macs_v2 runs from bayzid-system-2 2026-10-24_22:51

macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-entropy/budget-5/seed-1/log.txt

@@ -0,0 +1,107 @@
+[03:49:50][      0s][START] target=dhanyasi-P14-mouse-cerebellum variant=macs-v11-norestart-acq-entropy budget=5.0% seed=1 | {"task": "membrane", "target": "dhanyasi-P14-mouse-cerebellum", "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": 8, "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}
+[03:49:50][      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.
+[03:49:51][      0s][DATA] dhanyasi-P14-mouse-cerebellum: train=8649 val=1922 test=2883 patches
+[03:49:53][      2s][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=4.07
+[03:49:53][      3s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s
+[03:50:00][      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
+[03:50:00][      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
+[03:51:14][     84s][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
+[03:51:14][     84s][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
+[03:51:14][     84s][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
+[03:51:14][     84s][INIT] student initialised from teacher microns (weight 0.417)
+[03:51:18][     88s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[03:51:19][     88s][KD] distilling the mixture ensemble (prob space) into the student
+[03:55:24][    333s][KD] soft labels for 6000 patches from 8 teachers in 245s
+[03:58:22][    512s][KD] epoch 1/5 | patches=6000 | loss=0.5816 | 179s
+[04:01:21][    691s][KD] epoch 2/5 | patches=6000 | loss=0.4493 | 179s
+[04:04:20][    870s][KD] epoch 3/5 | patches=6000 | loss=0.4312 | 179s
+[04:07:19][   1048s][KD] epoch 4/5 | patches=6000 | loss=0.4244 | 179s
+[04:10:18][   1227s][KD] epoch 5/5 | patches=6000 | loss=0.4237 | 179s
+[04:10:24][   1233s][ROUND0] before any target label: val_dice=0.1287 val_vi=0.5096
+[04:10:24][   1233s][AL] patches=8649 (pool 8647 after dropping blank; free blank negatives: True) | budget=432 (5.0%) | rounds=6 | per_round=72 | acquisition=entropy | first_round=stratified
+[04:11:06][   1276s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': -0.5377973366328569, 'auroc_top20_error': 0.224696484375}
+[04:12:28][   1358s][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
+[04:12:42][   1372s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.548, uniform=0.270, top1=0.481, top2=0.492, top3=0.499 -> using mixture
+[04:12:45][   1374s][WEIGHTS] round 1: student kept (labeled Dice 0.242 vs best teacher mouse-brain 0.481); distilling the corrected ensemble into it
+[04:12:45][   1374s][WEIGHTS] round 1: teacher Dice on 72 labeled patches + 2 blank (1139017 fg px): c-elegans=0.423, flywire=0.157, berghia=0.194, microns=0.305, oct-glia=0.168, oct-sfl=0.436, snemi=0.411, mouse-brain=0.481 | weights: c-elegans=0.079, flywire=0.017, berghia=0.002, microns=0.200, oct-glia=0.005, oct-sfl=0.095, snemi=0.147, mouse-brain=0.456
+[04:16:49][   1619s][KD] soft labels for 6000 patches from 8 teachers in 245s
+[04:19:48][   1798s][KD] epoch 1/2 | patches=6000 | loss=0.5616 | 179s
+[04:22:47][   1977s][KD] epoch 2/2 | patches=6000 | loss=0.5540 | 179s
+[04:24:09][   2059s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 82s
+[04:24:16][   2066s][FT] round 1/6 epoch 1/5 | patches=74 | loss=1.3421
+[04:24:23][   2073s][FT] round 1/6 epoch 2/5 | patches=74 | loss=1.3098
+[04:24:30][   2079s][FT] round 1/6 epoch 3/5 | patches=74 | loss=1.2440
+[04:24:37][   2086s][FT] round 1/6 epoch 4/5 | patches=74 | loss=1.2534
+[04:24:43][   2093s][FT] round 1/6 epoch 5/5 | patches=74 | loss=1.2370
+[04:24:49][   2099s][AL] round 1/6 | labeled=72 | selection_seconds=124.6000 round_seconds=859.5000 restart_student_dice=0.2421 restart_teacher_dice=0.4806 val_dice=0.6204 val_vi=0.6482 val_ece=0.0404
+[04:25:31][   2141s][AL] round 2: 824/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[04:25:56][   2166s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.388, uniform=0.184, top1=0.368, top2=0.399, top3=0.425 -> using top3
+[04:25:56][   2166s][WEIGHTS] round 2: teacher Dice on 144 labeled patches + 2 blank (1961159 fg px): c-elegans=0.312, flywire=0.118, berghia=0.162, microns=0.220, oct-glia=0.106, oct-sfl=0.316, snemi=0.360, mouse-brain=0.368 | weights: c-elegans=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.108, snemi=0.335, mouse-brain=0.557
+[04:26:16][   2186s][CONS] round 2: on the 72 new patches student Dice=0.507 vs ensemble 0.287 -> pseudo-labels from the student | consistency x0.02
+[04:26:28][   2197s][FT] round 2/6 epoch 1/5 | patches=146 | loss=1.1134
+[04:26:38][   2208s][FT] round 2/6 epoch 2/5 | patches=146 | loss=0.9894
+[04:26:50][   2219s][FT] round 2/6 epoch 3/5 | patches=146 | loss=0.9750
+[04:27:00][   2230s][FT] round 2/6 epoch 4/5 | patches=146 | loss=0.9333
+[04:27:11][   2241s][FT] round 2/6 epoch 5/5 | patches=146 | loss=0.9466
+[04:27:17][   2247s][AL] round 2/6 | labeled=144 | selection_seconds=42.1000 round_seconds=142.0000 pseudo_student_dice=0.5069 pseudo_ensemble_dice=0.2872 consistency_scale=0.0230 val_dice=0.5836 val_vi=0.6030 val_ece=0.0512
+[04:27:59][   2289s][AL] round 3: 1378/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[04:30:53][   2463s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.380, flywire=0.188, berghia=0.176, microns=0.335, oct-glia=0.214, oct-sfl=0.499, snemi=0.431, mouse-brain=0.452 | blended with labeled Dice at 0.50
+[04:33:37][   2626s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.455, uniform=0.218, top1=0.465, top2=0.474, top3=0.473 -> using top3
+[04:33:37][   2626s][WEIGHTS] round 3: teacher Dice on 216 labeled patches + 2 blank (3836186 fg px): c-elegans=0.368, flywire=0.136, berghia=0.159, microns=0.249, oct-glia=0.137, oct-sfl=0.363, snemi=0.372, mouse-brain=0.465 | weights: c-elegans=0.085, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.000, snemi=0.186, mouse-brain=0.729
+[04:33:47][   2636s][CONS] round 3: on the 72 new patches student Dice=0.584 vs ensemble 0.653 -> pseudo-labels from the ensemble | consistency x1.00
+[04:34:01][   2651s][FT] round 3/6 epoch 1/5 | patches=218 | loss=1.4325
+[04:34:16][   2666s][FT] round 3/6 epoch 2/5 | patches=218 | loss=1.4260
+[04:34:32][   2681s][FT] round 3/6 epoch 3/5 | patches=218 | loss=1.2945
+[04:34:47][   2696s][FT] round 3/6 epoch 4/5 | patches=218 | loss=1.2789
+[04:35:02][   2712s][FT] round 3/6 epoch 5/5 | patches=218 | loss=1.2626
+[04:35:08][   2718s][AL] round 3/6 | labeled=216 | selection_seconds=42.1000 round_seconds=464.8000 pseudo_student_dice=0.5842 pseudo_ensemble_dice=0.6535 consistency_scale=1.0000 val_dice=0.6468 val_vi=0.5581 val_ece=0.0517
+[04:35:50][   2760s][AL] round 4: 1289/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[04:36:36][   2806s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.434, uniform=0.133, top1=0.451, top2=0.451, top3=0.450 -> using top3
+[04:36:36][   2806s][WEIGHTS] round 4: teacher Dice on 288 labeled patches + 2 blank (4894951 fg px): c-elegans=0.358, flywire=0.098, berghia=0.171, microns=0.172, oct-glia=0.055, oct-sfl=0.259, snemi=0.338, mouse-brain=0.451 | weights: c-elegans=0.092, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.000, snemi=0.158, mouse-brain=0.750
+[04:36:56][   2826s][CONS] round 4: on the 72 new patches student Dice=0.644 vs ensemble 0.369 -> pseudo-labels from the student | consistency x0.48
+[04:37:15][   2845s][FT] round 4/6 epoch 1/5 | patches=290 | loss=0.9723
+[04:37:34][   2864s][FT] round 4/6 epoch 2/5 | patches=290 | loss=0.9621
+[04:37:54][   2883s][FT] round 4/6 epoch 3/5 | patches=290 | loss=0.9220
+[04:38:13][   2903s][FT] round 4/6 epoch 4/5 | patches=290 | loss=0.9362
+[04:38:33][   2922s][FT] round 4/6 epoch 5/5 | patches=290 | loss=0.9013
+[04:38:39][   2929s][AL] round 4/6 | labeled=288 | selection_seconds=42.1000 round_seconds=204.8000 pseudo_student_dice=0.6440 pseudo_ensemble_dice=0.3688 consistency_scale=0.4800 val_dice=0.6459 val_vi=0.5368 val_ece=0.0249
+[04:39:21][   2971s][AL] round 5: 1384/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[04:42:25][   3155s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.388, flywire=0.168, berghia=0.173, microns=0.320, oct-glia=0.212, oct-sfl=0.460, snemi=0.435, mouse-brain=0.471 | blended with labeled Dice at 0.50
+[04:45:20][   3329s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.473, uniform=0.222, top1=0.462, top2=0.473, top3=0.472 -> using mixture
+[04:45:20][   3329s][WEIGHTS] round 5: teacher Dice on 360 labeled patches + 2 blank (6110421 fg px): c-elegans=0.380, flywire=0.136, berghia=0.177, microns=0.254, oct-glia=0.133, oct-sfl=0.368, snemi=0.396, mouse-brain=0.462 | weights: c-elegans=0.069, flywire=0.018, berghia=0.002, microns=0.161, oct-glia=0.005, oct-sfl=0.065, snemi=0.170, mouse-brain=0.510
+[04:45:41][   3350s][CONS] round 5: on the 72 new patches student Dice=0.607 vs ensemble 0.484 -> pseudo-labels from the student | consistency x0.36
+[04:46:04][   3374s][FT] round 5/6 epoch 1/5 | patches=362 | loss=0.8871
+[04:46:28][   3397s][FT] round 5/6 epoch 2/5 | patches=362 | loss=0.8923
+[04:46:51][   3421s][FT] round 5/6 epoch 3/5 | patches=362 | loss=0.8688
+[04:47:14][   3444s][FT] round 5/6 epoch 4/5 | patches=362 | loss=0.8893
+[04:47:38][   3467s][FT] round 5/6 epoch 5/5 | patches=362 | loss=0.8411
+[04:47:44][   3473s][AL] round 5/6 | labeled=360 | selection_seconds=42.1000 round_seconds=538.7000 pseudo_student_dice=0.6073 pseudo_ensemble_dice=0.4837 consistency_scale=0.3580 val_dice=0.6549 val_vi=0.5238 val_ece=0.0051
+[04:48:26][   3515s][AL] round 6: 1422/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[04:51:36][   3705s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.400, flywire=0.171, berghia=0.184, microns=0.326, oct-glia=0.207, oct-sfl=0.477, snemi=0.435, mouse-brain=0.478 | blended with labeled Dice at 0.50
+[04:54:35][   3885s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.480, uniform=0.220, top1=0.469, top2=0.481, top3=0.481 -> using mixture
