199468182} +[13:17:20][ 237s][AL] round 1: stratified cold start: 8 patches from the membrane-rich half (ensemble prevalence >= 0.071) + 7 from the membrane-poor half of 1855 candidates +[13:17:22][ 238s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.968, uniform=0.143, top1=0.968, top2=0.968, top3=0.968 -> using mixture +[13:17:22][ 238s][WEIGHTS] round 1: student kept (labeled Dice 0.970 vs best teacher lucchi 0.968); distilling the corrected ensemble into it +[13:17:22][ 238s][WEIGHTS] round 1: teacher Dice on 15 labeled patches + 0 blank (109027 fg px): fly-mito=0.395, kasthuri=0.000, lucchi=0.968, dros-vnc=0.016 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:17:44][ 261s][KD] soft labels for 1840 patches from 4 teachers in 23s +[13:18:16][ 293s][KD] epoch 1/2 | patches=1840 | loss=0.0450 | 32s +[13:18:46][ 323s][KD] epoch 2/2 | patches=1840 | loss=0.0443 | 30s +[13:19:02][ 338s][CONS] pseudo-labels for 1840 unlabeled patches from the current ensemble in 15s +[13:19:02][ 339s][FT] round 1/6 epoch 1/5 | patches=15 | loss=0.2086 +[13:19:03][ 339s][FT] round 1/6 epoch 2/5 | patches=15 | loss=0.1716 +[13:19:03][ 340s][FT] round 1/6 epoch 3/5 | patches=15 | loss=0.1611 +[13:19:04][ 340s][FT] round 1/6 epoch 4/5 | patches=15 | loss=0.1495 +[13:19:05][ 341s][FT] round 1/6 epoch 5/5 | patches=15 | loss=0.1662 +[13:19:06][ 342s][AL] round 1/6 | labeled=15 | selection_seconds=29.4000 round_seconds=133.4000 restart_student_dice=0.9698 restart_teacher_dice=0.9681 val_dice=0.9645 val_vi=0.0446 val_ece=0.0017 +[13:19:12][ 349s][AL] round 2: 283/1840 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:19:14][ 351s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.961, uniform=0.085, top1=0.961, top2=0.961, top3=0.961 -> using mixture +[13:19:14][ 351s][WEIGHTS] round 2: teacher Dice on 30 labeled patches + 0 blank (196592 fg px): fly-mito=0.362, kasthuri=0.000, lucchi=0.961, dros-vnc=0.009 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:19:15][ 352s][CONS] round 2: on the 15 new patches student Dice=0.946 vs ensemble 0.952 -> pseudo-labels from the ensemble | consistency x1.00 +[13:19:16][ 353s][FT] round 2/6 epoch 1/5 | patches=30 | loss=0.1723 +[13:19:18][ 354s][FT] round 2/6 epoch 2/5 | patches=30 | loss=0.1952 +[13:19:19][ 356s][FT] round 2/6 epoch 3/5 | patches=30 | loss=0.1984 +[13:19:20][ 357s][FT] round 2/6 epoch 4/5 | patches=30 | loss=0.1825 +[13:19:22][ 358s][FT] round 2/6 epoch 5/5 | patches=30 | loss=0.2224 +[13:19:23][ 359s][AL] round 2/6 | labeled=30 | selection_seconds=6.4000 round_seconds=15.8000 pseudo_student_dice=0.9459 pseudo_ensemble_dice=0.9525 consistency_scale=1.0000 val_dice=0.9326 val_vi=0.0477 val_ece=0.0014 +[13:19:29][ 366s][AL] round 3: 282/1825 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:19:32][ 369s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.955, uniform=0.072, top1=0.955, top2=0.955, top3=0.955 -> using mixture +[13:19:32][ 369s][WEIGHTS] round 3: teacher Dice on 45 labeled patches + 0 blank (284641 fg px): fly-mito=0.351, kasthuri=0.000, lucchi=0.955, dros-vnc=0.011 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:19:33][ 369s][CONS] round 3: on the 15 new patches student Dice=0.917 vs ensemble 0.941 -> pseudo-labels from the ensemble | consistency x1.00 +[13:19:35][ 371s][FT] round 3/6 epoch 1/5 | patches=45 | loss=0.1934 +[13:19:37][ 373s][FT] round 3/6 epoch 2/5 | patches=45 | loss=0.2079 +[13:19:39][ 375s][FT] round 3/6 epoch 3/5 | patches=45 | loss=0.1462 +[13:19:40][ 377s][FT] round 3/6 epoch 4/5 | patches=45 | loss=0.1664 +[13:19:42][ 379s][FT] round 3/6 epoch 5/5 | patches=45 | loss=0.1769 +[13:19:44][ 380s][AL] round 3/6 | labeled=45 | selection_seconds=6.3000 round_seconds=19.5000 pseudo_student_dice=0.9171 pseudo_ensemble_dice=0.9414 consistency_scale=1.0000 val_dice=0.9615 val_vi=0.0451 val_ece=0.0017 +[13:19:50][ 386s][AL] round 4: 273/1810 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:19:54][ 390s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.953, uniform=0.069, top1=0.953, top2=0.953, top3=0.953 -> using mixture +[13:19:54][ 390s][WEIGHTS] round 4: teacher Dice on 60 labeled patches + 0 blank (364598 fg px): fly-mito=0.346, kasthuri=0.000, lucchi=0.953, dros-vnc=0.012 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:19:54][ 391s][CONS] round 4: on the 15 new patches student Dice=0.901 vs ensemble 0.946 -> pseudo-labels from the ensemble | consistency x1.00 +[13:19:57][ 393s][FT] round 4/6 epoch 1/5 | patches=60 | loss=0.1831 +[13:19:59][ 396s][FT] round 4/6 epoch 2/5 | patches=60 | loss=0.1758 +[13:20:02][ 398s][FT] round 4/6 epoch 3/5 | patches=60 | loss=0.1748 +[13:20:04][ 401s][FT] round 4/6 epoch 4/5 | patches=60 | loss=0.1600 +[13:20:07][ 403s][FT] round 4/6 epoch 5/5 | patches=60 | loss=0.1690 +[13:20:08][ 404s][AL] round 4/6 | labeled=60 | selection_seconds=6.2000 round_seconds=23.0000 pseudo_student_dice=0.9011 pseudo_ensemble_dice=0.9464 consistency_scale=1.0000 val_dice=0.9057 val_vi=0.0503 val_ece=0.0020 +[13:20:14][ 410s][AL] round 5: 282/1795 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:20:18][ 415s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.954, uniform=0.069, top1=0.954, top2=0.954, top3=0.954 -> using mixture +[13:20:18][ 415s][WEIGHTS] round 5: teacher Dice on 75 labeled patches + 0 blank (438598 fg px): fly-mito=0.335, kasthuri=0.000, lucchi=0.954, dros-vnc=0.011 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:20:19][ 416s][CONS] round 5: on the 15 new patches student Dice=0.873 vs ensemble 0.956 -> pseudo-labels from the ensemble | consistency x1.00 +[13:20:22][ 419s][FT] round 5/6 epoch 1/5 | patches=75 | loss=0.2142 +[13:20:25][ 422s][FT] round 5/6 epoch 2/5 | patches=75 | loss=0.1819 +[13:20:28][ 425s][FT] round 5/6 epoch 3/5 | patches=75 | loss=0.1731 +[13:20:31][ 428s][FT] round 5/6 epoch 4/5 | patches=75 | loss=0.1487 +[13:20:34][ 431s][FT] round 5/6 epoch 5/5 | patches=75 | loss=0.1332 +[13:20:35][ 432s][AL] round 5/6 | labeled=75 | selection_seconds=6.2000 round_seconds=26.3000 pseudo_student_dice=0.8729 pseudo_ensemble_dice=0.9563 consistency_scale=1.0000 val_dice=0.9518 val_vi=0.0445 val_ece=0.0017 +[13:20:41][ 438s][AL] round 6: 285/1780 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:20:47][ 443s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.953, uniform=0.069, top1=0.953, top2=0.953, top3=0.953 -> using mixture +[13:20:47][ 443s][WEIGHTS] round 6: teacher Dice on 90 labeled patches + 0 blank (534078 fg px): fly-mito=0.344, kasthuri=0.000, lucchi=0.953, dros-vnc=0.012 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:20:48][ 444s][CONS] round 6: on the 15 new patches student Dice=0.934 vs ensemble 0.952 -> pseudo-labels from the ensemble | consistency x1.00 +[13:20:51][ 448s][FT] round 6/6 epoch 1/5 | patches=90 | loss=0.1742 +[13:20:55][ 451s][FT] round 6/6 epoch 2/5 | patches=90 | loss=0.1507 +[13:20:58][ 455s][FT] round 6/6 epoch 3/5 | patches=90 | loss=0.1472 +[13:21:02][ 458s][FT] round 6/6 epoch 4/5 | patches=90 | loss=0.1742 +[13:21:05][ 462s][FT] round 6/6 epoch 5/5 | patches=90 | loss=0.1719 +[13:21:07][ 463s][AL] round 6/6 | labeled=90 | selection_seconds=6.1000 round_seconds=29.9000 pseudo_student_dice=0.9342 pseudo_ensemble_dice=0.9518 consistency_scale=1.0000 val_dice=0.9564 val_vi=0.0433 val_ece=0.0014 +[13:21:10][ 467s][FT] final 1/10 epoch 1/1 | patches=90 | loss=0.1482 +[13:21:14][ 470s][FT] final 2/10 epoch 1/1 | patches=90 | loss=0.1507 +[13:21:17][ 474s][FT] final 3/10 epoch 1/1 | patches=90 | loss=0.1422 +[13:21:21][ 477s][FT] final 4/10 epoch 1/1 | patches=90 | loss=0.1488 +[13:21:24][ 481s][FT] final 5/10 epoch 1/1 | patches=90 | loss=0.1223 +[13:21:28][ 484s][FT] final 6/10 epoch 1/1 | patches=90 | loss=0.1140 +[13:21:31][ 488s][FT] final 7/10 epoch 1/1 | patches=90 | loss=0.1293 +[13:21:35][ 491s][FT] final 8/10 epoch 1/1 | patches=90 | loss=0.1250 +[13:21:38][ 495s][FT] final 9/10 epoch 1/1 | patches=90 | loss=0.1283 +[13:21:42][ 498s][FT] final 10/10 epoch 1/1 | patches=90 | loss=0.1129 +[13:21:43][ 500s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=35.3000 val_dice=0.9581 val_vi=0.0421 val_ece=0.0017 +[13:21:43][ 500s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:21:48][ 505s][TEST] raw student (no EMA, no TTA): patch_dice=0.9016 patch_vi=0.0798 +[13:21:49][ 505s][TEST] final model without TTA: patch_dice=0.9016 patch_vi=0.0798 +[13:21:49][ 505s][RESOURCES] {"phases": {"setup": {"seconds": 2.15, "peak_cpu_rss_mb": 1025.8, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 15.11, "peak_cpu_rss_mb": 1803.2, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 189.06, "peak_cpu_rss_mb": 2392.6, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4148.0, "kd_epochs": 5}, "active_learning": {"seconds": 292.0, "peak_cpu_rss_mb": 2665.0, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4156.0, "rounds": 6}, "inference": {"seconds": 5.69, "peak_cpu_rss_mb": 2665.0, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4156.0, "patches": 175, "images": 5, "patches_per_second": 30.8, "seconds_per_image": 1.138}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 2665.0, "total_seconds": 504.0} +[13:21:49][ 505s][TEST] patch_dice=0.8900 patch_iou=0.8427 patch_precision=0.9766 patch_recall=0.8623 patch_vi=0.0904 image_dice=0.9077 image_iou=0.8354 image_precision=0.9785 image_recall=0.8533 image_vi=0.1054 cal_ece=0.0072 cal_nll=0.0833 cal_brier=0.0107 cal_mean_confidence=0.9946 cal_accuracy=0.9874 cal_predicted_prevalence=0.0804 cal_true_prevalence=0.0886 cal_mean_probability=0.0801 +[13:21:49][ 506s][DONE] multiclass macs-v11-norestart-acq-entropy b5 s3 | test patch Dice=0.8900 VI=0.0904 | image Dice=0.9077 VI=0.1054 | 505s diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/result.json b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..1a5813d1705e873a508d4eba4ff49c665b129848 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/result.json @@ -0,0 +1,647 @@ +{ + "target": "multiclass", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 3, + "config": { + "task": "mitochondria", + "target": "multiclass", + "variant": "macs-v11-norestart-acq-entropy", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "entropy", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "multiclass", + "--variants", + "macs-v11-norestart-acq-entropy", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "alphas": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "mixture": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "per_source": [ + 0.014321466605441749, + 0.018762803993963635, + 0.9619825923792787, + 0.004933137021315701 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.004128299797828663, + "mmd2_mixture": 0.005874306045373645, + "mmd2_uniform": 