+[04:54:35][   3885s][WEIGHTS] round 6: teacher Dice on 432 labeled patches + 2 blank (7377505 fg px): c-elegans=0.392, flywire=0.135, berghia=0.185, microns=0.255, oct-glia=0.129, oct-sfl=0.373, snemi=0.403, mouse-brain=0.469 | weights: c-elegans=0.074, flywire=0.017, berghia=0.002, microns=0.153, oct-glia=0.004, oct-sfl=0.064, snemi=0.173, mouse-brain=0.513
+[04:54:56][   3906s][CONS] round 6: on the 72 new patches student Dice=0.659 vs ensemble 0.462 -> pseudo-labels from the student | consistency x0.53
+[04:55:24][   3933s][FT] round 6/6 epoch 1/5 | patches=434 | loss=0.8836
+[04:55:51][   3961s][FT] round 6/6 epoch 2/5 | patches=434 | loss=0.8524
+[04:56:19][   3988s][FT] round 6/6 epoch 3/5 | patches=434 | loss=0.8450
+[04:56:47][   4016s][FT] round 6/6 epoch 4/5 | patches=434 | loss=0.8465
+[04:57:14][   4044s][FT] round 6/6 epoch 5/5 | patches=434 | loss=0.8346
+[04:57:20][   4050s][AL] round 6/6 | labeled=432 | selection_seconds=42.2000 round_seconds=570.5000 pseudo_student_dice=0.6586 pseudo_ensemble_dice=0.4617 consistency_scale=0.5290 val_dice=0.6593 val_vi=0.5305 val_ece=0.0107
+[04:57:31][   4060s][CONS] final consolidation: pseudo-labels refreshed from the student
+[04:57:58][   4088s][FT] final 1/10 epoch 1/1 | patches=434 | loss=0.8176
+[04:58:26][   4116s][FT] final 2/10 epoch 1/1 | patches=434 | loss=0.8093
+[04:58:54][   4143s][FT] final 3/10 epoch 1/1 | patches=434 | loss=0.7822
+[04:59:22][   4171s][FT] final 4/10 epoch 1/1 | patches=434 | loss=0.7432
+[04:59:49][   4199s][FT] final 5/10 epoch 1/1 | patches=434 | loss=0.7414
+[05:00:17][   4227s][FT] final 6/10 epoch 1/1 | patches=434 | loss=0.6925
+[05:00:45][   4255s][FT] final 7/10 epoch 1/1 | patches=434 | loss=0.6934
+[05:01:12][   4282s][FT] final 8/10 epoch 1/1 | patches=434 | loss=0.6712
+[05:01:41][   4310s][FT] final 9/10 epoch 1/1 | patches=434 | loss=0.6439
+[05:02:08][   4338s][FT] final 10/10 epoch 1/1 | patches=434 | loss=0.6502
+[05:02:14][   4344s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=287.9000 val_dice=0.6657 val_vi=0.5277 val_ece=0.0145
+[05:02:14][   4344s][WEIGHTS] final: c-elegans=0.074, flywire=0.017, berghia=0.002, microns=0.153, oct-glia=0.004, oct-sfl=0.064, snemi=0.173, mouse-brain=0.513
+[05:04:29][   4479s][TEST] raw student (no EMA, no TTA): patch_dice=0.6755 patch_vi=0.4444
+[05:04:45][   4495s][TEST] final model without TTA: patch_dice=0.6755 patch_vi=0.4444
+[05:04:45][   4495s][RESOURCES] {"phases": {"setup": {"seconds": 1.57, "peak_cpu_rss_mb": 2160.6, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 85.57, "peak_cpu_rss_mb": 4184.1, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 1139.29, "peak_cpu_rss_mb": 5957.0, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4598.0, "kd_epochs": 5}, "active_learning": {"seconds": 3110.53, "peak_cpu_rss_mb": 5957.0, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4642.0, "rounds": 6}, "inference": {"seconds": 150.79, "peak_cpu_rss_mb": 5957.0, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4644.0, "patches": 2883, "images": 3, "patches_per_second": 19.1, "seconds_per_image": 50.263}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5957.0, "total_seconds": 4487.8}
+[05:04:45][   4495s][TEST] patch_dice=0.7316 patch_iou=0.6304 patch_precision=0.7851 patch_recall=0.7817 patch_vi=0.4133 image_dice=0.8190 image_iou=0.6967 image_precision=0.8642 image_recall=0.7816 image_vi=0.4620 cal_ece=0.0032 cal_nll=0.1825 cal_brier=0.0543 cal_mean_confidence=0.9228 cal_accuracy=0.9259 cal_predicted_prevalence=0.2157 cal_true_prevalence=0.2301 cal_mean_probability=0.2249
+[05:04:46][   4495s][DONE] dhanyasi-P14-mouse-cerebellum macs-v11-norestart-acq-entropy b5 s1 | test patch Dice=0.7316 VI=0.4133 | image Dice=0.8190 VI=0.4620 | 4495s

macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-entropy/budget-5/seed-1/result.json

@@ -0,0 +1,776 @@
+{
+  "target": "dhanyasi-P14-mouse-cerebellum",
+  "variant": "macs-v11-norestart-acq-entropy",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "membrane",
+    "target": "dhanyasi-P14-mouse-cerebellum",
+    "variant": "macs-v11-norestart-acq-entropy",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "dhanyasi-P14-mouse-cerebellum",
+      "--variants",
+      "macs-v11-norestart-acq-entropy",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.05083445087713844,
+    0.15580217281625375,
+    0.01024336472294703,
+    0.4172897252912645,
+    0.039642121317843636,
+    0.05395053407082265,
+    0.10715675218847985,
+    0.1650808787152502
+  ],
+  "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": [
+      0.05083445087713844,
+      0.15580217281625375,
+      0.01024336472294703,
+      0.4172897252912645,
+      0.039642121317843636,
+      0.05395053407082265,
+      0.10715675218847985,
+      0.1650808787152502
+    ],
+    "per_source": [
+      0.18317784495730796,
+      0.0008117673075475437,
+      0.07130128976808368,
+      0.2901325749523702,
+      0.19474134185881758,
+      0.1442779689307106,
+      0.11165161174087061,
+      0.003905600484291838
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.04944686886108997,
+      "mmd2_mixture": 0.06619593169520155,
+      "mmd2_uniform": 0.10385689512746193,
+      "mmd2_single_source": [
+        0.3515701146669005,
+        0.19141235024309128,
+        0.2296475179170791,
+        0.08478110169677128,
+        0.27262177079818134,
+        0.24630850341824578,
+        0.1500490344627728,
+        0.17311817557273296
+      ],
+      "median_squared_distance": 0.18230580278935338,
+      "gammas": [
+        1.3713222298736392,
+        2.7426444597472783,
+        5.485288919494557,
+        10.970577838989113,
+        21.941155677978227
+      ],
+      "max_weight": 0.4172897252912645,
+      "entropy": 1.6749062834111577,
+      "effective_sources": 5.338294838847327,
+      "min_weight": 0.01024336472294703
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.24983786823863818,
+        0.6076264103713149,
+        0.31213514946535237,
+        0.2194743346098349,
+        0.24579617653258468,
+        0.265599036130135,
+        0.28252505597774746,
+        0.503904484908046
+      ],
+      "standardized": [
+        -0.6514551732920085,
+        2.0580443558260613,
+        -0.17968362593338755,
+        -0.8813954044106882,
+        -0.6820625306688456,
+        -0.5320973069785436,
+        -0.4039181245665406,
+        1.2725678100239537
+      ],
+      "tau": 0.5,
+      "max_weight": 0.2901325749523702,
+      "entropy": 1.7283714073603158,
+      "effective_sources": 5.63147505732349,
+      "min_weight": 0.0008117673075475437
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "microns_basil",
+    "seconds": 6.6
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.4128695009061776,
+      0.10650556878992745,
+      0.15770951130321542,
+      0.22267897752458055,
+      0.11007038183068829,
+      0.2906751954472746,
+      0.2970256007290182,
+      0.44753146510338837
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.20976392808829658,
+      "mixture@raw": 0.2201044151609466,
+      "mixture@logit": 0.18899978060037625,
+      "per_source": 0.193177417318676,
+      "per_source@raw": 0.20883301715723235,
+      "per_source@logit": 0.15551528847303686,
+      "uniform": 0.19238477463512638,
+      "uniform@raw": 0.2103392831498283,
+      "uniform@logit": 0.15367871932302002
+    },
+    "val_patches": 1024
+  },
+  "temperatures": [
+    2.575598955154419,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    2.642308473587036,
+    3.547258138656616,
+    4.074584007263184
+  ],
+  "top_source": "microns_basil",
+  "weight_stats": {
+    "max_weight": 0.4172897252912645,
+    "entropy": 1.6749062834111577,
+    "effective_sources": 5.338294838847327,
+    "min_weight": 0.01024336472294703
+  },
+  "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": 1139.3,
+  "round0_val": {
+    "val_dice": 0.12873302341577902,
+    "val_vi": 0.5095556457765907,
+    "val_ece": 0.1409334247270805
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 72,
+      "selection_seconds": 124.6,
+      "round_seconds": 859.5,
+      "teacher_dice_labeled": [
+        0.423321230947482,
+        0.15747034038174015,
+        0.1937080422460344,
+        0.30535734387879976,
+        0.16796457367489764,
+        0.4355867030005887,
+        0.4106488934816602,
+        0.480569706780208
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1139017,
+      "evidence_nonempty_patches": 61,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.24208465449533073,
+      "restart_teacher_dice": 0.480569706780208,
+      "weights": [
+        0.07916149016146976,
+        0.016996268628103182,
+        0.0016054657774156285,
+        0.19974819501005328,
+        0.004803002789714615,
+        0.09497730470154105,
+        0.14700770716957254,
+        0.4557005657621298
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.547725247586474,
+        "uniform": 0.2700913480165708,
+        "top1": 0.480569706780208,
+        "top2": 0.4919855915340812,
+        "top3": 0.4986229536904216
+      },
+      "val_dice": 0.6203974319004413,
+      "val_vi": 0.6482077865842528,
+      "val_ece": 0.04038987229168493
+    },
+    {
+      "round": 2,
+      "n_labeled": 144,
+      "selection_seconds": 42.1,
+      "round_seconds": 142.0,
+      "teacher_dice_labeled": [
+        0.3121220380549426,
+        0.11784411213032195,
+        0.16213463696136268,
+        0.21973314935568444,
+        0.10574009869144996,
+        0.3155480158339058,
+        0.3603575734043396,
+        0.3678067279431099
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1961159,
+      "evidence_nonempty_patches": 104,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.10788468470323365,
+        0.3354200554780413,
+        0.5566952598187249
+      ],
+      "ensemble_choice": "top3",
+      "candidate_dice": {
+        "mixture": 0.3879878504056142,
+        "uniform": 0.18391130588268922,
+        "top1": 0.3678067279431099,
+        "top2": 0.39948449736175295,
+        "top3": 0.4246588769388564
+      },
+      "pseudo_student_dice": 0.506870089555582,
+      "pseudo_ensemble_dice": 0.28715103968797784,