0.24840696635360943, + "mmd2_single_source": [ + 0.6122003026141286, + 0.5667974971994383, + 0.004845544051820361, + 0.668933003056658 + ], + "median_squared_distance": 0.20473960170977523, + "gammas": [ + 1.221063233064128, + 2.442126466128256, + 4.884252932256512, + 9.768505864513024, + 19.537011729026048 + ], + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6529830716388123, + 0.6112535306587584, + 0.0030196383578795505, + 0.8176329290240223 + ], + "standardized": [ + 0.4264461417532283, + 0.2913877286688084, + -1.6771722494431436, + 0.9593383790211064 + ], + "tau": 0.5, + "max_weight": 0.9619825923792787, + "entropy": 0.19889667121630067, + "effective_sources": 1.2200558924950622, + "min_weight": 0.004933137021315701 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "lucchi-pp", + "seconds": 1.3 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.25842773603204217, + 0.17499716869513837, + 0.9675530639716557, + 0.17625312705123047 + ], + "ensemble_val_dice": { + "mixture": 0.9673733701024737, + "mixture@raw": 0.9675071736744472, + "mixture@logit": 0.9667425331047603, + "per_source": 0.9673222165448325, + "per_source@raw": 0.967496030160359, + "per_source@logit": 0.9664733615943364, + "uniform": 0.24779489094285445, + "uniform@raw": 0.20821287741695416, + "uniform@logit": 0.19508316544090382 + }, + "val_patches": 350 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 3.7969236373901367, + 2.767761468887329 + ], + "top_source": "lucchi-pp", + "weight_stats": { + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.39999999999999997, + "spearman_weight_taxonomy": -0.8944271909999159, + "top_source": "lucchi-pp", + "top_source_is_best_teacher": true + }, + "init_teacher": "lucchi-pp", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 189.1, + "round0_val": { + "val_dice": 0.964532719558221, + "val_vi": 0.043284516926611456, + "val_ece": 0.009747508076338396 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 15, + "selection_seconds": 29.4, + "round_seconds": 133.4, + "teacher_dice_labeled": [ + 0.3945454313577119, + 9.171955827860732e-06, + 0.9680883362682224, + 0.01620298768172638 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 109027, + "evidence_nonempty_patches": 14, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.9697621255984336, + "restart_teacher_dice": 0.9680883362682224, + "weights": [ + 3.8420009317991516e-05, + 7.639987786195338e-07, + 0.9999601910568676, + 6.249350358019625e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9680835162782008, + "uniform": 0.14307490547399257, + "top1": 0.9680883362682224, + "top2": 0.9680835162782008, + "top3": 0.9680835162782008 + }, + "val_dice": 0.9645227323877767, + "val_vi": 0.044619010034273925, + "val_ece": 0.0017458928573715086 + }, + { + "round": 2, + "n_labeled": 30, + "selection_seconds": 6.4, + "round_seconds": 15.8, + "teacher_dice_labeled": [ + 0.36209477620968805, + 5.086651101514296e-06, + 0.9610735805211097, + 0.009017173640082627 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 196592, + "evidence_nonempty_patches": 29, + "first_correction": false, + "restart": "", + "weights": [ + 2.979164485190767e-05, + 8.194894047372959e-07, + 0.9999687649934478, + 6.238722955845938e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9610709098620875, + "uniform": 0.0849203643271151, + "top1": 0.9610735805211097, + "top2": 0.9610735805211097, + "top3": 0.9610709098620875 + }, + "pseudo_student_dice": 0.9459372235236511, + "pseudo_ensemble_dice": 0.952488480843328, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9326183743773403, + "val_vi": 0.0477410063919867, + "val_ece": 0.001400668233203015 + }, + { + "round": 3, + "n_labeled": 45, + "selection_seconds": 6.3, + "round_seconds": 19.5, + "teacher_dice_labeled": [ + 0.35087425279619033, + 3.5131849832421078e-06, + 0.9549401420166936, + 0.01097529947685219 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 284641, + "evidence_nonempty_patches": 44, + "first_correction": false, + "restart": "", + "weights": [ + 2.8314060089773256e-05, + 8.713131946600871e-07, + 0.9999701381738917, + 6.764528238449629e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9549419916637906, + "uniform": 0.07214672486716943, + "top1": 0.9549401420166936, + "top2": 0.9549419916637906, + "top3": 0.9549419916637906 + }, + "pseudo_student_dice": 0.9170525164846199, + "pseudo_ensemble_dice": 0.9413887152679129, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9615215461421914, + "val_vi": 0.04507797894981182, + "val_ece": 0.001683667020509537 + }, + { + "round": 4, + "n_labeled": 60, + "selection_seconds": 6.2, + "round_seconds": 23.0, + "teacher_dice_labeled": [ + 0.34641884884108504, + 2.736614534159791e-06, + 0.9530518244220401, + 0.011959939506696696 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 364598, + "evidence_nonempty_patches": 58, + "first_correction": false, + "restart": "", + "weights": [ + 2.7596482581631467e-05, + 8.879163910088206e-07, + 0.9999708194316652, + 6.961693620906562e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9530518244220401, + "uniform": 0.06925237083169014, + "top1": 0.9530518244220401, + "top2": 0.9530518244220401, + "top3": 0.9530518244220401 + }, + "pseudo_student_dice": 0.9011273213420437, + "pseudo_ensemble_dice": 0.9463985032740879, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9056557971297992, + "val_vi": 0.05026376867483164, + "val_ece": 0.0019793680350681585 + }, + { + "round": 5, + "n_labeled": 75, + "selection_seconds": 6.2, + "round_seconds": 26.3, + "teacher_dice_labeled": [ + 0.3346199135065683, + 2.275680485357134e-06, + 0.9536079882434412, + 0.01110922601879814 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 438598, + "evidence_nonempty_patches": 73, + "first_correction": false, + "restart": "", + "weights": [ + 2.4389203438956514e-05, + 8.829906032953858e-07, + 0.9999740413600581, + 6.864458996066167e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9536091840011336, + "uniform": 0.06910263181058006, + "top1": 0.9536079882434412, + "top2": 0.9536091840011336, + "top3": 0.9536091840011336 + }, + "pseudo_student_dice": 0.8729114748954606, + "pseudo_ensemble_dice": 0.9562961231509, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9518048202755569, + "val_vi": 0.04450750110763561, + "val_ece": 0.0016885563000789516 + }, + { + "round": 6, + "n_labeled": 90, + "selection_seconds": 6.1, + "round_seconds": 29.9, + "teacher_dice_labeled": [ + 0.3442403938603118, + 1.8694803592393457e-06, + 0.9532876405214105, + 0.011610623283675266 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 534078, + "evidence_nonempty_patches": 88, + "first_correction": false, + "restart": "", + "weights": [ + 2.6938224835333994e-05, + 8.858179153993843e-07, + 0.9999714838490448, + 6.921082045030201e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9532877285264284, + "uniform": 0.06884778333670726, + "top1": 0.9532876405214105, + "top2": 0.9532877285264284, + "top3": 0.9532877285264284 + }, + "pseudo_student_dice": 0.9341540346371057, + "pseudo_ensemble_dice": 0.9517847472077299, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9563853420786778, + "val_vi": 0.04329670203377316, + "val_ece": 0.001406319267905886 + }, + { + "round": 7, + "n_labeled": 90, + "selection_seconds": 0.0, + "round_seconds": 35.3, + "final_consolidation": true, + "val_dice": 0.9581084082280827, + "val_vi": 0.04209837898533957, + "val_ece": 0.001678692111852077 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.1498049315012361, + "auroc_top20_error": 0.2730745199468182 + }, + "al_seconds": 292.0, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 2.6938224835333994e-05, + 8.858179153993843e-07, + 0.9999714838490448, + 6.921082045030201e-07 + ], + "weight_stats_final": { + "max_weight": 0.9999714838490448, + "entropy": 0.00033412074366636827, + "effective_sources": 1.0003341765682192, + "min_weight": 6.921082045030201e-07 + }, + "top_source_final": "lucchi-pp", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.9015696548776413, + "patch_iou": 0.8618806570071849, + "patch_precision": 0.9716983204129273, + "patch_recall": 0.8872217779065581, + "patch_vi": 0.07984149427657221, + "n_patches": 175, + "cal_ece": 0.007994530846354354, + "cal_nll": 0.08855378147702361, + "cal_brier": 0.009792356684570796, + "cal_mean_confidence": 0.9968617183463544, + "cal_accuracy": 0.9888671875, + "cal_predicted_prevalence": 0.08243303571428572, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08208372195704684 + }, + "test_notta": { + "patch_dice": 0.9015696548776413, + "patch_iou": 0.8618806570071849, + "patch_precision": 0.9716983204129273, + "patch_recall": 0.8872217779065581, + "patch_vi": 0.07984149427657221, + "n_patches": 175, + "cal_ece": 0.007994530846354354, + "cal_nll": 0.08855378147702361, + "cal_brier": 0.009792356684570796, + "cal_mean_confidence": 0.9968617183463544, + "cal_accuracy": 0.9888671875, + "cal_predicted_prevalence": 0.08243303571428572, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08208372195704684 + }, + "test": { + "patch_dice": 0.8900099898414046, + "patch_iou": 0.842657945984699, + "patch_precision": 0.9765654197182481, + "patch_recall": 0.862274458239762, + "patch_vi": 0.09043733614665139, + "n_patches": 175, + "image_dice": 0.9077466820486997, + "image_iou": 0.8353535378248867, + "image_precision": 0.9785348700607933, + "image_recall": 0.8533192309716462, + "image_vi": 0.10537477543400615, + "n_images": 5, + "cal_ece": 0.0072378983565131744, + "cal_nll": 0.08327452624592593, + "cal_brier": 0.010736460279384524, + "cal_mean_confidence": 0.9945569814356102, + "cal_accuracy": 0.9874441964285714, + "cal_predicted_prevalence": 0.08036272321428571, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08006140937958924 + }, + "eval_seconds": 5.7, + "total_seconds": 505.4, + "resources": { + "phases": { + "setup": { + "seconds": 2.15, + "peak_cpu_rss_mb": 1025.8, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 15.11, + "peak_cpu_rss_mb": 1803.2, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 189.06, + "peak_cpu_rss_mb": 2392.6, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4148.