+      "pseudo_source": "student",
+      "consistency_scale": 0.023,
+      "val_dice": 0.5836245726576526,
+      "val_vi": 0.6030459611865171,
+      "val_ece": 0.05115153999860124
+    },
+    {
+      "round": 3,
+      "n_labeled": 216,
+      "selection_seconds": 42.1,
+      "round_seconds": 464.8,
+      "teacher_dice_labeled": [
+        0.3678614412663967,
+        0.13648598384761942,
+        0.15898055108815634,
+        0.24919594486654167,
+        0.13671349199434196,
+        0.3634301167820999,
+        0.37207817186759584,
+        0.4654090843012749
+      ],
+      "teacher_dice_pool": [
+        0.38011472038533395,
+        0.18782671866421538,
+        0.1762019359249727,
+        0.335320945586393,
+        0.2141001779587749,
+        0.4985947426940896,
+        0.43097701284546114,
+        0.4516571637763781
+      ],
+      "evidence_fg_pixels": 3836186,
+      "evidence_nonempty_patches": 172,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.08465750023050535,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.18614007060340373,
+        0.729202429166091
+      ],
+      "ensemble_choice": "top3",
+      "candidate_dice": {
+        "mixture": 0.4550915721689207,
+        "uniform": 0.21818578097968497,
+        "top1": 0.4654090843012749,
+        "top2": 0.474151057114913,
+        "top3": 0.4729733104091749
+      },
+      "pseudo_student_dice": 0.5842168055752709,
+      "pseudo_ensemble_dice": 0.6534647165896713,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.6467763653471634,
+      "val_vi": 0.558056933432851,
+      "val_ece": 0.05167226492881565
+    },
+    {
+      "round": 4,
+      "n_labeled": 288,
+      "selection_seconds": 42.1,
+      "round_seconds": 204.8,
+      "teacher_dice_labeled": [
+        0.35821635711104094,
+        0.09778125362066968,
+        0.17137258333636496,
+        0.1724433457499402,
+        0.05513871750205011,
+        0.2586880798396885,
+        0.3382993663944764,
+        0.4506664101801657
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 4894951,
+      "evidence_nonempty_patches": 240,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.09163584339119596,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.1582808577862576,
+        0.7500832988225464
+      ],
+      "ensemble_choice": "top3",
+      "candidate_dice": {
+        "mixture": 0.43443073868488324,
+        "uniform": 0.1334549874746405,
+        "top1": 0.4506664101801657,
+        "top2": 0.4506745268961577,
+        "top3": 0.45005077329606585
+      },
+      "pseudo_student_dice": 0.6440377318350516,
+      "pseudo_ensemble_dice": 0.3687653529910971,
+      "pseudo_source": "student",
+      "consistency_scale": 0.48,
+      "val_dice": 0.6458938731498896,
+      "val_vi": 0.5368060920046469,
+      "val_ece": 0.024876896241767484
+    },
+    {
+      "round": 5,
+      "n_labeled": 360,
+      "selection_seconds": 42.1,
+      "round_seconds": 538.7,
+      "teacher_dice_labeled": [
+        0.38013299125106925,
+        0.1355929442454303,
+        0.17720215081782453,
+        0.2539448023511426,
+        0.13333927029771672,
+        0.3683974998441432,
+        0.39580615773577776,
+        0.462093229389536
+      ],
+      "teacher_dice_pool": [
+        0.38762516986152434,
+        0.16847478642657696,
+        0.17322085467147122,
+        0.3200939826971463,
+        0.211538626877524,
+        0.4602015716310891,
+        0.4350030382843811,
+        0.4713753858291866
+      ],
+      "evidence_fg_pixels": 6110421,
+      "evidence_nonempty_patches": 310,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.06914616887588221,
+        0.018372132368569834,
+        0.0018311974724859205,
+        0.16070128006889814,
+        0.004570412095666965,
+        0.06525880886321075,
+        0.17048933831294702,
+        0.5096306619423392
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.4731747180910907,
+        "uniform": 0.22201451964945562,
+        "top1": 0.462093229389536,
+        "top2": 0.47332982939518164,
+        "top3": 0.47240656144828863
+      },
+      "pseudo_student_dice": 0.6073316378424809,
+      "pseudo_ensemble_dice": 0.483669413330306,
+      "pseudo_source": "student",
+      "consistency_scale": 0.358,
+      "val_dice": 0.6548823077127274,
+      "val_vi": 0.5238367298145632,
+      "val_ece": 0.005050470148968892
+    },
+    {
+      "round": 6,
+      "n_labeled": 432,
+      "selection_seconds": 42.2,
+      "round_seconds": 570.5,
+      "teacher_dice_labeled": [
+        0.3924474982027606,
+        0.13520125388940998,
+        0.18469172221338315,
+        0.25488349411404726,
+        0.1294684251406094,
+        0.3729135753452115,
+        0.40291120307243233,
+        0.46865716029759097
+      ],
+      "teacher_dice_pool": [
+        0.3999496986695009,
+        0.17062220618934187,
+        0.18388714703370632,
+        0.325641113187999,
+        0.20701207561780072,
+        0.47749391841243344,
+        0.43473254729499794,
+        0.4778808675544609
+      ],
+      "evidence_fg_pixels": 7377505,
+      "evidence_nonempty_patches": 375,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.07373789940615798,
+        0.017254401160920353,
+        0.001860815823363887,
+        0.15294575848580463,
+        0.004145585803265005,
+        0.06437149923884866,
+        0.17258198504016561,
+        0.513102055041474
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.479814058220684,
+        "uniform": 0.2197636441812369,
+        "top1": 0.46865716029759097,
+        "top2": 0.48056416983213823,
+        "top3": 0.48105135534617427
+      },
+      "pseudo_student_dice": 0.6586200443659856,
+      "pseudo_ensemble_dice": 0.4617141635573474,
+      "pseudo_source": "student",
+      "consistency_scale": 0.529,
+      "val_dice": 0.6593283705355009,
+      "val_vi": 0.5304568455956489,
+      "val_ece": 0.01068904340930946
+    },
+    {
+      "round": 7,
+      "n_labeled": 432,
+      "selection_seconds": 0.0,
+      "round_seconds": 287.9,
+      "final_consolidation": true,
+      "val_dice": 0.6656920404478153,
+      "val_vi": 0.5277453584741619,
+      "val_ece": 0.014463817565412058
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": -0.5377973366328569,
+    "auroc_top20_error": 0.224696484375
+  },
+  "al_seconds": 3110.5,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.07373789940615798,
+    0.017254401160920353,
+    0.001860815823363887,
+    0.15294575848580463,
+    0.004145585803265005,
+    0.06437149923884866,
+    0.17258198504016561,
+    0.513102055041474
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.513102055041474,
+    "entropy": 1.4060874658688964,
+    "effective_sources": 4.079961148192515,
+    "min_weight": 0.001860815823363887
+  },
+  "top_source_final": "whole-mouse-brain",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.675490319683645,
+    "patch_iou": 0.568467878690645,
+    "patch_precision": 0.7172071866334969,
+    "patch_recall": 0.7691740397507477,
+    "patch_vi": 0.44435762694673336,
+    "n_patches": 2883,
+    "cal_ece": 0.011123024071693167,
+    "cal_nll": 0.20394394839630497,
+    "cal_brier": 0.060213044436333046,
+    "cal_mean_confidence": 0.9296774405866081,
+    "cal_accuracy": 0.918554416514915,
+    "cal_predicted_prevalence": 0.21542230315643426,
+    "cal_true_prevalence": 0.23013442215140478,
+    "cal_mean_probability": 0.22161547074418342
+  },
+  "test_notta": {
+    "patch_dice": 0.675490319683645,
+    "patch_iou": 0.568467878690645,
+    "patch_precision": 0.7172071866334969,
+    "patch_recall": 0.7691740397507477,
+    "patch_vi": 0.44435762694673336,
+    "n_patches": 2883,
+    "cal_ece": 0.011123024071693167,
+    "cal_nll": 0.20394394839630497,
+    "cal_brier": 0.060213044436333046,
+    "cal_mean_confidence": 0.9296774405866081,
+    "cal_accuracy": 0.918554416514915,
+    "cal_predicted_prevalence": 0.21542230315643426,
+    "cal_true_prevalence": 0.23013442215140478,
+    "cal_mean_probability": 0.22161547074418342
+  },
+  "test": {
+    "patch_dice": 0.7316377151657691,
+    "patch_iou": 0.6303840026333201,
+    "patch_precision": 0.7850571730891608,
+    "patch_recall": 0.7817328455606284,
+    "patch_vi": 0.4133368944747202,
+    "n_patches": 2883,
+    "image_dice": 0.8189727425328098,
+    "image_iou": 0.6966907438405338,
+    "image_precision": 0.8641969001002137,
+    "image_recall": 0.7815801609355528,
+    "image_vi": 0.4620376655571004,
+    "n_images": 3,
+    "cal_ece": 0.0031798602784341846,
+    "cal_nll": 0.18247887747829966,
+    "cal_brier": 0.05429650188685861,
+    "cal_mean_confidence": 0.9227941310674863,
+    "cal_accuracy": 0.9259042094823101,
+    "cal_predicted_prevalence": 0.21568583615157821,
+    "cal_true_prevalence": 0.23013442215140478,
+    "cal_mean_probability": 0.22487676152217542
+  },
+  "eval_seconds": 150.8,
+  "total_seconds": 4494.7,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 1.57,
+        "peak_cpu_rss_mb": 2160.6,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 85.57,
+        "peak_cpu_rss_mb": 4184.1,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 1139.29,
+        "peak_cpu_rss_mb": 5957.0,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4598.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 3110.53,
+        "peak_cpu_rss_mb": 5957.0,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4642.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 150.79,
+        "peak_cpu_rss_mb": 5957.0,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4644.0,
+        "patches": 2883,
+        "images": 3,
+        "patches_per_second": 19.1,
+        "seconds_per_image": 50.263
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5957.0,
+    "total_seconds": 4487.8
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/dhanyasi-P14-mouse-cerebellum/macs-v11-norestart-acq-entropy_b5_s1.pth"
+}

macs_v2_r26/runs/membrane/dhanyasi-P14-mouse-cerebellum/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