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 292.0, + "peak_cpu_rss_mb": 2665.0, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4156.0, + "rounds": 6 + }, + "inference": { + "seconds": 5.69, + "peak_cpu_rss_mb": 2665.0, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4156.0, + "patches": 175, + "images": 5, + "patches_per_second": 30.8, + "seconds_per_image": 1.138 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 2665.0, + "total_seconds": 504.0 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/multiclass/macs-v11-norestart-acq-entropy_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..2145ed352e608fa389454d9b448f404b1ce49f26 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-entropy/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,15,29.4,133.4,"[0.3945454313577119, 9.171955827860732e-06, 0.9680883362682224, 0.01620298768172638]",,109027,14,True,kept,0.9697621255984336,0.9680883362682224,"[3.8420009317991516e-05, 7.639987786195338e-07, 0.9999601910568676, 6.249350358019625e-07]",mixture,"{'mixture': 0.9680835162782008, 'uniform': 0.14307490547399257, 'top1': 0.9680883362682224, 'top2': 0.9680835162782008, 'top3': 0.9680835162782008}",0.9645227323877767,0.044619010034273925,0.0017458928573715086,,,,, +2,30,6.4,15.8,"[0.36209477620968805, 5.086651101514296e-06, 0.9610735805211097, 0.009017173640082627]",,196592,29,False,,,,"[2.979164485190767e-05, 8.194894047372959e-07, 0.9999687649934478, 6.238722955845938e-07]",mixture,"{'mixture': 0.9610709098620875, 'uniform': 0.0849203643271151, 'top1': 0.9610735805211097, 'top2': 0.9610735805211097, 'top3': 0.9610709098620875}",0.9326183743773403,0.0477410063919867,0.001400668233203015,0.9459372235236511,0.952488480843328,ensemble,1.0, +3,45,6.3,19.5,"[0.35087425279619033, 3.5131849832421078e-06, 0.9549401420166936, 0.01097529947685219]",,284641,44,False,,,,"[2.8314060089773256e-05, 8.713131946600871e-07, 0.9999701381738917, 6.764528238449629e-07]",mixture,"{'mixture': 0.9549419916637906, 'uniform': 0.07214672486716943, 'top1': 0.9549401420166936, 'top2': 0.9549419916637906, 'top3': 0.9549419916637906}",0.9615215461421914,0.04507797894981182,0.001683667020509537,0.9170525164846199,0.9413887152679129,ensemble,1.0, +4,60,6.2,23.0,"[0.34641884884108504, 2.736614534159791e-06, 0.9530518244220401, 0.011959939506696696]",,364598,58,False,,,,"[2.7596482581631467e-05, 8.879163910088206e-07, 0.9999708194316652, 6.961693620906562e-07]",mixture,"{'mixture': 0.9530518244220401, 'uniform': 0.06925237083169014, 'top1': 0.9530518244220401, 'top2': 0.9530518244220401, 'top3': 0.9530518244220401}",0.9056557971297992,0.05026376867483164,0.0019793680350681585,0.9011273213420437,0.9463985032740879,ensemble,1.0, +5,75,6.2,26.3,"[0.3346199135065683, 2.275680485357134e-06, 0.9536079882434412, 0.01110922601879814]",,438598,73,False,,,,"[2.4389203438956514e-05, 8.829906032953858e-07, 0.9999740413600581, 6.864458996066167e-07]",mixture,"{'mixture': 0.9536091840011336, 'uniform': 0.06910263181058006, 'top1': 0.9536079882434412, 'top2': 0.9536091840011336, 'top3': 0.9536091840011336}",0.9518048202755569,0.04450750110763561,0.0016885563000789516,0.8729114748954606,0.9562961231509,ensemble,1.0, +6,90,6.1,29.9,"[0.3442403938603118, 1.8694803592393457e-06, 0.9532876405214105, 0.011610623283675266]",,534078,88,False,,,,"[2.6938224835333994e-05, 8.858179153993843e-07, 0.9999714838490448, 6.921082045030201e-07]",mixture,"{'mixture': 0.9532877285264284, 'uniform': 0.06884778333670726, 'top1': 0.9532876405214105, 'top2': 0.9532877285264284, 'top3': 0.9532877285264284}",0.9563853420786778,0.04329670203377316,0.001406319267905886,0.9341540346371057,0.9517847472077299,ensemble,1.0, +7,90,0.0,35.3,,,,,,,,,,,,0.9581084082280827,0.04209837898533957,0.001678692111852077,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/log.txt b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..f75d8a3d6dc677bbbffb15d05b0fa87917074c1d --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/log.txt @@ -0,0 +1,103 @@ +[08:25:59][ 0s][START] target=multiclass variant=macs-v11-norestart-acq-margin budget=5.0% seed=2 | {"task": "mitochondria", "target": "multiclass", "variant": "macs-v11-norestart-acq-margin", "budget": 5.0, "seed": 2, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "margin", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[08:25:59][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[08:25:59][ 0s][DATA] multiclass: train=1855 val=350 test=175 patches +[08:26:00][ 1s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, dros-vnc=2.77 +[08:26:00][ 1s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[08:26:01][ 2s][WEIGHTS] mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 | eff.sources=1.19 | MMD2 mix=0.0059 uniform=0.2484 +[08:26:01][ 2s][WEIGHTS] per-source: fly-mito=0.014, kasthuri=0.019, lucchi=0.962, dros-vnc=0.005 | min-MMD source (NeuroADDA rule) = lucchi +[08:26:07][ 8s][TEACHERS] target-val Dice: fly-mito=0.258, kasthuri=0.175, lucchi=0.968, dros-vnc=0.176 +[08:26:07][ 8s][TEACHERS] ensemble target-val Dice: mixture=0.967, mixture@raw=0.968, mixture@logit=0.967, per_source=0.967, per_source@raw=0.967, per_source@logit=0.966, uniform=0.248, uniform@raw=0.208, uniform@logit=0.195 +[08:26:07][ 8s][WEIGHTS] using mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 +[08:26:07][ 8s][INIT] student initialised from teacher lucchi (weight 0.968) +[08:26:08][ 9s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[08:26:08][ 9s][KD] distilling the mixture ensemble (prob space) into the student +[08:26:19][ 20s][KD] soft labels for 1855 patches from 4 teachers in 11s +[08:26:36][ 37s][KD] epoch 1/5 | patches=1855 | loss=0.1028 | 17s +[08:26:53][ 54s][KD] epoch 2/5 | patches=1855 | loss=0.0735 | 17s +[08:27:09][ 70s][KD] epoch 3/5 | patches=1855 | loss=0.0710 | 16s +[08:27:26][ 87s][KD] epoch 4/5 | patches=1855 | loss=0.0695 | 17s +[08:27:43][ 104s][KD] epoch 5/5 | patches=1855 | loss=0.0687 | 17s +[08:27:44][ 105s][ROUND0] before any target label: val_dice=0.9665 val_vi=0.0425 +[08:27:44][ 105s][AL] patches=1855 (pool 1855 after dropping blank; free blank negatives: True) | budget=92 (5.0%) | rounds=6 | per_round=15 | acquisition=margin | first_round=stratified +[08:27:47][ 108s][AL] round-1 uncertainty quality on 1855 patches: {'spearman_uncertainty_error': -0.10042322840297344, 'auroc_top20_error': 0.2929541343059118} +[08:27:58][ 119s][AL] round 1: stratified cold start: 8 patches from the membrane-rich half (ensemble prevalence >= 0.071) + 7 from the membrane-poor half of 1855 candidates +[08:27:59][ 120s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.971, uniform=0.135, top1=0.971, top2=0.971, top3=0.971 -> using mixture +[08:27:59][ 120s][WEIGHTS] round 1: student kept (labeled Dice 0.973 vs best teacher lucchi 0.971); distilling the corrected ensemble into it +[08:27:59][ 120s][WEIGHTS] round 1: teacher Dice on 15 labeled patches + 0 blank (92186 fg px): fly-mito=0.353, kasthuri=0.000, lucchi=0.971, dros-vnc=0.006 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:28:10][ 131s][KD] soft labels for 1840 patches from 4 teachers in 11s +[08:28:26][ 147s][KD] epoch 1/2 | patches=1840 | loss=0.0450 | 16s +[08:28:43][ 164s][KD] epoch 2/2 | patches=1840 | loss=0.0441 | 16s +[08:28:53][ 174s][CONS] pseudo-labels for 1840 unlabeled patches from the current ensemble in 11s +[08:28:54][ 175s][FT] round 1/6 epoch 1/5 | patches=15 | loss=0.1524 +[08:28:54][ 175s][FT] round 1/6 epoch 2/5 | patches=15 | loss=0.1165 +[08:28:55][ 176s][FT] round 1/6 epoch 3/5 | patches=15 | loss=0.1215 +[08:28:55][ 176s][FT] round 1/6 epoch 4/5 | patches=15 | loss=0.1068 +[08:28:56][ 177s][FT] round 1/6 epoch 5/5 | patches=15 | loss=0.0948 +[08:28:57][ 178s][AL] round 1/6 | labeled=15 | selection_seconds=13.9000 round_seconds=72.0000 restart_student_dice=0.9734 restart_teacher_dice=0.9708 val_dice=0.9560 val_vi=0.0479 val_ece=0.0025 +[08:29:00][ 181s][AL] round 2: 285/1840 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:29:01][ 182s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.963, uniform=0.108, top1=0.963, top2=0.963, top3=0.963 -> using mixture +[08:29:01][ 182s][WEIGHTS] round 2: teacher Dice on 30 labeled patches + 0 blank (183893 fg px): fly-mito=0.343, kasthuri=0.000, lucchi=0.963, dros-vnc=0.011 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:29:02][ 183s][CONS] round 2: on the 15 new patches student Dice=0.930 vs ensemble 0.955 -> pseudo-labels from the ensemble | consistency x1.00 +[08:29:03][ 184s][FT] round 2/6 epoch 1/5 | patches=30 | loss=0.1882 +[08:29:04][ 185s][FT] round 2/6 epoch 2/5 | patches=30 | loss=0.1541 +[08:29:04][ 185s][FT] round 2/6 epoch 3/5 | patches=30 | loss=0.1579 +[08:29:05][ 186s][FT] round 2/6 epoch 4/5 | patches=30 | loss=0.1478 +[08:29:06][ 187s][FT] round 2/6 epoch 5/5 | patches=30 | loss=0.1783 +[08:29:07][ 188s][AL] round 2/6 | labeled=30 | selection_seconds=3.1000 round_seconds=9.4000 pseudo_student_dice=0.9301 pseudo_ensemble_dice=0.9545 consistency_scale=1.0000 val_dice=0.9540 val_vi=0.0425 val_ece=0.0012 +[08:29:10][ 191s][AL] round 3: 281/1825 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:29:12][ 193s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.959, uniform=0.080, top1=0.959, top2=0.959, top3=0.959 -> using mixture +[08:29:12][ 193s][WEIGHTS] round 3: teacher Dice on 45 labeled patches + 0 blank (266018 fg px): fly-mito=0.341, kasthuri=0.000, lucchi=0.959, dros-vnc=0.008 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:29:13][ 194s][CONS] round 3: on the 15 new patches student Dice=0.925 vs ensemble 0.952 -> pseudo-labels from the ensemble | consistency x1.00 +[08:29:14][ 195s][FT] round 3/6 epoch 1/5 | patches=45 | loss=0.2068 +[08:29:15][ 196s][FT] round 3/6 epoch 2/5 | patches=45 | loss=0.1946 +[08:29:16][ 197s][FT] round 3/6 epoch 3/5 | patches=45 | loss=0.1724 +[08:29:17][ 198s][FT] round 3/6 epoch 4/5 | patches=45 | loss=0.2027 +[08:29:18][ 199s][FT] round 3/6 epoch 5/5 | patches=45 | loss=0.1897 +[08:29:19][ 200s][AL] round 3/6 | labeled=45 | selection_seconds=3.0000 round_seconds=11.1000 pseudo_student_dice=0.9254 pseudo_ensemble_dice=0.9516 consistency_scale=1.0000 val_dice=0.9587 val_vi=0.0460 val_ece=0.0016 +[08:29:22][ 203s][AL] round 4: 284/1810 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:29:24][ 205s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.958, uniform=0.068, top1=0.958, top2=0.958, top3=0.958 -> using mixture +[08:29:24][ 205s][WEIGHTS] round 4: teacher Dice on 60 labeled patches + 0 blank (357787 fg px): fly-mito=0.343, kasthuri=0.000, lucchi=0.958, dros-vnc=0.007 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:29:25][ 206s][CONS] round 4: on the 15 new patches student Dice=0.924 vs ensemble 0.955 -> pseudo-labels from the ensemble | consistency x1.00 +[08:29:26][ 207s][FT] round 4/6 epoch 1/5 | patches=60 | loss=0.1694 +[08:29:27][ 208s][FT] round 4/6 epoch 2/5 | patches=60 | loss=0.1583 +[08:29:29][ 210s][FT] round 4/6 epoch 3/5 | patches=60 | loss=0.1294 +[08:29:30][ 211s][FT] round 4/6 epoch 4/5 | patches=60 | loss=0.1537 +[08:29:31][ 212s][FT] round 4/6 epoch 5/5 | patches=60 | loss=0.1351 +[08:29:32][ 213s][AL] round 4/6 | labeled=60 | selection_seconds=2.9000 round_seconds=12.6000 pseudo_student_dice=0.9242 pseudo_ensemble_dice=0.9548 consistency_scale=1.0000 val_dice=0.9602 val_vi=0.0437 val_ece=0.0011 +[08:29:35][ 216s][AL] round 5: 285/1795 