+1,72,124.6,859.5,"[0.423321230947482, 0.15747034038174015, 0.1937080422460344, 0.30535734387879976, 0.16796457367489764, 0.4355867030005887, 0.4106488934816602, 0.480569706780208]",,1139017,61,True,kept,0.24208465449533073,0.480569706780208,"[0.07916149016146976, 0.016996268628103182, 0.0016054657774156285, 0.19974819501005328, 0.004803002789714615, 0.09497730470154105, 0.14700770716957254, 0.4557005657621298]",mixture,"{'mixture': 0.547725247586474, 'uniform': 0.2700913480165708, 'top1': 0.480569706780208, 'top2': 0.4919855915340812, 'top3': 0.4986229536904216}",0.6203974319004413,0.6482077865842528,0.04038987229168493,,,,,
+2,144,42.1,142.0,"[0.3121220380549426, 0.11784411213032195, 0.16213463696136268, 0.21973314935568444, 0.10574009869144996, 0.3155480158339058, 0.3603575734043396, 0.3678067279431099]",,1961159,104,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.10788468470323365, 0.3354200554780413, 0.5566952598187249]",top3,"{'mixture': 0.3879878504056142, 'uniform': 0.18391130588268922, 'top1': 0.3678067279431099, 'top2': 0.39948449736175295, 'top3': 0.4246588769388564}",0.5836245726576526,0.6030459611865171,0.05115153999860124,0.506870089555582,0.28715103968797784,student,0.023,
+3,216,42.1,464.8,"[0.3678614412663967, 0.13648598384761942, 0.15898055108815634, 0.24919594486654167, 0.13671349199434196, 0.3634301167820999, 0.37207817186759584, 0.4654090843012749]","[0.38011472038533395, 0.18782671866421538, 0.1762019359249727, 0.335320945586393, 0.2141001779587749, 0.4985947426940896, 0.43097701284546114, 0.4516571637763781]",3836186,172,False,,,,"[0.08465750023050535, 0.0, 0.0, 0.0, 0.0, 0.0, 0.18614007060340373, 0.729202429166091]",top3,"{'mixture': 0.4550915721689207, 'uniform': 0.21818578097968497, 'top1': 0.4654090843012749, 'top2': 0.474151057114913, 'top3': 0.4729733104091749}",0.6467763653471634,0.558056933432851,0.05167226492881565,0.5842168055752709,0.6534647165896713,ensemble,1.0,
+4,288,42.1,204.8,"[0.35821635711104094, 0.09778125362066968, 0.17137258333636496, 0.1724433457499402, 0.05513871750205011, 0.2586880798396885, 0.3382993663944764, 0.4506664101801657]",,4894951,240,False,,,,"[0.09163584339119596, 0.0, 0.0, 0.0, 0.0, 0.0, 0.1582808577862576, 0.7500832988225464]",top3,"{'mixture': 0.43443073868488324, 'uniform': 0.1334549874746405, 'top1': 0.4506664101801657, 'top2': 0.4506745268961577, 'top3': 0.45005077329606585}",0.6458938731498896,0.5368060920046469,0.024876896241767484,0.6440377318350516,0.3687653529910971,student,0.48,
+5,360,42.1,538.7,"[0.38013299125106925, 0.1355929442454303, 0.17720215081782453, 0.2539448023511426, 0.13333927029771672, 0.3683974998441432, 0.39580615773577776, 0.462093229389536]","[0.38762516986152434, 0.16847478642657696, 0.17322085467147122, 0.3200939826971463, 0.211538626877524, 0.4602015716310891, 0.4350030382843811, 0.4713753858291866]",6110421,310,False,,,,"[0.06914616887588221, 0.018372132368569834, 0.0018311974724859205, 0.16070128006889814, 0.004570412095666965, 0.06525880886321075, 0.17048933831294702, 0.5096306619423392]",mixture,"{'mixture': 0.4731747180910907, 'uniform': 0.22201451964945562, 'top1': 0.462093229389536, 'top2': 0.47332982939518164, 'top3': 0.47240656144828863}",0.6548823077127274,0.5238367298145632,0.005050470148968892,0.6073316378424809,0.483669413330306,student,0.358,
+6,432,42.2,570.5,"[0.3924474982027606, 0.13520125388940998, 0.18469172221338315, 0.25488349411404726, 0.1294684251406094, 0.3729135753452115, 0.40291120307243233, 0.46865716029759097]","[0.3999496986695009, 0.17062220618934187, 0.18388714703370632, 0.325641113187999, 0.20701207561780072, 0.47749391841243344, 0.43473254729499794, 0.4778808675544609]",7377505,375,False,,,,"[0.07373789940615798, 0.017254401160920353, 0.001860815823363887, 0.15294575848580463, 0.004145585803265005, 0.06437149923884866, 0.17258198504016561, 0.513102055041474]",mixture,"{'mixture': 0.479814058220684, 'uniform': 0.2197636441812369, 'top1': 0.46865716029759097, 'top2': 0.48056416983213823, 'top3': 0.48105135534617427}",0.6593283705355009,0.5304568455956489,0.01068904340930946,0.6586200443659856,0.4617141635573474,student,0.529,
+7,432,0.0,287.9,,,,,,,,,,,,0.6656920404478153,0.5277453584741619,0.014463817565412058,,,,,True

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-1/log.txt

@@ -0,0 +1,108 @@
+[05:04:49][      0s][START] target=octopus-vulgaris-vertical-lobe-sfltract variant=macs-v11-norestart-acq-entropy budget=5.0% seed=1 | {"task": "membrane", "target": "octopus-vulgaris-vertical-lobe-sfltract", "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": 8, "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}
+[05:04:49][      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.
+[05:04:49][      0s][DATA] octopus-vulgaris-vertical-lobe-sfltract: train=3584 val=768 test=768 patches
+[05:04:51][      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, snemi=3.55, mouse-brain=4.07
+[05:04:51][      3s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s
+[05:04:58][      9s][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
+[05:04:58][      9s][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
+[05:05:53][     65s][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
+[05:05:53][     65s][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
+[05:05:53][     65s][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
+[05:05:53][     65s][INIT] student initialised from teacher oct-glia (weight 0.636)
+[05:05:57][     68s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[05:05:57][     68s][KD] distilling the mixture ensemble (prob space) into the student
+[05:08:22][    214s][KD] soft labels for 3584 patches from 8 teachers in 146s
+[05:10:09][    320s][KD] epoch 1/5 | patches=3584 | loss=0.5720 | 106s
+[05:11:55][    427s][KD] epoch 2/5 | patches=3584 | loss=0.4388 | 106s
+[05:13:42][    533s][KD] epoch 3/5 | patches=3584 | loss=0.4006 | 106s
+[05:15:28][    640s][KD] epoch 4/5 | patches=3584 | loss=0.3569 | 107s
+[05:17:15][    746s][KD] epoch 5/5 | patches=3584 | loss=0.3310 | 106s
+[05:17:19][    751s][ROUND0] before any target label: val_dice=0.8502 val_vi=0.1470
+[05:17:19][    751s][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
+[05:17:32][    763s][AL] round-1 uncertainty quality on 2338 patches: {'spearman_uncertainty_error': 0.44560086267226684, 'auroc_top20_error': 0.593673156908451}
+[05:18:54][    845s][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
+[05:19:03][    854s][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
+[05:19:04][    855s][WEIGHTS] round 1: student kept (labeled Dice 0.750 vs best teacher oct-glia 0.731); distilling the corrected ensemble into it
+[05:19:04][    855s][WEIGHTS] round 1: teacher Dice on 29 labeled patches + 29 blank (237130 fg px): c-elegans=0.296, cerebellum=0.619, 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
+[05:20:38][    949s][KD] soft labels for 2309 patches from 8 teachers in 94s
+[05:21:47][   1018s][KD] epoch 1/2 | patches=2309 | loss=0.7212 | 69s
+[05:22:55][   1087s][KD] epoch 2/2 | patches=2309 | loss=0.6145 | 69s
+[05:24:17][   1169s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 82s
+[05:24:22][   1173s][FT] round 1/6 epoch 1/5 | patches=58 | loss=1.4306
+[05:24:26][   1177s][FT] round 1/6 epoch 2/5 | patches=58 | loss=1.3428
+[05:24:30][   1182s][FT] round 1/6 epoch 3/5 | patches=58 | loss=1.3158
+[05:24:35][   1186s][FT] round 1/6 epoch 4/5 | patches=58 | loss=1.2565
+[05:24:39][   1190s][FT] round 1/6 epoch 5/5 | patches=58 | loss=1.3613
+[05:24:44][   1195s][AL] round 1/6 | labeled=29 | selection_seconds=94.3000 round_seconds=439.9000 restart_student_dice=0.7496 restart_teacher_dice=0.7308 val_dice=0.3973 val_vi=0.1807 val_ece=0.0784
+[05:24:56][   1207s][AL] round 2: 518/2309 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[05:25:14][   1225s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.588, uniform=0.611, top1=0.588, top2=0.587, top3=0.587 -> using uniform
+[05:25:14][   1225s][WEIGHTS] round 2: teacher Dice on 58 labeled patches + 58 blank (646819 fg px): c-elegans=0.348, cerebellum=0.462, flywire=0.176, berghia=0.317, microns=0.481, oct-glia=0.588, snemi=0.179, mouse-brain=0.323 | 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
+[05:25:30][   1242s][CONS] round 2: on the 29 new patches student Dice=0.611 vs ensemble 0.523 -> pseudo-labels from the student | consistency x0.37
+[05:25:38][   1249s][FT] round 2/6 epoch 1/5 | patches=116 | loss=1.0340
+[05:25:46][   1257s][FT] round 2/6 epoch 2/5 | patches=116 | loss=1.0324
+[05:25:54][   1265s][FT] round 2/6 epoch 3/5 | patches=116 | loss=0.9938
+[05:26:01][   1273s][FT] round 2/6 epoch 4/5 | patches=116 | loss=0.9853
+[05:26:09][   1280s][FT] round 2/6 epoch 5/5 | patches=116 | loss=0.9211
+[05:26:14][   1285s][AL] round 2/6 | labeled=58 | selection_seconds=12.5000 round_seconds=85.6000 pseudo_student_dice=0.6108 pseudo_ensemble_dice=0.5226 consistency_scale=0.3690 val_dice=0.6369 val_vi=0.1633 val_ece=0.0094
+[05:26:26][   1297s][AL] round 3: 566/2280 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[05:29:13][   1465s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.441, cerebellum=0.666, flywire=0.237, berghia=0.399, microns=0.581, oct-glia=0.806, snemi=0.334, mouse-brain=0.310 | blended with labeled Dice at 0.50
+[05:31:51][   1622s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.692, uniform=0.709, top1=0.692, top2=0.692, top3=0.691 -> using uniform
+[05:31:51][   1622s][WEIGHTS] round 3: teacher Dice on 87 labeled patches + 87 blank (1020182 fg px): c-elegans=0.388, cerebellum=0.565, flywire=0.205, berghia=0.354, microns=0.525, oct-glia=0.692, snemi=0.257, mouse-brain=0.318 | 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
+[05:32:07][   1639s][CONS] round 3: on the 29 new patches student Dice=0.685 vs ensemble 0.565 -> pseudo-labels from the student | consistency x0.62
+[05:32:18][   1650s][FT] round 3/6 epoch 1/5 | patches=174 | loss=0.9333
+[05:32:29][   1661s][FT] round 3/6 epoch 2/5 | patches=174 | loss=0.9096
+[05:32:41][   1672s][FT] round 3/6 epoch 3/5 | patches=174 | loss=0.8885
+[05:32:52][   1683s][FT] round 3/6 epoch 4/5 | patches=174 | loss=0.8886
+[05:33:03][   1694s][FT] round 3/6 epoch 5/5 | patches=174 | loss=0.8893