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:29:37][ 218s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.957, uniform=0.064, top1=0.957, top2=0.957, top3=0.957 -> using mixture +[08:29:37][ 218s][WEIGHTS] round 5: teacher Dice on 75 labeled patches + 0 blank (446155 fg px): fly-mito=0.345, kasthuri=0.000, lucchi=0.957, dros-vnc=0.006 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:29:38][ 219s][CONS] round 5: on the 15 new patches student Dice=0.940 vs ensemble 0.952 -> pseudo-labels from the ensemble | consistency x1.00 +[08:29:40][ 221s][FT] round 5/6 epoch 1/5 | patches=75 | loss=0.1520 +[08:29:41][ 222s][FT] round 5/6 epoch 2/5 | patches=75 | loss=0.1581 +[08:29:43][ 224s][FT] round 5/6 epoch 3/5 | patches=75 | loss=0.1336 +[08:29:45][ 226s][FT] round 5/6 epoch 4/5 | patches=75 | loss=0.1378 +[08:29:46][ 227s][FT] round 5/6 epoch 5/5 | patches=75 | loss=0.1384 +[08:29:47][ 228s][AL] round 5/6 | labeled=75 | selection_seconds=3.0000 round_seconds=14.0000 pseudo_student_dice=0.9405 pseudo_ensemble_dice=0.9519 consistency_scale=1.0000 val_dice=0.9565 val_vi=0.0449 val_ece=0.0016 +[08:29:50][ 231s][AL] round 6: 285/1780 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[08:29:52][ 233s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.955, uniform=0.071, top1=0.955, top2=0.955, top3=0.955 -> using mixture +[08:29:52][ 233s][WEIGHTS] round 6: teacher Dice on 90 labeled patches + 0 blank (540593 fg px): fly-mito=0.354, kasthuri=0.000, lucchi=0.955, dros-vnc=0.007 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:29:53][ 234s][CONS] round 6: on the 15 new patches student Dice=0.942 vs ensemble 0.944 -> pseudo-labels from the ensemble | consistency x1.00 +[08:29:55][ 236s][FT] round 6/6 epoch 1/5 | patches=90 | loss=0.1505 +[08:29:57][ 238s][FT] round 6/6 epoch 2/5 | patches=90 | loss=0.1433 +[08:29:59][ 240s][FT] round 6/6 epoch 3/5 | patches=90 | loss=0.1210 +[08:30:01][ 242s][FT] round 6/6 epoch 4/5 | patches=90 | loss=0.1265 +[08:30:03][ 244s][FT] round 6/6 epoch 5/5 | patches=90 | loss=0.1214 +[08:30:04][ 245s][AL] round 6/6 | labeled=90 | selection_seconds=2.9000 round_seconds=16.0000 pseudo_student_dice=0.9418 pseudo_ensemble_dice=0.9438 consistency_scale=1.0000 val_dice=0.9633 val_vi=0.0423 val_ece=0.0013 +[08:30:06][ 247s][FT] final 1/10 epoch 1/1 | patches=90 | loss=0.1235 +[08:30:08][ 249s][FT] final 2/10 epoch 1/1 | patches=90 | loss=0.1184 +[08:30:10][ 250s][FT] final 3/10 epoch 1/1 | patches=90 | loss=0.1115 +[08:30:11][ 252s][FT] final 4/10 epoch 1/1 | patches=90 | loss=0.1169 +[08:30:13][ 254s][FT] final 5/10 epoch 1/1 | patches=90 | loss=0.1127 +[08:30:15][ 256s][FT] final 6/10 epoch 1/1 | patches=90 | loss=0.1138 +[08:30:17][ 258s][FT] final 7/10 epoch 1/1 | patches=90 | loss=0.1125 +[08:30:19][ 260s][FT] final 8/10 epoch 1/1 | patches=90 | loss=0.0987 +[08:30:21][ 262s][FT] final 9/10 epoch 1/1 | patches=90 | loss=0.1057 +[08:30:23][ 264s][FT] final 10/10 epoch 1/1 | patches=90 | loss=0.1001 +[08:30:24][ 265s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=18.9000 val_dice=0.9643 val_vi=0.0414 val_ece=0.0015 +[08:30:24][ 265s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[08:30:26][ 267s][TEST] raw student (no EMA, no TTA): patch_dice=0.9046 patch_vi=0.0778 +[08:30:27][ 268s][TEST] final model without TTA: patch_dice=0.9046 patch_vi=0.0778 +[08:30:27][ 268s][RESOURCES] {"phases": {"setup": {"seconds": 0.59, "peak_cpu_rss_mb": 1026.7, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 8.07, "peak_cpu_rss_mb": 1804.3, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 95.3, "peak_cpu_rss_mb": 2393.6, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4148.0, "kd_epochs": 5}, "active_learning": {"seconds": 159.59, "peak_cpu_rss_mb": 2671.2, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4156.0, "rounds": 6}, "inference": {"seconds": 3.1, "peak_cpu_rss_mb": 2671.2, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4156.0, "patches": 175, "images": 5, "patches_per_second": 56.5, "seconds_per_image": 0.62}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 2671.2, "total_seconds": 266.6} +[08:30:27][ 268s][TEST] patch_dice=0.8898 patch_iou=0.8425 patch_precision=0.9664 patch_recall=0.8716 patch_vi=0.0918 image_dice=0.9052 image_iou=0.8301 image_precision=0.9628 image_recall=0.8627 image_vi=0.1136 cal_ece=0.0083 cal_nll=0.0792 cal_brier=0.0109 cal_mean_confidence=0.9953 cal_accuracy=0.9870 cal_predicted_prevalence=0.0820 cal_true_prevalence=0.0886 cal_mean_probability=0.0814 +[08:30:27][ 268s][DONE] multiclass macs-v11-norestart-acq-margin b5 s2 | test patch Dice=0.8898 VI=0.0918 | image Dice=0.9052 VI=0.1136 | 268s diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/result.json b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/result.json new file mode 100644 index 0000000000000000000000000000000000000000..8b2b0b44a6a5517850b123a3c778b9f2a0cb006a --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/result.json @@ -0,0 +1,647 @@ +{ + "target": "multiclass", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 2, + "config": { + "task": "mitochondria", + "target": "multiclass", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 2, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "margin", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "multiclass", + "--variants", + "macs-v11-norestart-acq-margin", + "--budgets", + "5", + "--seeds", + "2", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "alphas": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "mixture": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "per_source": [ + 0.014321466605441749, + 0.018762803993963635, + 0.9619825923792787, + 0.004933137021315701 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.004128299797828663, + "mmd2_mixture": 0.005874306045373645, + "mmd2_uniform": 0.24840696635360943, + "mmd2_single_source": [ + 0.6122003026141286, + 0.5667974971994383, + 0.004845544051820361, + 0.668933003056658 + ], + "median_squared_distance": 0.20473960170977523, + "gammas": [ + 1.221063233064128, + 2.442126466128256, + 4.884252932256512, + 9.768505864513024, + 19.537011729026048 + ], + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6529830716388123, + 0.6112535306587584, + 0.0030196383578795505, + 0.8176329290240223 + ], + "standardized": [ + 0.4264461417532283, + 0.2913877286688084, + -1.6771722494431436, + 0.9593383790211064 + ], + "tau": 0.5, + "max_weight": 0.9619825923792787, + "entropy": 0.19889667121630067, + "effective_sources": 1.2200558924950622, + "min_weight": 0.004933137021315701 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "lucchi-pp", + "seconds": 1.0 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.25842773603204217, + 0.17499716869513837, + 0.9675530639716557, + 0.17625312705123047 + ], + "ensemble_val_dice": { + "mixture": 0.9673733701024737, + "mixture@raw": 0.9675071736744472, + "mixture@logit": 0.9667425331047603, + "per_source": 0.9673222165448325, + "per_source@raw": 0.967496030160359, + "per_source@logit": 0.9664733615943364, + "uniform": 0.24779489094285445, + "uniform@raw": 0.20821287741695416, + "uniform@logit": 0.19508316544090382 + }, + "val_patches": 350 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 3.7969236373901367, + 2.767761468887329 + ], + "top_source": "lucchi-pp", + "weight_stats": { + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.39999999999999997, + "spearman_weight_taxonomy": -0.8944271909999159, + "top_source": "lucchi-pp", + "top_source_is_best_teacher": true + }, + "init_teacher": "lucchi-pp", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 95.3, + "round0_val": { + "val_dice": 0.9664821314694002, + "val_vi": 0.04254604989486815, + "val_ece": 0.008540277282563461 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 15, + "selection_seconds": 13.9, + "round_seconds": 72.0, + "teacher_dice_labeled": [ + 0.3531319282015097, + 1.084751646110623e-05, + 0.9708165777595404, + 0.005894504591224218 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 92186, + "evidence_nonempty_patches": 15, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.973380797753302, + "restart_teacher_dice": 0.9708165777595404, + "weights": [ + 2.4709178198489446e-05, + 7.43459534993782e-07, + 0.9999739988035825, + 5.485586840192426e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9708165777595404, + "uniform": 0.13523545594303746, + "top1": 0.9708165777595404, + "top2": 0.9708165777595404, + "top3": 0.9708165777595404 + }, + "val_dice": 0.9560329865678461, + "val_vi": 0.04787947142319161, + "val_ece": 0.0024928581906216857 + }, + { + "round": 2, + "n_labeled": 30, + "selection_seconds": 3.1, + "round_seconds": 9.4, + "teacher_dice_labeled": [ + 0.3427255907384892, + 5.437915320782625e-06, + 0.9627224375633177, + 0.010776759401952038 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 183893, + "evidence_nonempty_patches": 30, + "first_correction": false, + "restart": "", + "weights": [ + 2.4144418009402606e-05, + 8.060953787888235e-07, + 0.9999744249195928, + 6.245670190848431e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9627224375633177, + "uniform": 0.1080237229374881, + "top1": 0.9627224375633177, + "top2": 0.9627224375633177, + "top3": 0.9627224375633177 + }, + "pseudo_student_dice": 0.9301396005008776, + "pseudo_ensemble_dice": 0.9545253055841839, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9540084000745441, + "val_vi": 0.04251018094040054, + "val_ece": 0.001181723519984802 + }, + { + "round": 3, + "n_labeled": 45, + "selection_seconds": 3.0, + "round_seconds": 11.1, + "teacher_dice_labeled": [ + 0.3406656176578363, + 3.759129986955819e-06, + 0.959221720209133, + 0.00755973086010192 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 266018, + "evidence_nonempty_patches": 44, + "first_correction": false, + "restart": "", + "weights": [ + 2.4494785959788307e-05, + 8.347999174389686e-07, + 0.9999740440729998, + 6.263411229128656e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.959221720209133, + "uniform": 0.08037189635391487, + "top1": 0.959221720209133, + "top2": 0.959221720209133, + "top3": 0.959221720209133 + }, + "pseudo_student_dice": 0.9254436706078345, + "pseudo_ensemble_dice": 0.9515697118133075, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9586854984366555, + "val_vi": 0.04603169284170308, + "val_ece": 0.001573164729802531 + }, + { + "round": 4, + "n_labeled": 60, + "selection_seconds": 2.9, + "round_seconds": 12.6, + "teacher_dice_labeled": [ + 0.34307255343832094, + 2.7949512001520453e-06, + 0.9580933502089982, + 0.007275734464928621 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 357787, + "evidence_nonempty_patches": 59, + "first_correction": false, + "restart": "", + "weights": [ + 2.5376218058359684e-05, + 