+[05:33:07][   1698s][AL] round 3/6 | labeled=87 | selection_seconds=12.3000 round_seconds=409.1000 pseudo_student_dice=0.6845 pseudo_ensemble_dice=0.5650 consistency_scale=0.6150 val_dice=0.8477 val_vi=0.1608 val_ece=0.0074
+[05:33:19][   1710s][AL] round 4: 568/2251 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[05:36:11][   1882s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.447, cerebellum=0.668, flywire=0.234, berghia=0.398, microns=0.567, oct-glia=0.791, snemi=0.341, mouse-brain=0.320 | blended with labeled Dice at 0.50
+[05:38:54][   2045s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.699, uniform=0.708, top1=0.698, top2=0.699, top3=0.698 -> using uniform
+[05:38:54][   2045s][WEIGHTS] round 4: teacher Dice on 116 labeled patches + 116 blank (1253998 fg px): c-elegans=0.367, cerebellum=0.569, flywire=0.206, berghia=0.348, microns=0.519, oct-glia=0.698, snemi=0.259, mouse-brain=0.300 | 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
+[05:39:10][   2061s][CONS] round 4: on the 29 new patches student Dice=0.717 vs ensemble 0.649 -> pseudo-labels from the student | consistency x0.72
+[05:39:24][   2076s][FT] round 4/6 epoch 1/5 | patches=232 | loss=0.8830
+[05:39:39][   2090s][FT] round 4/6 epoch 2/5 | patches=232 | loss=0.8655
+[05:39:53][   2104s][FT] round 4/6 epoch 3/5 | patches=232 | loss=0.8936
+[05:40:08][   2119s][FT] round 4/6 epoch 4/5 | patches=232 | loss=0.8549
+[05:40:22][   2133s][FT] round 4/6 epoch 5/5 | patches=232 | loss=0.7787
+[05:40:26][   2138s][AL] round 4/6 | labeled=116 | selection_seconds=12.2000 round_seconds=434.8000 pseudo_student_dice=0.7174 pseudo_ensemble_dice=0.6485 consistency_scale=0.7250 val_dice=0.8907 val_vi=0.1518 val_ece=0.0106
+[05:40:38][   2149s][AL] round 5: 572/2222 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[05:43:34][   2326s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.428, cerebellum=0.669, flywire=0.232, berghia=0.383, microns=0.566, oct-glia=0.815, snemi=0.326, mouse-brain=0.303 | blended with labeled Dice at 0.50
+[05:46:21][   2492s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.698, uniform=0.702, top1=0.698, top2=0.699, top3=0.698 -> using mixture
+[05:46:21][   2492s][WEIGHTS] round 5: teacher Dice on 145 labeled patches + 145 blank (1654235 fg px): c-elegans=0.367, cerebellum=0.566, flywire=0.204, berghia=0.340, microns=0.509, oct-glia=0.698, snemi=0.254, mouse-brain=0.298 | weights: c-elegans=0.005, cerebellum=0.012, flywire=0.000, berghia=0.006, microns=0.007, oct-glia=0.968, snemi=0.000, mouse-brain=0.001
+[05:46:37][   2508s][CONS] round 5: on the 29 new patches student Dice=0.783 vs ensemble 0.496 -> pseudo-labels from the student | consistency x0.94
+[05:46:55][   2526s][FT] round 5/6 epoch 1/5 | patches=290 | loss=0.7943
+[05:47:13][   2544s][FT] round 5/6 epoch 2/5 | patches=290 | loss=0.7149
+[05:47:31][   2562s][FT] round 5/6 epoch 3/5 | patches=290 | loss=0.6657
+[05:47:49][   2580s][FT] round 5/6 epoch 4/5 | patches=290 | loss=0.7386
+[05:48:07][   2598s][FT] round 5/6 epoch 5/5 | patches=290 | loss=0.5905
+[05:48:11][   2603s][AL] round 5/6 | labeled=145 | selection_seconds=12.0000 round_seconds=460.5000 pseudo_student_dice=0.7831 pseudo_ensemble_dice=0.4959 consistency_scale=0.9440 val_dice=0.8796 val_vi=0.1516 val_ece=0.0127
+[05:48:23][   2614s][AL] round 6: 561/2193 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[05:51:24][   2796s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.432, cerebellum=0.673, flywire=0.240, berghia=0.391, microns=0.593, oct-glia=0.838, snemi=0.311, mouse-brain=0.294 | blended with labeled Dice at 0.50
+[05:54:16][   2967s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.681, uniform=0.689, top1=0.680, top2=0.681, top3=0.680 -> using uniform
+[05:54:16][   2967s][WEIGHTS] round 6: teacher Dice on 174 labeled patches + 174 blank (2203858 fg px): c-elegans=0.356, cerebellum=0.551, flywire=0.203, berghia=0.333, microns=0.501, oct-glia=0.680, snemi=0.258, mouse-brain=0.289 | 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
+[05:54:32][   2983s][CONS] round 6: on the 29 new patches student Dice=0.586 vs ensemble 0.339 -> pseudo-labels from the student | consistency x0.29
+[05:54:53][   3004s][FT] round 6/6 epoch 1/5 | patches=348 | loss=0.6190
+[05:55:14][   3026s][FT] round 6/6 epoch 2/5 | patches=348 | loss=0.5244
+[05:55:36][   3047s][FT] round 6/6 epoch 3/5 | patches=348 | loss=0.6222
+[05:55:57][   3068s][FT] round 6/6 epoch 4/5 | patches=348 | loss=0.4741
+[05:56:18][   3090s][FT] round 6/6 epoch 5/5 | patches=348 | loss=0.5008
+[05:56:23][   3094s][AL] round 6/6 | labeled=174 | selection_seconds=11.9000 round_seconds=487.1000 pseudo_student_dice=0.5858 pseudo_ensemble_dice=0.3394 consistency_scale=0.2860 val_dice=0.8889 val_vi=0.1497 val_ece=0.0118
+[05:56:33][   3104s][CONS] final consolidation: pseudo-labels refreshed from the student
+[05:56:54][   3125s][FT] final 1/10 epoch 1/1 | patches=348 | loss=0.4606
+[05:57:15][   3146s][FT] final 2/10 epoch 1/1 | patches=348 | loss=0.4388
+[05:57:36][   3168s][FT] final 3/10 epoch 1/1 | patches=348 | loss=0.4371
+[05:57:58][   3189s][FT] final 4/10 epoch 1/1 | patches=348 | loss=0.4126
+[05:58:19][   3210s][FT] final 5/10 epoch 1/1 | patches=348 | loss=0.3581
+[05:58:40][   3231s][FT] final 6/10 epoch 1/1 | patches=348 | loss=0.3504
+[05:59:01][   3253s][FT] final 7/10 epoch 1/1 | patches=348 | loss=0.3270
+[05:59:23][   3274s][FT] final 8/10 epoch 1/1 | patches=348 | loss=0.3318
+[05:59:44][   3295s][FT] final 9/10 epoch 1/1 | patches=348 | loss=0.3389
+[06:00:05][   3317s][FT] final 10/10 epoch 1/1 | patches=348 | loss=0.3184
+[06:00:10][   3321s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=222.6000 val_dice=0.8926 val_vi=0.1435 val_ece=0.0113
+[06:00:10][   3321s][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
+[06:00:46][   3357s][TEST] raw student (no EMA, no TTA): patch_dice=0.8728 patch_vi=0.1771
+[06:00:50][   3362s][TEST] final model without TTA: patch_dice=0.8728 patch_vi=0.1771
+[06:00:50][   3362s][RESOURCES] {"phases": {"setup": {"seconds": 1.48, "peak_cpu_rss_mb": 2173.9, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 65.54, "peak_cpu_rss_mb": 4182.5, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 678.3, "peak_cpu_rss_mb": 5201.6, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4598.0, "kd_epochs": 5}, "active_learning": {"seconds": 2570.54, "peak_cpu_rss_mb": 5201.6, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4606.0, "rounds": 6}, "inference": {"seconds": 40.49, "peak_cpu_rss_mb": 5201.6, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4606.0, "patches": 768, "images": 6, "patches_per_second": 19.0, "seconds_per_image": 6.748}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5201.6, "total_seconds": 3356.3}
+[06:00:50][   3362s][TEST] patch_dice=0.8733 patch_iou=0.8121 patch_precision=0.8897 patch_recall=0.8868 patch_vi=0.1731 image_dice=0.8078 image_iou=0.6844 image_precision=0.7892 image_recall=0.8329 image_vi=0.1954 cal_ece=0.0053 cal_nll=0.0703 cal_brier=0.0211 cal_mean_confidence=0.9763 cal_accuracy=0.9711 cal_predicted_prevalence=0.0766 cal_true_prevalence=0.0718 cal_mean_probability=0.0757
+[06:00:51][   3362s][DONE] octopus-vulgaris-vertical-lobe-sfltract macs-v11-norestart-acq-entropy b5 s1 | test patch Dice=0.8733 VI=0.1731 | image Dice=0.8078 VI=0.1954 | 3362s

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy/budget-5/seed-1/result.json

@@ -0,0 +1,785 @@
+{
+  "target": "octopus-vulgaris-vertical-lobe-sfltract",
+  "variant": "macs-v11-norestart-acq-entropy",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "membrane",
+    "target": "octopus-vulgaris-vertical-lobe-sfltract",
+    "variant": "macs-v11-norestart-acq-entropy",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "octopus-vulgaris-vertical-lobe-sfltract",
+      "--variants",
+      "macs-v11-norestart-acq-entropy",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.09888716021456875,
+    0.02852231397957182,
+    0.006877159468562143,
+    0.147623932500446,
+    0.03162958721157525,
+    0.636412589592954,
+    0.019031438719961654,
+    0.031015818312360402
+  ],
+  "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": [
+      0.09888716021456875,
+      0.02852231397957182,
+      0.006877159468562143,
+      0.147623932500446,
+      0.03162958721157525,
+      0.636412589592954,
+      0.019031438719961654,
+      0.031015818312360402
+    ],
+    "per_source": [
+      0.3196471036584027,
+      0.016524109438155077,
+      0.0007715730533314212,
+      0.17539928949217498,
+      0.027481458110301667,
+      0.4363205819091585,
+      0.016643531248023022,
+      0.007212353090452595
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.00048686699692031786,
+      "mmd2_mixture": 0.012755679331756475,
+      "mmd2_uniform": 0.09698314484068699,
+      "mmd2_single_source": [
+        0.07244562949164945,
+        0.2568973175367495,
+        0.39290896557558747,
+        0.05287206442937192,
+        0.2649845830320105,
+        0.011934176639413208,
+        0.33925886828153656,
+        0.26844309711822195
+      ],
+      "median_squared_distance": 0.1834397705913593,
+      "gammas": [
+        1.3628451409095685,
+        2.725690281819137,
+        5.451380563638274,
+        10.902761127276548,
+        21.805522254553097
+      ],
+      "max_weight": 0.636412589592954,
+      "entropy": 1.2268812334836157,
+      "effective_sources": 3.4105761415276636,
+      "min_weight": 0.006877159468562143
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.049620160284728154,
+        0.42663804362685775,
+        0.8166050888075048,
+        0.12600028070223912,
+        0.36189803378480956,
+        0.01001954120360149,
+        0.42572157074448214,
+        0.5321462950254623
+      ],
+      "standardized": [
+        -1.1548909896059252,
+        0.32630755724364063,
+        1.8583797643241364,
+        -0.8548147110811022,
+        0.07196204012648774,
+        -1.3104708228948398,
+        0.3227069898393355,
+        0.740820172048267
+      ],
+      "tau": 0.5,
+      "max_weight": 0.4363205819091585,