8.442639969547949e-07, + 0.9999731478663377, + 6.316516069238657e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9580918833381201, + "uniform": 0.0677963813489074, + "top1": 0.9580933502089982, + "top2": 0.9580918833381201, + "top3": 0.9580918833381201 + }, + "pseudo_student_dice": 0.9241767006839885, + "pseudo_ensemble_dice": 0.954796512431512, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9601968278753901, + "val_vi": 0.043720087299644196, + "val_ece": 0.0010501619925133393 + }, + { + "round": 5, + "n_labeled": 75, + "selection_seconds": 3.0, + "round_seconds": 14.0, + "teacher_dice_labeled": [ + 0.34538076899292985, + 2.2413684899452208e-06, + 0.9568501293008739, + 0.005970182589759339 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 446155, + "evidence_nonempty_patches": 74, + "first_correction": false, + "restart": "", + "weights": [ + 2.6293608203527377e-05, + 8.548200542867155e-07, + 0.9999722203143128, + 6.312574295060962e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9568523829108754, + "uniform": 0.063512627873769, + "top1": 0.9568501293008739, + "top2": 0.9568523829108754, + "top3": 0.9568523829108754 + }, + "pseudo_student_dice": 0.9404803827858177, + "pseudo_ensemble_dice": 0.951871627570948, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9565177012960709, + "val_vi": 0.04485171179356281, + "val_ece": 0.0015691427606498998 + }, + { + "round": 6, + "n_labeled": 90, + "selection_seconds": 2.9, + "round_seconds": 16.0, + "teacher_dice_labeled": [ + 0.35436745114156276, + 1.84678686786794e-06, + 0.9545989983579333, + 0.0069656049077946244 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 540593, + "evidence_nonempty_patches": 89, + "first_correction": false, + "restart": "", + "weights": [ + 2.942076857121487e-05, + 8.742751823441658e-07, + 0.9999690528704682, + 6.520857783162038e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9546018364663353, + "uniform": 0.07087319240627804, + "top1": 0.9545989983579333, + "top2": 0.9546018364663353, + "top3": 0.9546018364663353 + }, + "pseudo_student_dice": 0.9417882365773185, + "pseudo_ensemble_dice": 0.9438394180250572, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9633493335225106, + "val_vi": 0.042278783816662883, + "val_ece": 0.00133770239531772 + }, + { + "round": 7, + "n_labeled": 90, + "selection_seconds": 0.0, + "round_seconds": 18.9, + "final_consolidation": true, + "val_dice": 0.9642885030556532, + "val_vi": 0.04144729652420767, + "val_ece": 0.0015398298322362664 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.10042322840297344, + "auroc_top20_error": 0.2929541343059118 + }, + "al_seconds": 159.6, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 2.942076857121487e-05, + 8.742751823441658e-07, + 0.9999690528704682, + 6.520857783162038e-07 + ], + "weight_stats_final": { + "max_weight": 0.9999690528704682, + "entropy": 0.00035940109368310774, + "effective_sources": 1.0003594656859942, + "min_weight": 6.520857783162038e-07 + }, + "top_source_final": "lucchi-pp", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.9046381406063272, + "patch_iou": 0.8670207090934244, + "patch_precision": 0.9684005330651791, + "patch_recall": 0.8906199472419738, + "patch_vi": 0.07782599806861844, + "n_patches": 175, + "cal_ece": 0.00792163530158271, + "cal_nll": 0.07965875210638623, + "cal_brier": 0.009601323427730788, + "cal_mean_confidence": 0.9970551359177154, + "cal_accuracy": 0.9891796875, + "cal_predicted_prevalence": 0.08264508928571429, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08232324545444795 + }, + "test_notta": { + "patch_dice": 0.9046381406063272, + "patch_iou": 0.8670207090934244, + "patch_precision": 0.9684005330651791, + "patch_recall": 0.8906199472419738, + "patch_vi": 0.07782599806861844, + "n_patches": 175, + "cal_ece": 0.00792163530158271, + "cal_nll": 0.07965875210638623, + "cal_brier": 0.009601323427730788, + "cal_mean_confidence": 0.9970551359177154, + "cal_accuracy": 0.9891796875, + "cal_predicted_prevalence": 0.08264508928571429, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08232324545444795 + }, + "test": { + "patch_dice": 0.8897726125744861, + "patch_iou": 0.8425408008711338, + "patch_precision": 0.9664409479317618, + "patch_recall": 0.8715995651501255, + "patch_vi": 0.09181329884812649, + "n_patches": 175, + "image_dice": 0.9051561172164501, + "image_iou": 0.8300951328573942, + "image_precision": 0.9628138809918202, + "image_recall": 0.8626918894978696, + "image_vi": 0.1135688119666963, + "n_images": 5, + "cal_ece": 0.008291749076722204, + "cal_nll": 0.0791665905974264, + "cal_brier": 0.0108667219880463, + "cal_mean_confidence": 0.9952756778263262, + "cal_accuracy": 0.98703125, + "cal_predicted_prevalence": 0.08198102678571428, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08144396278481261 + }, + "eval_seconds": 3.1, + "total_seconds": 267.7, + "resources": { + "phases": { + "setup": { + "seconds": 0.59, + "peak_cpu_rss_mb": 1026.7, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 8.07, + "peak_cpu_rss_mb": 1804.3, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 95.3, + "peak_cpu_rss_mb": 2393.6, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4148.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 159.59, + "peak_cpu_rss_mb": 2671.2, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4156.0, + "rounds": 6 + }, + "inference": { + "seconds": 3.1, + "peak_cpu_rss_mb": 2671.2, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4156.0, + "patches": 175, + "images": 5, + "patches_per_second": 56.5, + "seconds_per_image": 0.62 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 2671.2, + "total_seconds": 266.6 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/multiclass/macs-v11-norestart-acq-margin_b5_s2.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/rounds.csv b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..023d6851026341ce3997d3cc3d531207870eb7b4 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-2/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,15,13.9,72.0,"[0.3531319282015097, 1.084751646110623e-05, 0.9708165777595404, 0.005894504591224218]",,92186,15,True,kept,0.973380797753302,0.9708165777595404,"[2.4709178198489446e-05, 7.43459534993782e-07, 0.9999739988035825, 5.485586840192426e-07]",mixture,"{'mixture': 0.9708165777595404, 'uniform': 0.13523545594303746, 'top1': 0.9708165777595404, 'top2': 0.9708165777595404, 'top3': 0.9708165777595404}",0.9560329865678461,0.04787947142319161,0.0024928581906216857,,,,, +2,30,3.1,9.4,"[0.3427255907384892, 5.437915320782625e-06, 0.9627224375633177, 0.010776759401952038]",,183893,30,False,,,,"[2.4144418009402606e-05, 8.060953787888235e-07, 0.9999744249195928, 6.245670190848431e-07]",mixture,"{'mixture': 0.9627224375633177, 'uniform': 0.1080237229374881, 'top1': 0.9627224375633177, 'top2': 0.9627224375633177, 'top3': 0.9627224375633177}",0.9540084000745441,0.04251018094040054,0.001181723519984802,0.9301396005008776,0.9545253055841839,ensemble,1.0, +3,45,3.0,11.1,"[0.3406656176578363, 3.759129986955819e-06, 0.959221720209133, 0.00755973086010192]",,266018,44,False,,,,"[2.4494785959788307e-05, 8.347999174389686e-07, 0.9999740440729998, 6.263411229128656e-07]",mixture,"{'mixture': 0.959221720209133, 'uniform': 0.08037189635391487, 'top1': 0.959221720209133, 'top2': 0.959221720209133, 'top3': 0.959221720209133}",0.9586854984366555,0.04603169284170308,0.001573164729802531,0.9254436706078345,0.9515697118133075,ensemble,1.0, +4,60,2.9,12.6,"[0.34307255343832094, 2.7949512001520453e-06, 0.9580933502089982, 0.007275734464928621]",,357787,59,False,,,,"[2.5376218058359684e-05, 8.442639969547949e-07, 0.9999731478663377, 6.316516069238657e-07]",mixture,"{'mixture': 0.9580918833381201, 'uniform': 0.0677963813489074, 'top1': 0.9580933502089982, 'top2': 0.9580918833381201, 'top3': 0.9580918833381201}",0.9601968278753901,0.043720087299644196,0.0010501619925133393,0.9241767006839885,0.954796512431512,ensemble,1.0, +5,75,3.0,14.0,"[0.34538076899292985, 2.2413684899452208e-06, 0.9568501293008739, 0.005970182589759339]",,446155,74,False,,,,"[2.6293608203527377e-05, 8.548200542867155e-07, 0.9999722203143128, 6.312574295060962e-07]",mixture,"{'mixture': 0.9568523829108754, 'uniform': 0.063512627873769, 'top1': 0.9568501293008739, 'top2': 0.9568523829108754, 'top3': 0.9568523829108754}",0.9565177012960709,0.04485171179356281,0.0015691427606498998,0.9404803827858177,0.951871627570948,ensemble,1.0, +6,90,2.9,16.0,"[0.35436745114156276, 1.84678686786794e-06, 0.9545989983579333, 0.0069656049077946244]",,540593,89,False,,,,"[2.942076857121487e-05, 8.742751823441658e-07, 0.9999690528704682, 6.520857783162038e-07]",mixture,"{'mixture': 0.9546018364663353, 'uniform': 0.07087319240627804, 'top1': 0.9545989983579333, 'top2': 0.9546018364663353, 'top3': 0.9546018364663353}",0.9633493335225106,0.042278783816662883,0.00133770239531772,0.9417882365773185,0.9438394180250572,ensemble,1.0, +7,90,0.0,18.9,,,,,,,,,,,,0.9642885030556532,0.04144729652420767,0.0015398298322362664,,,,,True diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/log.txt b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/log.txt new file mode 100644 index 0000000000000000000000000000000000000000..ea427a4f424b7e3a17012e4c60db2883afe6be45 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/log.txt @@ -0,0 +1,103 @@ +[13:21:51][ 0s][START] target=multiclass variant=macs-v11-norestart-acq-margin budget=5.0% seed=3 | {"task": "mitochondria", "target": "multiclass", "variant": "macs-v11-norestart-acq-margin", "budget": 5.0, "seed": 3, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 5, "batch_size": 8, "eval_batch_size": 16, "amp": true, "weighting": "mixture", "feature_n": 512, "feature_seed": 0, "source_split": "train", "kernel_scales": "0.25,0.5,1,2,4", "lam": 0.01, "tau": 0.5, "calibration": "source-val", "ensemble_space": "prob", "student_init": "top-source", "init_compatible": false, "weight_update": "labeled", "reliability_beta": 10.0, "update_cap": 512, "kd_after_update": 2, "reinit_after_update": true, "min_consistency": 64, "consistency_cap": 2000, "select_ensemble": true, "pseudo_source": "auto", "final_epochs": 10, "evidence": "pooled", "min_evidence_fg": 2000, "restart_rule": "evidence", "ema": 0.0, "conf_weight": false, "skip_blank": true, "free_blank": true, "aug_intensity": false, "final_pool": false, "final_pool_cap": 8000, "tta": true, "blank_std": 0.01, "kd": true, "kd_epochs": 5, "kd_lr": 0.0001, "kd_cap": 6000, "kd_kl_weight": 0.0, "rounds": 6, "label_schedule": "rounds", "eval_test_rounds": "", "acquisition": "margin", "first_round": "stratified", "cold_cap": 2000, "recover": false, "recover_margin": 0.1, "recover_max": 2, "recover_big": 0.3, "recover_epochs": 3, "adabn": true, "adabn_cap": 512, "restart_margin": 