+      "entropy": 1.3076009419864412,
+      "effective_sources": 3.6972930433455575,
+      "min_weight": 0.0007715730533314212
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+    "seconds": 6.6
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.47571341080584045,
+      0.6329400594773938,
+      0.1419064245794933,
+      0.4669233743357945,
+      0.7079863797747427,
+      0.8450584862058909,
+      0.22902171788078127,
+      0.11473171105409108
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.8488016622672679,
+      "mixture@raw": 0.8468775415451569,
+      "mixture@logit": 0.8498181877900208,
+      "per_source": 0.8430718482924652,
+      "per_source@raw": 0.6192917168334541,
+      "per_source@logit": 0.8479512124278182,
+      "uniform": 0.7183086201637726,
+      "uniform@raw": 0.6771618878909939,
+      "uniform@logit": 0.8065091143369424
+    },
+    "val_patches": 768
+  },
+  "temperatures": [
+    2.575598955154419,
+    4.557764530181885,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    3.547258138656616,
+    4.074584007263184
+  ],
+  "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+  "weight_stats": {
+    "max_weight": 0.636412589592954,
+    "entropy": 1.2268812334836157,
+    "effective_sources": 3.4105761415276636,
+    "min_weight": 0.006877159468562143
+  },
+  "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": 678.3,
+  "round0_val": {
+    "val_dice": 0.8502116365713305,
+    "val_vi": 0.1469760206899403,
+    "val_ece": 0.06606864095266592
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 29,
+      "selection_seconds": 94.3,
+      "round_seconds": 439.9,
+      "teacher_dice_labeled": [
+        0.29568020341690904,
+        0.6193671974528233,
+        0.19588368058977745,
+        0.33035580668548686,
+        0.5508702113809155,
+        0.7308332381787659,
+        0.15051414530012766,
+        0.23402934928630323
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 237130,
+      "evidence_nonempty_patches": 23,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.749598658200551,
+      "restart_teacher_dice": 0.7308332381787659,
+      "weights": [
+        0.125,
+        0.125,
+        0.125,
+        0.125,
+        0.125,
+        0.125,
+        0.125,
+        0.125
+      ],
+      "ensemble_choice": "uniform",
+      "candidate_dice": {
+        "mixture": 0.7308488562528247,
+        "uniform": 0.7388948429036649,
+        "top1": 0.7308332381787659,
+        "top2": 0.7311335062716425,
+        "top3": 0.7306909902224555
+      },
+      "val_dice": 0.39731069928710405,
+      "val_vi": 0.18067060453166736,
+      "val_ece": 0.07838129414619284
+    },
+    {
+      "round": 2,
+      "n_labeled": 58,
+      "selection_seconds": 12.5,
+      "round_seconds": 85.6,
+      "teacher_dice_labeled": [
+        0.3476127955077974,
+        0.4623686045382735,
+        0.1755232866300641,
+        0.31664754608108453,
+        0.48104076607485885,
+        0.5880462888045375,
+        0.1790241268093611,
+        0.32257295870994507
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 646819,
+      "evidence_nonempty_patches": 51,
+      "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": 0.5883446508492198,
+        "uniform": 0.6113743713600025,
+        "top1": 0.5880462888045375,
+        "top2": 0.5871022090819282,
+        "top3": 0.587255051506554
+      },
+      "pseudo_student_dice": 0.6107994294456149,
+      "pseudo_ensemble_dice": 0.5225904590692428,
+      "pseudo_source": "student",
+      "consistency_scale": 0.369,
+      "val_dice": 0.6369395682370388,
+      "val_vi": 0.16326645853342284,
+      "val_ece": 0.009399857270229506
+    },
+    {
+      "round": 3,
+      "n_labeled": 87,
+      "selection_seconds": 12.3,
+      "round_seconds": 409.1,
+      "teacher_dice_labeled": [
+        0.3880088072732191,
+        0.565341164182688,
+        0.20471024332465787,
+        0.35380388386182804,
+        0.5253267179901164,
+        0.6915026222402474,
+        0.2565538563065365,
+        0.3179727238418305
+      ],
+      "teacher_dice_pool": [
+        0.4412735330757333,
+        0.6659325718988698,
+        0.2372073842200266,
+        0.39943171418340423,
+        0.5811530177876038,
+        0.8057331885912369,
+        0.33433901329823956,
+        0.3096649357020063
+      ],
+      "evidence_fg_pixels": 1020182,
+      "evidence_nonempty_patches": 80,
+      "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": 0.6919442120407975,
+        "uniform": 0.7091946157908361,
+        "top1": 0.6915026222402474,
+        "top2": 0.6921985279311229,
+        "top3": 0.6912061210138836
+      },
+      "pseudo_student_dice": 0.6845415260994039,
+      "pseudo_ensemble_dice": 0.5650033092900414,
+      "pseudo_source": "student",
+      "consistency_scale": 0.615,
+      "val_dice": 0.8476713037914542,
+      "val_vi": 0.1607710772373184,
+      "val_ece": 0.007436648340219965
+    },
+    {
+      "round": 4,
+      "n_labeled": 116,
+      "selection_seconds": 12.2,
+      "round_seconds": 434.8,
+      "teacher_dice_labeled": [
+        0.3669914562867691,
+        0.5693699554122069,
+        0.2055175935989612,
+        0.347743730912508,
+        0.5187721250757013,
+        0.6984750534228014,
+        0.2589420340565617,
+        0.3004582926329548
+      ],
+      "teacher_dice_pool": [
+        0.44670708753866184,
+        0.667642860577924,
+        0.23361163069008736,
+        0.39822489431136593,
+        0.5667288786894037,
+        0.7912791062085153,
+        0.3405437241505429,
+        0.32008342376085336
+      ],
+      "evidence_fg_pixels": 1253998,
+      "evidence_nonempty_patches": 108,
+      "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": 0.6988568524984016,
+        "uniform": 0.7081510561372487,
+        "top1": 0.6984750534228014,
+        "top2": 0.6991990657433986,
+        "top3": 0.6983183514463073
+      },
+      "pseudo_student_dice": 0.717398226943492,
+      "pseudo_ensemble_dice": 0.6485005341745833,
+      "pseudo_source": "student",
+      "consistency_scale": 0.725,
+      "val_dice": 0.8906511127825961,
+      "val_vi": 0.15183013964667516,
+      "val_ece": 0.010564493907424609
+    },
+    {
+      "round": 5,
+      "n_labeled": 145,
+      "selection_seconds": 12.0,
+      "round_seconds": 460.5,
+      "teacher_dice_labeled": [
+        0.3668099150383525,
+        0.5657863433890937,
+        0.20371422005742845,
+        0.3401057868287816,
+        0.508806826757005,
+        0.6980405494858268,
+        0.254115069915782,
+        0.2978415150971249
+      ],
+      "teacher_dice_pool": [
+        0.4282138051931642,
+        0.668509289478364,
+        0.23248299174518863,
+        0.38337992924291425,
+        0.5661442227675124,
+        0.8152971118400434,
+        0.32553445749354537,
+        0.3033784254312041
+      ],
+      "evidence_fg_pixels": 1654235,
+      "evidence_nonempty_patches": 137,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.005480660578475023,
+        0.011561692285882852,
+        7.460818807273828e-05,
+        0.006264338487527958,
+        0.007252225134167767,
+        0.9681622203186412,
+        0.0003417725105441578,
+        0.0008624824966881743
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.6984906391263901,
+        "uniform": 0.7020901946833366,
+        "top1": 0.6980405494858268,
+        "top2": 0.6987377441865222,
+        "top3": 0.6979569026901176
+      },
+      "pseudo_student_dice": 0.7831485684638488,
+      "pseudo_ensemble_dice": 0.4959069721765298,
+      "pseudo_source": "student",
+      "consistency_scale": 0.944,
+      "val_dice": 0.8796156527618478,
+      "val_vi": 0.15158004082917956,
+      "val_ece": 0.012681466394150536
+    },
+    {
+      "round": 6,
+      "n_labeled": 174,
+      "selection_seconds": 11.9,
+      "round_seconds": 487.1,
+      "teacher_dice_labeled": [
+        0.356145961605246,
+        0.551214355755117,
+        0.2025672126275297,
+        0.3327589734690903,
+        0.5010580750877105,
+        0.6802621501411376,
+        0.25820683333981315,
+        0.28948928888955194
+      ],
+      "teacher_dice_pool": [
+        0.431692128592248,
+        0.6726023749138484,
+        0.24020601597533023,
+        0.3911778628654421,
+        0.5933084714965258,
+        0.8380355437195518,
+        0.31126908803516595,
+        0.2935730368159814
+      ],
+      "evidence_fg_pixels": 2203858,
+      "evidence_nonempty_patches": 166,
+      "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": 0.6807795554829543,
+        "uniform": 0.68870702244862,
+        "top1": 0.6802621501411376,
+        "top2": 0.6810272289642794,
+        "top3": 0.6801802090041148
+      },
+      "pseudo_student_dice": 0.5857522693305887,
+      "pseudo_ensemble_dice": 0.33939174177213893,
+      "pseudo_source": "student",
+      "consistency_scale": 0.286,
+      "val_dice": 0.8888998968722844,
+      "val_vi": 0.14966503109444784,
+      "val_ece": 0.011809566908036034
+    },
+    {
+      "round": 7,
+      "n_labeled": 174,
+      "selection_seconds": 0.0,
+      "round_seconds": 222.6,
+      "final_consolidation": true,
+      "val_dice": 0.8925997424783859,
+      "val_vi": 0.1434862563966719,
+      "val_ece": 0.01128403642853145
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.44560086267226684,
+    "auroc_top20_error": 0.593673156908451
+  },
+  "al_seconds": 2570.5,
+  "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": 0.8727808053306628,
+    "patch_iou": 0.8096397365891068,
+    "patch_precision": 0.884013050033685,
+    "patch_recall": 0.8888230542795013,
+    "patch_vi": 0.1771062671777847,
+    "n_patches": 768,
+    "cal_ece": 0.006959328183459772,
+    "cal_nll": 0.07563690589200811,
+    "cal_brier": 0.02189482180963423,
+    "cal_mean_confidence": 0.9771665425389284,
+    "cal_accuracy": 0.9702072143554688,
+    "cal_predicted_prevalence": 0.07643890380859375,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.07524883666029399
+  },
+  "test_notta": {
+    "patch_dice": 0.8727808053306628,
+    "patch_iou": 0.8096397365891068,
+    "patch_precision": 0.884013050033685,
+    "patch_recall": 0.8888230542795013,
+    "patch_vi": 0.1771062671777847,
+    "n_patches": 768,
+    "cal_ece": 0.006959328183459772,
+    "cal_nll": 0.07563690589200811,
+    "cal_brier": 0.02189482180963423,
+    "cal_mean_confidence": 0.9771665425389284,
+    "cal_accuracy": 0.9702072143554688,
+    "cal_predicted_prevalence": 0.07643890380859375,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.07524883666029399
+  },
+  "test": {
+    "patch_dice": 0.8733466141713727,
+    "patch_iou": 0.8120811888493558,
+    "patch_precision": 0.889693337909526,
+    "patch_recall": 0.8867975267013571,
+    "patch_vi": 0.17312031856958507,
+    "n_patches": 768,
+    "image_dice": 0.8077534707124677,
+    "image_iou": 0.684351592405526,
+    "image_precision": 0.7892406397670163,
+    "image_recall": 0.8329104380944133,
+    "image_vi": 0.19538874245769558,
+    "n_images": 6,
+    "cal_ece": 0.005300650660982887,
+    "cal_nll": 0.0702677819857817,
+    "cal_brier": 0.02107877091438942,
+    "cal_mean_confidence": 0.9763216560626121,
+    "cal_accuracy": 0.9711151123046875,
+    "cal_predicted_prevalence": 0.07657623291015625,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.0756733806638027
+  },
+  "eval_seconds": 40.5,
+  "total_seconds": 3361.6,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 1.48,
+        "peak_cpu_rss_mb": 2173.9,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 65.54,
+        "peak_cpu_rss_mb": 4182.5,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 678.3,
+        "peak_cpu_rss_mb": 5201.6,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4598.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2570.54,
+        "peak_cpu_rss_mb": 5201.6,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4606.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 40.49,
+        "peak_cpu_rss_mb": 5201.6,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4606.0,
+        "patches": 768,
+        "images": 6,
+        "patches_per_second": 19.0,
+        "seconds_per_image": 6.748
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5201.6,
+    "total_seconds": 3356.3
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-entropy_b5_s1.pth"
+}

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/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
+1,29,94.3,439.9,"[0.29568020341690904, 0.6193671974528233, 0.19588368058977745, 0.33035580668548686, 0.5508702113809155, 0.7308332381787659, 0.15051414530012766, 0.23402934928630323]",,237130,23,True,kept,0.749598658200551,0.7308332381787659,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.7308488562528247, 'uniform': 0.7388948429036649, 'top1': 0.7308332381787659, 'top2': 0.7311335062716425, 'top3': 0.7306909902224555}",0.39731069928710405,0.18067060453166736,0.07838129414619284,,,,,
+2,58,12.5,85.6,"[0.3476127955077974, 0.4623686045382735, 0.1755232866300641, 0.31664754608108453, 0.48104076607485885, 0.5880462888045375, 0.1790241268093611, 0.32257295870994507]",,646819,51,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.5883446508492198, 'uniform': 0.6113743713600025, 'top1': 0.5880462888045375, 'top2': 0.5871022090819282, 'top3': 0.587255051506554}",0.6369395682370388,0.16326645853342284,0.009399857270229506,0.6107994294456149,0.5225904590692428,student,0.369,
+3,87,12.3,409.1,"[0.3880088072732191, 0.565341164182688, 0.20471024332465787, 0.35380388386182804, 0.5253267179901164, 0.6915026222402474, 0.2565538563065365, 0.3179727238418305]","[0.4412735330757333, 0.6659325718988698, 0.2372073842200266, 0.39943171418340423, 0.5811530177876038, 0.8057331885912369, 0.33433901329823956, 0.3096649357020063]",1020182,80,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6919442120407975, 'uniform': 0.7091946157908361, 'top1': 0.6915026222402474, 'top2': 0.6921985279311229, 'top3': 0.6912061210138836}",0.8476713037914542,0.1607710772373184,0.007436648340219965,0.6845415260994039,0.5650033092900414,student,0.615,
+4,116,12.2,434.8,"[0.3669914562867691, 0.5693699554122069, 0.2055175935989612, 0.347743730912508, 0.5187721250757013, 0.6984750534228014, 0.2589420340565617, 0.3004582926329548]","[0.44670708753866184, 0.667642860577924, 0.23361163069008736, 0.39822489431136593, 0.5667288786894037, 0.7912791062085153, 0.3405437241505429, 0.32008342376085336]",1253998,108,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6988568524984016, 'uniform': 0.7081510561372487, 'top1': 0.6984750534228014, 'top2': 0.6991990657433986, 'top3': 0.6983183514463073}",0.8906511127825961,0.15183013964667516,0.010564493907424609,0.717398226943492,0.6485005341745833,student,0.725,
+5,145,12.0,460.5,"[0.3668099150383525, 0.5657863433890937, 0.20371422005742845, 0.3401057868287816, 0.508806826757005, 0.6980405494858268, 0.254115069915782, 0.2978415150971249]","[0.4282138051931642, 0.668509289478364, 0.23248299174518863, 0.38337992924291425, 0.5661442227675124, 0.8152971118400434, 0.32553445749354537, 0.3033784254312041]",1654235,137,False,,,,"[0.005480660578475023, 0.011561692285882852, 7.460818807273828e-05, 0.006264338487527958, 0.007252225134167767, 0.9681622203186412, 0.0003417725105441578, 0.0008624824966881743]",mixture,"{'mixture': 0.6984906391263901, 'uniform': 0.7020901946833366, 'top1': 0.6980405494858268, 'top2': 0.6987377441865222, 'top3': 0.6979569026901176}",0.8796156527618478,0.15158004082917956,0.012681466394150536,0.7831485684638488,0.4959069721765298,student,0.944,
+6,174,11.9,487.1,"[0.356145961605246, 0.551214355755117, 0.2025672126275297, 0.3327589734690903, 0.5010580750877105, 0.6802621501411376, 0.25820683333981315, 0.28948928888955194]","[0.431692128592248, 0.6726023749138484, 0.24020601597533023, 0.3911778628654421, 0.5933084714965258, 0.8380355437195518, 0.31126908803516595, 0.2935730368159814]",2203858,166,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6807795554829543, 'uniform': 0.68870702244862, 'top1': 0.6802621501411376, 'top2': 0.6810272289642794, 'top3': 0.6801802090041148}",0.8888998968722844,0.14966503109444784,0.011809566908036034,0.5857522693305887,0.33939174177213893,student,0.286,
+7,174,0.0,222.6,,,,,,,,,,,,0.8925997424783859,0.1434862563966719,0.01128403642853145,,,,,True

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-1/log.txt

@@ -0,0 +1,108 @@
+[00:56:03][      0s][START] target=octopus-vulgaris-vertical-lobe-sfltract variant=macs-v11-norestart-acq-laplace budget=5.0% seed=1 | {"task": "membrane", "target": "octopus-vulgaris-vertical-lobe-sfltract", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 8, "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}
+[00:56:03][      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.
+[00:56:04][      1s][DATA] octopus-vulgaris-vertical-lobe-sfltract: train=3584 val=768 test=768 patches
+[00:56:22][     19s][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=4.07
+[00:56:56][     53s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 34s
+[00:57:02][     59s][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
+[00:57:02][     59s][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
+[00:57:57][    114s][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
+[00:57:57][    114s][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
+[00:57:57][    114s][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
+[00:57:57][    114s][INIT] student initialised from teacher oct-glia (weight 0.636)
+[00:58:01][    117s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[00:58:01][    117s][KD] distilling the mixture ensemble (prob space) into the student
+[01:00:26][    263s][KD] soft labels for 3584 patches from 8 teachers in 145s
+[01:02:13][    369s][KD] epoch 1/5 | patches=3584 | loss=0.5717 | 106s
+[01:03:59][    476s][KD] epoch 2/5 | patches=3584 | loss=0.4387 | 106s
+[01:05:46][    582s][KD] epoch 3/5 | patches=3584 | loss=0.4024 | 106s
+[01:07:32][    689s][KD] epoch 4/5 | patches=3584 | loss=0.3575 | 107s
+[01:09:19][    796s][KD] epoch 5/5 | patches=3584 | loss=0.3323 | 107s
+[01:09:23][    800s][ROUND0] before any target label: val_dice=0.4068 val_vi=0.1562
+[01:09:23][    800s][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
+[01:09:49][    825s][AL] round-1 uncertainty quality on 2338 patches: {'spearman_uncertainty_error': -0.42973054159052243, 'auroc_top20_error': 0.01460704062277854}
+[01:11:11][    907s][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
+[01:11:20][    917s][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
+[01:11:21][    917s][WEIGHTS] round 1: student kept (labeled Dice 0.724 vs best teacher oct-glia 0.731); distilling the corrected ensemble into it
+[01:11:21][    917s][WEIGHTS] round 1: teacher Dice on 29 labeled patches + 29 blank (237130 fg px): c-elegans=0.296, cerebellum=0.619, 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
+[01:12:55][   1012s][KD] soft labels for 2309 patches from 8 teachers in 94s
+[01:14:04][   1080s][KD] epoch 1/2 | patches=2309 | loss=0.7251 | 69s
+[01:15:13][   1149s][KD] epoch 2/2 | patches=2309 | loss=0.6221 | 69s
+[01:16:34][   1231s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 82s
+[01:16:39][   1235s][FT] round 1/6 epoch 1/5 | patches=58 | loss=1.3947
+[01:16:43][   1240s][FT] round 1/6 epoch 2/5 | patches=58 | loss=1.3601
+[01:16:47][   1244s][FT] round 1/6 epoch 3/5 | patches=58 | loss=1.3657
+[01:16:52][   1248s][FT] round 1/6 epoch 4/5 | patches=58 | loss=1.2920
+[01:16:56][   1253s][FT] round 1/6 epoch 5/5 | patches=58 | loss=1.3077
+[01:17:01][   1257s][AL] round 1/6 | labeled=29 | selection_seconds=107.5000 round_seconds=452.9000 restart_student_dice=0.7237 restart_teacher_dice=0.7308 val_dice=0.8265 val_vi=0.1609 val_ece=0.0489
+[01:17:26][   1282s][AL] round 2: 571/2309 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[01:17:44][   1300s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.601, uniform=0.616, top1=0.601, top2=0.600, top3=0.600 -> using uniform
+[01:17:44][   1300s][WEIGHTS] round 2: teacher Dice on 58 labeled patches + 58 blank (672023 fg px): c-elegans=0.317, cerebellum=0.466, flywire=0.176, berghia=0.291, microns=0.517, oct-glia=0.601, snemi=0.186, mouse-brain=0.309 | 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