9.0, "skip_empty": true, "empty_prevalence": 0.005, "score_pool_cap": 8000, "lap_precision": "pool", "lap_score": "entropy+var", "lap_scale": 1.0, "deconf_prev": false, "diversity": "none", "div_mult": 4, "evidence_explore_only": false, "pseudo_switch_margin": 0.05, "cons_gate": true, "cons_gate_lo": 0.5, "cons_gate_hi": 0.8, "pool_evidence": 0.5, "pool_evidence_n": 2000, "oracle_evidence": false, "typicality": 0.0, "evidence_rounds": 0, "explore_frac": 0.0, "strat_bins": 10, "dump_signals": "", "mc_passes": 5, "mc_dropout": 0.2, "ft_epochs": 5, "ft_lr": 0.0001, "ft_cap": 10000, "ft_loss": "combined", "augment": true, "l2sp": 0.0, "consistency": 0.5, "eval_rounds": true, "val_cap": 1024, "image_level": true} +[13:21:51][ 0s][DATA] MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation. +[13:21:51][ 0s][DATA] multiclass: train=1855 val=350 test=175 patches +[13:21:54][ 2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, dros-vnc=2.77 +[13:21:54][ 2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s +[13:21:55][ 4s][WEIGHTS] mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 | eff.sources=1.19 | MMD2 mix=0.0059 uniform=0.2484 +[13:21:55][ 4s][WEIGHTS] per-source: fly-mito=0.014, kasthuri=0.019, lucchi=0.962, dros-vnc=0.005 | min-MMD source (NeuroADDA rule) = lucchi +[13:22:06][ 15s][TEACHERS] target-val Dice: fly-mito=0.258, kasthuri=0.175, lucchi=0.968, dros-vnc=0.176 +[13:22:06][ 15s][TEACHERS] ensemble target-val Dice: mixture=0.967, mixture@raw=0.968, mixture@logit=0.967, per_source=0.967, per_source@raw=0.967, per_source@logit=0.966, uniform=0.248, uniform@raw=0.208, uniform@logit=0.195 +[13:22:06][ 15s][WEIGHTS] using mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008 +[13:22:06][ 15s][INIT] student initialised from teacher lucchi (weight 0.968) +[13:22:09][ 18s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches +[13:22:09][ 18s][KD] distilling the mixture ensemble (prob space) into the student +[13:22:32][ 40s][KD] soft labels for 1855 patches from 4 teachers in 23s +[13:23:06][ 74s][KD] epoch 1/5 | patches=1855 | loss=0.1033 | 34s +[13:23:39][ 108s][KD] epoch 2/5 | patches=1855 | loss=0.0740 | 34s +[13:24:13][ 141s][KD] epoch 3/5 | patches=1855 | loss=0.0711 | 34s +[13:24:46][ 175s][KD] epoch 4/5 | patches=1855 | loss=0.0696 | 34s +[13:25:19][ 208s][KD] epoch 5/5 | patches=1855 | loss=0.0683 | 33s +[13:25:20][ 209s][ROUND0] before any target label: val_dice=0.9646 val_vi=0.0433 +[13:25:20][ 209s][AL] patches=1855 (pool 1855 after dropping blank; free blank negatives: True) | budget=92 (5.0%) | rounds=6 | per_round=15 | acquisition=margin | first_round=stratified +[13:25:27][ 216s][AL] round-1 uncertainty quality on 1855 patches: {'spearman_uncertainty_error': -0.12135425581258891, 'auroc_top20_error': 0.2865643231304626} +[13:25:50][ 238s][AL] round 1: stratified cold start: 8 patches from the membrane-rich half (ensemble prevalence >= 0.071) + 7 from the membrane-poor half of 1855 candidates +[13:25:51][ 240s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.968, uniform=0.143, top1=0.968, top2=0.968, top3=0.968 -> using mixture +[13:25:51][ 240s][WEIGHTS] round 1: student kept (labeled Dice 0.970 vs best teacher lucchi 0.968); distilling the corrected ensemble into it +[13:25:51][ 240s][WEIGHTS] round 1: teacher Dice on 15 labeled patches + 0 blank (109027 fg px): fly-mito=0.395, kasthuri=0.000, lucchi=0.968, dros-vnc=0.016 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:26:14][ 262s][KD] soft labels for 1840 patches from 4 teachers in 22s +[13:26:47][ 296s][KD] epoch 1/2 | patches=1840 | loss=0.0450 | 33s +[13:27:20][ 329s][KD] epoch 2/2 | patches=1840 | loss=0.0443 | 33s +[13:27:41][ 350s][CONS] pseudo-labels for 1840 unlabeled patches from the current ensemble in 21s +[13:27:42][ 351s][FT] round 1/6 epoch 1/5 | patches=15 | loss=0.2088 +[13:27:43][ 352s][FT] round 1/6 epoch 2/5 | patches=15 | loss=0.1714 +[13:27:44][ 352s][FT] round 1/6 epoch 3/5 | patches=15 | loss=0.1610 +[13:27:44][ 353s][FT] round 1/6 epoch 4/5 | patches=15 | loss=0.1490 +[13:27:45][ 354s][FT] round 1/6 epoch 5/5 | patches=15 | loss=0.1664 +[13:27:46][ 355s][AL] round 1/6 | labeled=15 | selection_seconds=29.3000 round_seconds=144.9000 restart_student_dice=0.9697 restart_teacher_dice=0.9681 val_dice=0.9647 val_vi=0.0445 val_ece=0.0017 +[13:27:52][ 361s][AL] round 2: 283/1840 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:27:54][ 363s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.961, uniform=0.085, top1=0.961, top2=0.961, top3=0.961 -> using mixture +[13:27:54][ 363s][WEIGHTS] round 2: teacher Dice on 30 labeled patches + 0 blank (197342 fg px): fly-mito=0.358, kasthuri=0.000, lucchi=0.961, dros-vnc=0.009 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:27:55][ 363s][CONS] round 2: on the 15 new patches student Dice=0.944 vs ensemble 0.953 -> pseudo-labels from the ensemble | consistency x1.00 +[13:27:56][ 365s][FT] round 2/6 epoch 1/5 | patches=30 | loss=0.1608 +[13:27:57][ 366s][FT] round 2/6 epoch 2/5 | patches=30 | loss=0.1735 +[13:27:59][ 367s][FT] round 2/6 epoch 3/5 | patches=30 | loss=0.2296 +[13:28:00][ 369s][FT] round 2/6 epoch 4/5 | patches=30 | loss=0.1947 +[13:28:01][ 370s][FT] round 2/6 epoch 5/5 | patches=30 | loss=0.1665 +[13:28:03][ 371s][AL] round 2/6 | labeled=30 | selection_seconds=6.0000 round_seconds=14.9000 pseudo_student_dice=0.9443 pseudo_ensemble_dice=0.9533 consistency_scale=1.0000 val_dice=0.9464 val_vi=0.0466 val_ece=0.0019 +[13:28:09][ 377s][AL] round 3: 271/1825 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:28:12][ 380s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.959, uniform=0.074, top1=0.959, top2=0.959, top3=0.959 -> using mixture +[13:28:12][ 380s][WEIGHTS] round 3: teacher Dice on 45 labeled patches + 0 blank (273200 fg px): fly-mito=0.335, kasthuri=0.000, lucchi=0.959, dros-vnc=0.008 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:28:13][ 381s][CONS] round 3: on the 15 new patches student Dice=0.880 vs ensemble 0.952 -> pseudo-labels from the ensemble | consistency x1.00 +[13:28:15][ 383s][FT] round 3/6 epoch 1/5 | patches=45 | loss=0.1915 +[13:28:16][ 385s][FT] round 3/6 epoch 2/5 | patches=45 | loss=0.2314 +[13:28:18][ 387s][FT] round 3/6 epoch 3/5 | patches=45 | loss=0.1720 +[13:28:20][ 389s][FT] round 3/6 epoch 4/5 | patches=45 | loss=0.1663 +[13:28:22][ 391s][FT] round 3/6 epoch 5/5 | patches=45 | loss=0.1755 +[13:28:23][ 392s][AL] round 3/6 | labeled=45 | selection_seconds=6.3000 round_seconds=19.5000 pseudo_student_dice=0.8799 pseudo_ensemble_dice=0.9519 consistency_scale=1.0000 val_dice=0.9572 val_vi=0.0430 val_ece=0.0013 +[13:28:29][ 398s][AL] round 4: 284/1810 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:28:33][ 402s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.957, uniform=0.062, top1=0.957, top2=0.957, top3=0.957 -> using mixture +[13:28:33][ 402s][WEIGHTS] round 4: teacher Dice on 60 labeled patches + 0 blank (366894 fg px): fly-mito=0.341, kasthuri=0.000, lucchi=0.957, dros-vnc=0.009 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:28:34][ 403s][CONS] round 4: on the 15 new patches student Dice=0.932 vs ensemble 0.953 -> pseudo-labels from the ensemble | consistency x1.00 +[13:28:36][ 405s][FT] round 4/6 epoch 1/5 | patches=60 | loss=0.1762 +[13:28:39][ 407s][FT] round 4/6 epoch 2/5 | patches=60 | loss=0.1660 +[13:28:41][ 410s][FT] round 4/6 epoch 3/5 | patches=60 | loss=0.1585 +[13:28:44][ 412s][FT] round 4/6 epoch 4/5 | patches=60 | loss=0.1487 +[13:28:45][ 414s][FT] round 4/6 epoch 5/5 | patches=60 | loss=0.1477 +[13:28:46][ 415s][AL] round 4/6 | labeled=60 | selection_seconds=6.2000 round_seconds=21.9000 pseudo_student_dice=0.9319 pseudo_ensemble_dice=0.9529 consistency_scale=1.0000 val_dice=0.9630 val_vi=0.0449 val_ece=0.0018 +[13:28:52][ 421s][AL] round 5: 285/1795 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:28:57][ 426s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.954, uniform=0.052, top1=0.954, top2=0.954, top3=0.954 -> using mixture +[13:28:57][ 426s][WEIGHTS] round 5: teacher Dice on 75 labeled patches + 0 blank (453825 fg px): fly-mito=0.339, kasthuri=0.000, lucchi=0.954, dros-vnc=0.007 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:28:58][ 427s][CONS] round 5: on the 15 new patches student Dice=0.945 vs ensemble 0.941 -> pseudo-labels from the ensemble | consistency x1.00 +[13:29:01][ 430s][FT] round 5/6 epoch 1/5 | patches=75 | loss=0.1347 +[13:29:04][ 433s][FT] round 5/6 epoch 2/5 | patches=75 | loss=0.1555 +[13:29:07][ 436s][FT] round 5/6 epoch 3/5 | patches=75 | loss=0.1541 +[13:29:10][ 439s][FT] round 5/6 epoch 4/5 | patches=75 | loss=0.1501 +[13:29:13][ 441s][FT] round 5/6 epoch 5/5 | patches=75 | loss=0.1638 +[13:29:14][ 443s][AL] round 5/6 | labeled=75 | selection_seconds=6.2000 round_seconds=26.4000 pseudo_student_dice=0.9449 pseudo_ensemble_dice=0.9407 consistency_scale=1.0000 val_dice=0.9528 val_vi=0.0434 val_ece=0.0015 +[13:29:20][ 449s][AL] round 6: 279/1780 candidates skipped as agreed-empty (predicted prevalence < 0.005) +[13:29:25][ 454s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.952, uniform=0.091, top1=0.952, top2=0.952, top3=0.952 -> using mixture +[13:29:25][ 454s][WEIGHTS] round 6: teacher Dice on 90 labeled patches + 0 blank (551201 fg px): fly-mito=0.341, kasthuri=0.000, lucchi=0.952, dros-vnc=0.016 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:29:26][ 455s][CONS] round 6: on the 15 new patches student Dice=0.924 vs ensemble 0.943 -> pseudo-labels from the ensemble | consistency x1.00 +[13:29:30][ 458s][FT] round 6/6 epoch 1/5 | patches=90 | loss=0.1687 +[13:29:33][ 462s][FT] round 6/6 epoch 2/5 | patches=90 | loss=0.1609 +[13:29:37][ 465s][FT] round 6/6 epoch 3/5 | patches=90 | loss=0.1521 +[13:29:40][ 469s][FT] round 6/6 epoch 4/5 | patches=90 | loss=0.1583 +[13:29:44][ 473s][FT] round 6/6 epoch 5/5 | patches=90 | loss=0.1500 +[13:29:45][ 474s][AL] round 6/6 | labeled=90 | selection_seconds=6.1000 round_seconds=29.8000 pseudo_student_dice=0.9242 pseudo_ensemble_dice=0.9427 consistency_scale=1.0000 val_dice=0.9556 val_vi=0.0441 val_ece=0.0010 +[13:29:49][ 477s][FT] final 1/10 epoch 1/1 | patches=90 | loss=0.1325 +[13:29:52][ 481s][FT] final 2/10 epoch 1/1 | patches=90 | loss=0.1251 +[13:29:56][ 484s][FT] final 3/10 epoch 1/1 | patches=90 | loss=0.1262 +[13:29:59][ 488s][FT] final 4/10 epoch 1/1 | patches=90 | loss=0.1276 +[13:30:03][ 492s][FT] final 5/10 epoch 1/1 | patches=90 | loss=0.1145 +[13:30:06][ 495s][FT] final 6/10 epoch 1/1 | patches=90 | loss=0.1159 +[13:30:10][ 499s][FT] final 7/10 epoch 1/1 | patches=90 | loss=0.1161 +[13:30:14][ 502s][FT] final 8/10 epoch 1/1 | patches=90 | loss=0.1173 +[13:30:17][ 506s][FT] final 9/10 epoch 1/1 | patches=90 | loss=0.1098 +[13:30:21][ 509s][FT] final 10/10 epoch 1/1 | patches=90 | loss=0.1159 +[13:30:22][ 510s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=35.4000 val_dice=0.9630 val_vi=0.0417 val_ece=0.0015 +[13:30:22][ 510s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000 +[13:30:27][ 515s][TEST] raw student (no EMA, no TTA): patch_dice=0.9052 patch_vi=0.0768 +[13:30:28][ 516s][TEST] final model without TTA: patch_dice=0.9052 patch_vi=0.0768 +[13:30:28][ 516s][RESOURCES] {"phases": {"setup": {"seconds": 2.18, "peak_cpu_rss_mb": 1025.7, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 15.34, "peak_cpu_rss_mb": 1803.5, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 190.04, "peak_cpu_rss_mb": 2393.1, "peak_gpu_alloc_mb": 3829.8, "peak_gpu_reserved_mb": 4148.0, "kd_epochs": 5}, "active_learning": {"seconds": 301.43, "peak_cpu_rss_mb": 2670.2, "peak_gpu_alloc_mb": 3950.0, "peak_gpu_reserved_mb": 4156.0, "rounds": 6}, "inference": {"seconds": 5.69, "peak_cpu_rss_mb": 2670.2, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4156.0, "patches": 175, "images": 5, "patches_per_second": 30.8, "seconds_per_image": 1.138}}, "peak_gpu_alloc_mb": 3950.0, "peak_cpu_rss_mb": 2670.2, "total_seconds": 514.7} +[13:30:28][ 516s][TEST] patch_dice=0.8784 patch_iou=0.8286 patch_precision=0.9633 patch_recall=0.8551 patch_vi=0.0950 image_dice=0.9013 image_iou=0.8257 image_precision=0.9688 image_recall=0.8529 image_vi=0.1099 cal_ece=0.0077 cal_nll=0.0592 cal_brier=0.0111 cal_mean_confidence=0.9941 cal_accuracy=0.9865 cal_predicted_prevalence=0.0811 cal_true_prevalence=0.0886 cal_mean_probability=0.0809 +[13:30:28][ 516s][DONE] multiclass macs-v11-norestart-acq-margin b5 s3 | test patch Dice=0.8784 VI=0.0950 | image Dice=0.9013 VI=0.1099 | 516s diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/result.json b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/result.json new file mode 100644 index 0000000000000000000000000000000000000000..51aa55d198fb65cd8b064b2bb45f28704c72296e --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/result.json @@ -0,0 +1,647 @@ +{ + "target": "multiclass", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 3, + "config": { + "task": "mitochondria", + "target": "multiclass", + "variant": "macs-v11-norestart-acq-margin", + "budget": 5.0, + "seed": 3, + "sources": "", + "run_tag": "", + "arch": "unet", + "teacher_ckpt": "model_best.pth", + "models_dir": "/home/user/connectomics/MACS-2/models", + "cache_dir": "/home/user/connectomics/MACS-2/cache/macs_v2", + "output_dir": "experiments/macs_v2/outputs-r26", + "checkpoint_dir": "/home/user/connectomics/MACS-2/models/experiments/macs_v2", + "device": "cuda:0", + "num_workers": 5, + "batch_size": 8, + "eval_batch_size": 16, + "amp": true, + "weighting": "mixture", + "feature_n": 512, + "feature_seed": 0, + "source_split": "train", + "kernel_scales": "0.25,0.5,1,2,4", + "lam": 0.01, + "tau": 0.5, + "calibration": "source-val", + "ensemble_space": "prob", + "student_init": "top-source", + "init_compatible": false, + "weight_update": "labeled", + "reliability_beta": 10.0, + "update_cap": 512, + "kd_after_update": 2, + "reinit_after_update": true, + "min_consistency": 64, + "consistency_cap": 2000, + "select_ensemble": true, + "pseudo_source": "auto", + "final_epochs": 10, + "evidence": "pooled", + "min_evidence_fg": 2000, + "restart_rule": "evidence", + "ema": 0.0, + "conf_weight": false, + "skip_blank": true, + "free_blank": true, + "aug_intensity": false, + "final_pool": false, + "final_pool_cap": 8000, + "tta": true, + "blank_std": 0.01, + "kd": true, + "kd_epochs": 5, + "kd_lr": 0.0001, + "kd_cap": 6000, + "kd_kl_weight": 0.0, + "rounds": 6, + "label_schedule": "rounds", + "eval_test_rounds": "", + "acquisition": "margin", + "first_round": "stratified", + "cold_cap": 2000, + "recover": false, + "recover_margin": 0.1, + "recover_max": 2, + "recover_big": 0.3, + "recover_epochs": 3, + "adabn": true, + "adabn_cap": 512, + "restart_margin": 9.0, + "skip_empty": true, + "empty_prevalence": 0.005, + "score_pool_cap": 8000, + "lap_precision": "pool", + "lap_score": "entropy+var", + "lap_scale": 1.0, + "deconf_prev": false, + "diversity": "none", + "div_mult": 4, + "evidence_explore_only": false, + "pseudo_switch_margin": 0.05, + "cons_gate": true, + "cons_gate_lo": 0.5, + "cons_gate_hi": 0.8, + "pool_evidence": 0.5, + "pool_evidence_n": 2000, + "oracle_evidence": false, + "typicality": 0.0, + "evidence_rounds": 0, + "explore_frac": 0.0, + "strat_bins": 10, + "dump_signals": "", + "mc_passes": 5, + "mc_dropout": 0.2, + "ft_epochs": 5, + "ft_lr": 0.0001, + "ft_cap": 10000, + "ft_loss": "combined", + "augment": true, + "l2sp": 0.0, + "consistency": 0.5, + "eval_rounds": true, + "val_cap": 1024, + "image_level": true + }, + "commit": "aecf380", + "data_access": "MMD weights and distillation use the unlabeled target train pool only; teacher temperatures are fitted on each teacher's own source validation set; target validation labels are used only for reported diagnostics, per-round monitoring and the explicitly named best-single oracle; test labels are used once, for the final evaluation.", + "environment": { + "host": "pc2", + "python": "3.14.7", + "torch": "2.14.1+cu130", + "argv": [ + "experiments/macs_v2/run.py", + "sweep", + "--targets", + "multiclass", + "--variants", + "macs-v11-norestart-acq-margin", + "--budgets", + "5", + "--seeds", + "3", + "--num-workers", + "5", + "--output-dir", + "experiments/macs_v2/outputs-v3", + "--task", + "mitochondria", + "--output-dir", + "experiments/macs_v2/outputs-r26" + ], + "cwd": "/home/user/connectomics/MACS-2", + "gpu": "NVIDIA GeForce RTX 4090", + "cuda": "13.0", + "commit": "aecf380", + "dirty": true + }, + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "alphas": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "weights_all": { + "sources": [ + "fly-brain-mito-benchmark", + "kasthuri-pp", + "lucchi-pp", + "drosophila-vnc" + ], + "mixture": [ + 0.01152107526225956, + 0.011843427771872628, + 0.968396188229585, + 0.008239308736282923 + ], + "per_source": [ + 0.014321466605441749, + 0.018762803993963635, + 0.9619825923792787, + 0.004933137021315701 + ], + "uniform": [ + 0.25, + 0.25, + 0.25, + 0.25 + ], + "mixture_info": { + "objective": 0.004128299797828663, + "mmd2_mixture": 0.005874306045373645, + "mmd2_uniform": 0.24840696635360943, + "mmd2_single_source": [ + 0.6122003026141286, + 0.5667974971994383, + 0.004845544051820361, + 0.668933003056658 + ], + "median_squared_distance": 0.20473960170977523, + "gammas": [ + 1.221063233064128, + 2.442126466128256, + 4.884252932256512, + 9.768505864513024, + 19.537011729026048 + ], + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "per_source_info": { + "mmd2_own_rkhs": [ + 0.6529830716388123, + 0.6112535306587584, + 0.0030196383578795505, + 0.8176329290240223 + ], + "standardized": [ + 0.4264461417532283, + 0.2913877286688084, + -1.6771722494431436, + 0.9593383790211064 + ], + "tau": 0.5, + "max_weight": 0.9619825923792787, + "entropy": 0.19889667121630067, + "effective_sources": 1.2200558924950622, + "min_weight": 0.004933137021315701 + }, + "lam": 0.01, + "tau": 0.5, + "single_mmd_source": "lucchi-pp", + "seconds": 1.4 + }, + "teacher_diagnostics": { + "teacher_val_dice": [ + 0.25842773603204217, + 0.17499716869513837, + 0.9675530639716557, + 0.17625312705123047 + ], + "ensemble_val_dice": { + "mixture": 0.9673733701024737, + "mixture@raw": 0.9675071736744472, + "mixture@logit": 0.9667425331047603, + "per_source": 0.9673222165448325, + "per_source@raw": 0.967496030160359, + "per_source@logit": 0.9664733615943364, + "uniform": 0.24779489094285445, + "uniform@raw": 0.20821287741695416, + "uniform@logit": 0.19508316544090382 + }, + "val_patches": 350 + }, + "temperatures": [ + 1.3819772005081177, + 2.5080416202545166, + 3.7969236373901367, + 2.767761468887329 + ], + "top_source": "lucchi-pp", + "weight_stats": { + "max_weight": 0.968396188229585, + "entropy": 0.17460062475449822, + "effective_sources": 1.1907705573517349, + "min_weight": 0.008239308736282923 + }, + "correlations": { + "spearman_weight_teacher_dice": 0.39999999999999997, + "spearman_weight_taxonomy": -0.8944271909999159, + "top_source": "lucchi-pp", + "top_source_is_best_teacher": true + }, + "init_teacher": "lucchi-pp", + "student_init": "top-source", + "blank_fraction": 0.0, + "kd_seconds": 190.0, + "round0_val": { + "val_dice": 0.9645588601012774, + "val_vi": 0.04328153081053755, + "val_ece": 0.009755093868701359 + }, + "rounds": [ + { + "round": 1, + "n_labeled": 15, + "selection_seconds": 29.3, + "round_seconds": 144.9, + "teacher_dice_labeled": [ + 0.3945454313577119, + 9.171955827860732e-06, + 0.9680883362682224, + 0.01620298768172638 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 109027, + "evidence_nonempty_patches": 14, + "first_correction": true, + "restart": "kept", + "restart_student_dice": 0.969730043704533, + "restart_teacher_dice": 0.9680883362682224, + "weights": [ + 3.8420009317991516e-05, + 7.639987786195338e-07, + 0.9999601910568676, + 6.249350358019625e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9680835162782008, + "uniform": 0.14307490547399257, + "top1": 0.9680883362682224, + "top2": 0.9680835162782008, + "top3": 0.9680835162782008 + }, + "val_dice": 0.9646718113609869, + "val_vi": 0.04446548615290779, + "val_ece": 0.001694864422380892 + }, + { + "round": 2, + "n_labeled": 30, + "selection_seconds": 6.0, + "round_seconds": 14.9, + "teacher_dice_labeled": [ + 0.3579099625216992, + 5.067319337397324e-06, + 0.961406755419766, + 0.008983697187361285 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 197342, + "evidence_nonempty_patches": 29, + "first_correction": false, + "restart": "", + "weights": [ + 2.8475652254189208e-05, + 8.16764536171835e-07, + 0.9999700859933408, + 6.215898688641217e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.961406755419766, + "uniform": 0.0846112782684655, + "top1": 0.961406755419766, + "top2": 0.961406755419766, + "top3": 0.961406755419766 + }, + "pseudo_student_dice": 0.9443227505892139, + "pseudo_ensemble_dice": 0.9532819103012677, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9463882777703261, + "val_vi": 0.046568383953680265, + "val_ece": 0.0019284904371264454 + }, + { + "round": 3, + "n_labeled": 45, + "selection_seconds": 6.3, + "round_seconds": 19.5, + "teacher_dice_labeled": [ + 0.334846145721047, + 3.6368000523699207e-06, + 0.9587246522139048, + 0.007838921282798833 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 273200, + "evidence_nonempty_patches": 44, + "first_correction": false, + "restart": "", + "weights": [ + 2.3225190757088682e-05, + 8.389598020190152e-07, + 0.9999753046265767, + 6.312228641801486e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9587265716013154, + "uniform": 0.07402254314899613, + "top1": 0.9587246522139048, + "top2": 0.9587265716013154, + "top3": 0.9587265716013154 + }, + "pseudo_student_dice": 0.8799494374020794, + "pseudo_ensemble_dice": 0.9518636903628929, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9571846395854329, + "val_vi": 0.04296463556814065, + "val_ece": 0.001267404570853909 + }, + { + "round": 4, + "n_labeled": 60, + "selection_seconds": 6.2, + "round_seconds": 21.9, + "teacher_dice_labeled": [ + 0.3412943407859468, + 2.712519089353091e-06, + 0.957265958913553, + 0.009216452411489247 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 366894, + "evidence_nonempty_patches": 59, + "first_correction": false, + "restart": "", + "weights": [ + 2.513608442753139e-05, + 8.512778234358265e-07, + 0.9999733632569481, + 6.493808010523167e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9572673965705273, + "uniform": 0.06248878426736196, + "top1": 0.957265958913553, + "top2": 0.9572673965705273, + "top3": 0.9572673965705273 + }, + "pseudo_student_dice": 0.9318661924753322, + "pseudo_ensemble_dice": 0.9529371218503485, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9630231512104639, + "val_vi": 0.04489372802810563, + "val_ece": 0.0017532539362966177 + }, + { + "round": 5, + "n_labeled": 75, + "selection_seconds": 6.2, + "round_seconds": 26.4, + "teacher_dice_labeled": [ + 0.33902580528091814, + 2.1949463555110714e-06, + 0.9540560867686285, + 0.007458396236351508 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 453825, + "evidence_nonempty_patches": 74, + "first_correction": false, + "restart": "", + "weights": [ + 2.5373809797839377e-05, + 8.790412395696186e-07, + 0.9999730882717588, + 6.588772036961855e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9540560867686285, + "uniform": 0.051831891234340906, + "top1": 0.9540560867686285, + "top2": 0.9540560867686285, + "top3": 0.9540560867686285 + }, + "pseudo_student_dice": 0.9448709516465572, + "pseudo_ensemble_dice": 0.9407407831070338, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9527958166733158, + "val_vi": 0.043406650994652615, + "val_ece": 0.0015177299427026481 + }, + { + "round": 6, + "n_labeled": 90, + "selection_seconds": 6.1, + "round_seconds": 29.8, + "teacher_dice_labeled": [ + 0.34052004619608056, + 1.8084229105481692e-06, + 0.9520722013446463, + 0.016366461973472624 + ], + "teacher_dice_pool": null, + "evidence_fg_pixels": 551201, + "evidence_nonempty_patches": 89, + "first_correction": false, + "restart": "", + "weights": [ + 2.6271844217384235e-05, + 8.96650192034205e-07, + 0.9999720968104048, + 7.34695185747632e-07 + ], + "ensemble_choice": "mixture", + "candidate_dice": { + "mixture": 0.9520731627877532, + "uniform": 0.09109870974219077, + "top1": 0.9520722013446463, + "top2": 0.9520731627877532, + "top3": 0.9520731627877532 + }, + "pseudo_student_dice": 0.9241799918541361, + "pseudo_ensemble_dice": 0.9426643204855331, + "pseudo_source": "ensemble", + "consistency_scale": 1.0, + "val_dice": 0.9555770518205333, + "val_vi": 0.044131102459704336, + "val_ece": 0.0010309637004614577 + }, + { + "round": 7, + "n_labeled": 90, + "selection_seconds": 0.0, + "round_seconds": 35.4, + "final_consolidation": true, + "val_dice": 0.963027841173535, + "val_vi": 0.041729577899810986, + "val_ece": 0.0014772768619997602 + } + ], + "uncertainty_quality": { + "spearman_uncertainty_error": -0.12135425581258891, + "auroc_top20_error": 0.2865643231304626 + }, + "al_seconds": 301.4, + "source_archs": [ + "unet", + "unet", + "unet", + "unet" + ], + "alphas_final": [ + 2.6271844217384235e-05, + 8.96650192034205e-07, + 0.9999720968104048, + 7.34695185747632e-07 + ], + "weight_stats_final": { + "max_weight": 0.9999720968104048, + "entropy": 0.00032785446693755626, + "effective_sources": 1.0003279082170873, + "min_weight": 7.34695185747632e-07 + }, + "top_source_final": "lucchi-pp", + "ensemble_choice_final": "", + "restart": "kept", + "test_raw_student": { + "patch_dice": 0.9052153478097684, + "patch_iou": 0.8682025653632535, + "patch_precision": 0.9724860943271315, + "patch_recall": 0.8932489326941735, + "patch_vi": 0.07676159361526944, + "n_patches": 175, + "cal_ece": 0.007511254251670856, + "cal_nll": 0.061821238321234276, + "cal_brier": 0.009365646389635138, + "cal_mean_confidence": 0.9968699254128839, + "cal_accuracy": 0.9893917410714286, + "cal_predicted_prevalence": 0.08301339285714286, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08280276414339341 + }, + "test_notta": { + "patch_dice": 0.9052153478097684, + "patch_iou": 0.8682025653632535, + "patch_precision": 0.9724860943271315, + "patch_recall": 0.8932489326941735, + "patch_vi": 0.07676159361526944, + "n_patches": 175, + "cal_ece": 0.007511254251670856, + "cal_nll": 0.061821238321234276, + "cal_brier": 0.009365646389635138, + "cal_mean_confidence": 0.9968699254128839, + "cal_accuracy": 0.9893917410714286, + "cal_predicted_prevalence": 0.08301339285714286, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08280276414339341 + }, + "test": { + "patch_dice": 0.878351039391282, + "patch_iou": 0.8286347674570593, + "patch_precision": 0.9632866435654643, + "patch_recall": 0.855115361170372, + "patch_vi": 0.09501608484524254, + "n_patches": 175, + "image_dice": 0.9012982904929572, + "image_iou": 0.825731968256313, + "image_precision": 0.9688168951078062, + "image_recall": 0.8528783062353946, + "image_vi": 0.10985392283216525, + "n_images": 5, + "cal_ece": 0.007677805701579884, + "cal_nll": 0.0592324220305252, + "cal_brier": 0.011119732447192372, + "cal_mean_confidence": 0.9940822700911102, + "cal_accuracy": 0.9865122767857143, + "cal_predicted_prevalence": 0.08111607142857143, + "cal_true_prevalence": 0.08861049107142857, + "cal_mean_probability": 0.08087745002127328 + }, + "eval_seconds": 5.7, + "total_seconds": 516.1, + "resources": { + "phases": { + "setup": { + "seconds": 2.18, + "peak_cpu_rss_mb": 1025.7, + "peak_gpu_alloc_mb": 592.4, + "peak_gpu_reserved_mb": 594.0 + }, + "features_and_weights": { + "seconds": 15.34, + "peak_cpu_rss_mb": 1803.5, + "peak_gpu_alloc_mb": 2298.2, + "peak_gpu_reserved_mb": 2640.0 + }, + "distillation": { + "seconds": 190.04, + "peak_cpu_rss_mb": 2393.1, + "peak_gpu_alloc_mb": 3829.8, + "peak_gpu_reserved_mb": 4148.0, + "kd_epochs": 5 + }, + "active_learning": { + "seconds": 301.43, + "peak_cpu_rss_mb": 2670.2, + "peak_gpu_alloc_mb": 3950.0, + "peak_gpu_reserved_mb": 4156.0, + "rounds": 6 + }, + "inference": { + "seconds": 5.69, + "peak_cpu_rss_mb": 2670.2, + "peak_gpu_alloc_mb": 2149.5, + "peak_gpu_reserved_mb": 4156.0, + "patches": 175, + "images": 5, + "patches_per_second": 30.8, + "seconds_per_image": 1.138 + } + }, + "peak_gpu_alloc_mb": 3950.0, + "peak_cpu_rss_mb": 2670.2, + "total_seconds": 514.7 + }, + "student_parameters": 31042369, + "teacher_parameters": [ + 31042369, + 31042369, + 31042369, + 31042369 + ], + "checkpoint": "/home/user/connectomics/MACS-2/models/experiments/macs_v2/multiclass/macs-v11-norestart-acq-margin_b5_s3.pth" +} \ No newline at end of file diff --git a/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/rounds.csv b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/rounds.csv new file mode 100644 index 0000000000000000000000000000000000000000..9265657e3caaa86dcec766bb4eafae6d2e86cfd6 --- /dev/null +++ b/macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-margin/budget-5/seed-3/rounds.csv @@ -0,0 +1,8 @@ +round,n_labeled,selection_seconds,round_seconds,teacher_dice_labeled,teacher_dice_pool,evidence_fg_pixels,evidence_nonempty_patches,first_correction,restart,restart_student_dice,restart_teacher_dice,weights,ensemble_choice,candidate_dice,val_dice,val_vi,val_ece,pseudo_student_dice,pseudo_ensemble_dice,pseudo_source,consistency_scale,final_consolidation +1,15,29.3,144.9,"[0.3945454313577119, 9.171955827860732e-06, 0.9680883362682224, 0.01620298768172638]",,109027,14,True,kept,0.969730043704533,0.9680883362682224,"[3.8420009317991516e-05, 7.639987786195338e-07, 0.9999601910568676, 6.249350358019625e-07]",mixture,"{'mixture': 0.9680835162782008, 'uniform': 0.14307490547399257, 'top1': 0.9680883362682224, 'top2': 0.9680835162782008, 'top3': 0.9680835162782008}",0.9646718113609869,0.04446548615290779,0.001694864422380892,,,,, +2,30,6.0,14.9,"[0.3579099625216992, 5.067319337397324e-06, 0.961406755419766, 0.008983697187361285]",,197342,29,False,,,,"[2.8475652254189208e-05, 8.16764536171835e-07, 0.9999700859933408, 6.215898688641217e-07]",mixture,"{'mixture': 0.961406755419766, 'uniform': 0.0846112782684655, 'top1': 0.961406755419766, 'top2': 0.961406755419766, 'top3': 0.961406755419766}",0.9463882777703261,0.046568383953680265,0.0019284904371264454,0.9443227505892139,0.9532819103012677,ensemble,1.0, +3,45,6.3,19.5,"[0.334846145721047, 3.6368000523699207e-06, 0.9587246522139048, 0.007838921282798833]",,273200,44,False,,,,"[2.3225190757088682e-05, 8.389598020190152e-07, 0.9999753046265767, 6.312228641801486e-07]",mixture,"{'mixture': 0.9587265716013154, 'uniform': 0.07402254314899613, 'top1': 0.9587246522139048, 'top2': 0.9587265716013154, 'top3': 0.9587265716013154}",0.9571846395854329,0.04296463556814065,0.001267404570853909,0.8799494374020794,0.9518636903628929,ensemble,1.0, +4,60,6.2,21.9,"[0.3412943407859468, 2.712519089353091e-06, 0.957265958913553, 0.009216452411489247]",,366894,59,False,,,,"[2.513608442753139e-05, 8.512778234358265e-07, 0.9999733632569481, 6.493808010523167e-07]",mixture,"{'mixture': 0.9572673965705273, 'uniform': 0.06248878426736196, 'top1': 0.957265958913553, 'top2': 0.9572673965705273, 'top3': 0.9572673965705273}",0.9630231512104639,0.04489372802810563,0.0017532539362966177,0.9318661924753322,0.9529371218503485,ensemble,1.0, +5,75,6.2,26.4,"[0.33902580528091814, 2.1949463555110714e-06, 0.9540560867686285, 0.007458396236351508]",,453825,74,False,,,,"[2.5373809797839377e-05, 8.790412395696186e-07, 0.9999730882717588, 6.588772036961855e-07]",mixture,"{'mixture': 0.9540560867686285, 'uniform': 0.051831891234340906, 'top1': 0.9540560867686285, 'top2': 0.9540560867686285, 'top3': 0.9540560867686285}",0.9527958166733158,0.043406650994652615,0.0015177299427026481,0.9448709516465572,0.9407407831070338,ensemble,1.0, +6,90,6.1,29.8,"[0.34052004619608056, 1.8084229105481692e-06, 0.9520722013446463, 0.016366461973472624]",,551201,89,False,,,,"[2.6271844217384235e-05, 8.96650192034205e-07, 0.9999720968104048, 7.34695185747632e-07]",mixture,"{'mixture': 0.9520731627877532, 'uniform': 0.09109870974219077, 'top1': 0.9520722013446463, 'top2': 0.9520731627877532, 'top3': 0.9520731627877532}",0.9555770518205333,0.044131102459704336,0.0010309637004614577,0.9241799918541361,0.9426643204855331,ensemble,1.0, +7,90,0.0,35.4,,,,,,,,,,,,0.963027841173535,0.041729577899810986,0.0014772768619997602,,,,,True