+[01:18:00][   1316s][CONS] round 2: on the 29 new patches student Dice=0.614 vs ensemble 0.545 -> pseudo-labels from the student | consistency x0.38
+[01:18:07][   1324s][FT] round 2/6 epoch 1/5 | patches=116 | loss=1.0646
+[01:18:15][   1332s][FT] round 2/6 epoch 2/5 | patches=116 | loss=0.9956
+[01:18:23][   1339s][FT] round 2/6 epoch 3/5 | patches=116 | loss=0.9742
+[01:18:30][   1347s][FT] round 2/6 epoch 4/5 | patches=116 | loss=0.9444
+[01:18:38][   1354s][FT] round 2/6 epoch 5/5 | patches=116 | loss=0.9440
+[01:18:42][   1359s][AL] round 2/6 | labeled=58 | selection_seconds=25.1000 round_seconds=97.3000 pseudo_student_dice=0.6139 pseudo_ensemble_dice=0.5450 consistency_scale=0.3800 val_dice=0.8588 val_vi=0.1514 val_ece=0.0125
+[01:19:07][   1384s][AL] round 3: 602/2280 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[01:21:55][   1551s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.371, cerebellum=0.616, flywire=0.239, berghia=0.365, microns=0.592, oct-glia=0.873, snemi=0.275, mouse-brain=0.247 | blended with labeled Dice at 0.50
+[01:24:33][   1709s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.650, uniform=0.638, top1=0.649, top2=0.650, top3=0.649 -> using mixture
+[01:24:33][   1709s][WEIGHTS] round 3: teacher Dice on 87 labeled patches + 87 blank (1597523 fg px): c-elegans=0.352, cerebellum=0.508, flywire=0.197, berghia=0.306, microns=0.487, oct-glia=0.649, snemi=0.272, mouse-brain=0.308 | weights: c-elegans=0.008, cerebellum=0.011, flywire=0.000, berghia=0.007, microns=0.009, oct-glia=0.963, snemi=0.001, mouse-brain=0.002
+[01:24:49][   1725s][CONS] round 3: on the 29 new patches student Dice=0.383 vs ensemble 0.260 -> pseudo-labels from the student | consistency x0.00
+[01:24:54][   1731s][FT] round 3/6 epoch 1/5 | patches=174 | loss=1.1783
+[01:25:00][   1736s][FT] round 3/6 epoch 2/5 | patches=174 | loss=1.0471
+[01:25:05][   1742s][FT] round 3/6 epoch 3/5 | patches=174 | loss=1.0678
+[01:25:11][   1748s][FT] round 3/6 epoch 4/5 | patches=174 | loss=1.0251
+[01:25:17][   1753s][FT] round 3/6 epoch 5/5 | patches=174 | loss=0.9424
+[01:25:21][   1758s][AL] round 3/6 | labeled=87 | selection_seconds=24.9000 round_seconds=394.2000 pseudo_student_dice=0.3832 pseudo_ensemble_dice=0.2603 consistency_scale=0.0000 val_dice=0.8525 val_vi=0.1750 val_ece=0.0105
+[01:25:45][   1782s][AL] round 4: 537/2251 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[01:28:38][   1954s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.443, cerebellum=0.666, flywire=0.236, berghia=0.392, microns=0.582, oct-glia=0.788, snemi=0.338, mouse-brain=0.314 | blended with labeled Dice at 0.50
+[01:31:20][   2117s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.631, uniform=0.664, top1=0.630, top2=0.631, top3=0.630 -> using uniform
+[01:31:20][   2117s][WEIGHTS] round 4: teacher Dice on 116 labeled patches + 116 blank (1870770 fg px): c-elegans=0.372, cerebellum=0.550, flywire=0.199, berghia=0.328, microns=0.494, oct-glia=0.630, snemi=0.295, mouse-brain=0.318 | 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
+[01:31:36][   2133s][CONS] round 4: on the 29 new patches student Dice=0.716 vs ensemble 0.677 -> pseudo-labels from the student | consistency x0.72
+[01:31:51][   2147s][FT] round 4/6 epoch 1/5 | patches=232 | loss=1.0101
+[01:32:05][   2162s][FT] round 4/6 epoch 2/5 | patches=232 | loss=0.9749
+[01:32:19][   2176s][FT] round 4/6 epoch 3/5 | patches=232 | loss=0.9456
+[01:32:34][   2190s][FT] round 4/6 epoch 4/5 | patches=232 | loss=0.9118
+[01:32:48][   2204s][FT] round 4/6 epoch 5/5 | patches=232 | loss=0.9304
+[01:32:52][   2209s][AL] round 4/6 | labeled=116 | selection_seconds=24.6000 round_seconds=447.0000 pseudo_student_dice=0.7162 pseudo_ensemble_dice=0.6767 consistency_scale=0.7210 val_dice=0.8798 val_vi=0.1542 val_ece=0.0075
+[01:33:16][   2233s][AL] round 5: 569/2222 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[01:36:13][   2409s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.415, cerebellum=0.661, flywire=0.244, berghia=0.384, microns=0.603, oct-glia=0.851, snemi=0.310, mouse-brain=0.282 | blended with labeled Dice at 0.50
+[01:38:59][   2576s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.671, uniform=0.681, top1=0.670, top2=0.671, top3=0.670 -> using uniform
+[01:38:59][   2576s][WEIGHTS] round 5: teacher Dice on 145 labeled patches + 145 blank (2186789 fg px): c-elegans=0.366, cerebellum=0.549, flywire=0.204, berghia=0.329, microns=0.509, oct-glia=0.670, snemi=0.271, 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
+[01:39:15][   2592s][CONS] round 5: on the 29 new patches student Dice=0.726 vs ensemble 0.597 -> pseudo-labels from the student | consistency x0.75
+[01:39:33][   2609s][FT] round 5/6 epoch 1/5 | patches=290 | loss=0.9539
+[01:39:51][   2627s][FT] round 5/6 epoch 2/5 | patches=290 | loss=0.9196
+[01:40:09][   2645s][FT] round 5/6 epoch 3/5 | patches=290 | loss=0.9494
+[01:40:27][   2663s][FT] round 5/6 epoch 4/5 | patches=290 | loss=0.8943
+[01:40:45][   2681s][FT] round 5/6 epoch 5/5 | patches=290 | loss=0.8669
+[01:40:49][   2686s][AL] round 5/6 | labeled=145 | selection_seconds=24.3000 round_seconds=472.3000 pseudo_student_dice=0.7263 pseudo_ensemble_dice=0.5966 consistency_scale=0.7540 val_dice=0.8752 val_vi=0.1569 val_ece=0.0086
+[01:41:13][   2709s][AL] round 6: 553/2193 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[01:44:14][   2891s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.434, cerebellum=0.668, flywire=0.240, berghia=0.395, microns=0.591, oct-glia=0.824, snemi=0.320, mouse-brain=0.299 | blended with labeled Dice at 0.50
+[01:47:06][   3063s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.665, uniform=0.683, top1=0.664, top2=0.665, top3=0.664 -> using uniform
+[01:47:06][   3063s][WEIGHTS] round 6: teacher Dice on 174 labeled patches + 174 blank (2478728 fg px): c-elegans=0.368, cerebellum=0.555, flywire=0.202, berghia=0.336, microns=0.505, oct-glia=0.664, snemi=0.273, 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
+[01:47:22][   3078s][CONS] round 6: on the 29 new patches student Dice=0.793 vs ensemble 0.596 -> pseudo-labels from the student | consistency x0.98
+[01:47:43][   3100s][FT] round 6/6 epoch 1/5 | patches=348 | loss=0.8826
+[01:48:04][   3121s][FT] round 6/6 epoch 2/5 | patches=348 | loss=0.9147
+[01:48:26][   3142s][FT] round 6/6 epoch 3/5 | patches=348 | loss=0.8845
+[01:48:47][   3164s][FT] round 6/6 epoch 4/5 | patches=348 | loss=0.8723
+[01:49:08][   3185s][FT] round 6/6 epoch 5/5 | patches=348 | loss=0.8485
+[01:49:13][   3189s][AL] round 6/6 | labeled=174 | selection_seconds=24.0000 round_seconds=499.4000 pseudo_student_dice=0.7929 pseudo_ensemble_dice=0.5957 consistency_scale=0.9760 val_dice=0.8927 val_vi=0.1440 val_ece=0.0078
+[01:49:23][   3199s][CONS] final consolidation: pseudo-labels refreshed from the student
+[01:49:44][   3220s][FT] final 1/10 epoch 1/1 | patches=348 | loss=0.8445
+[01:50:05][   3242s][FT] final 2/10 epoch 1/1 | patches=348 | loss=0.8160
+[01:50:26][   3263s][FT] final 3/10 epoch 1/1 | patches=348 | loss=0.7961
+[01:50:48][   3284s][FT] final 4/10 epoch 1/1 | patches=348 | loss=0.7927
+[01:51:09][   3305s][FT] final 5/10 epoch 1/1 | patches=348 | loss=0.7625
+[01:51:30][   3327s][FT] final 6/10 epoch 1/1 | patches=348 | loss=0.7530
+[01:51:51][   3348s][FT] final 7/10 epoch 1/1 | patches=348 | loss=0.7819
+[01:52:13][   3369s][FT] final 8/10 epoch 1/1 | patches=348 | loss=0.7434
+[01:52:34][   3391s][FT] final 9/10 epoch 1/1 | patches=348 | loss=0.7514
+[01:52:55][   3412s][FT] final 10/10 epoch 1/1 | patches=348 | loss=0.7475
+[01:53:00][   3416s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=222.6000 val_dice=0.8887 val_vi=0.1467 val_ece=0.0078
+[01:53:00][   3416s][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
+[01:53:36][   3453s][TEST] raw student (no EMA, no TTA): patch_dice=0.8645 patch_vi=0.1770
+[01:53:40][   3457s][TEST] final model without TTA: patch_dice=0.8645 patch_vi=0.1770
+[01:53:40][   3457s][RESOURCES] {"phases": {"setup": {"seconds": 16.85, "peak_cpu_rss_mb": 2160.3, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 98.94, "peak_cpu_rss_mb": 4183.7, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 678.09, "peak_cpu_rss_mb": 5202.8, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4598.0, "kd_epochs": 5}, "active_learning": {"seconds": 2616.44, "peak_cpu_rss_mb": 5202.8, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4606.0, "rounds": 6}, "inference": {"seconds": 40.52, "peak_cpu_rss_mb": 5202.8, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4606.0, "patches": 768, "images": 6, "patches_per_second": 19.0, "seconds_per_image": 6.753}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5202.8, "total_seconds": 3450.8}
+[01:53:40][   3457s][TEST] patch_dice=0.8750 patch_iou=0.8143 patch_precision=0.8867 patch_recall=0.8961 patch_vi=0.1732 image_dice=0.8095 image_iou=0.6867 image_precision=0.7810 image_recall=0.8463 image_vi=0.1962 cal_ece=0.0100 cal_nll=0.0766 cal_brier=0.0215 cal_mean_confidence=0.9717 cal_accuracy=0.9708 cal_predicted_prevalence=0.0786 cal_true_prevalence=0.0718 cal_mean_probability=0.0855
+[01:53:41][   3457s][DONE] octopus-vulgaris-vertical-lobe-sfltract macs-v11-norestart-acq-laplace b5 s1 | test patch Dice=0.8750 VI=0.1732 | image Dice=0.8095 VI=0.1962 | 3457s

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/budget-5/seed-1/result.json

@@ -0,0 +1,785 @@
+{
+  "target": "octopus-vulgaris-vertical-lobe-sfltract",
+  "variant": "macs-v11-norestart-acq-laplace",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "membrane",
+    "target": "octopus-vulgaris-vertical-lobe-sfltract",
+    "variant": "macs-v11-norestart-acq-laplace",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "octopus-vulgaris-vertical-lobe-sfltract",
+      "--variants",
+      "macs-v11-norestart-acq-laplace",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.09888716021456875,
+    0.02852231397957182,
+    0.006877159468562143,
+    0.147623932500446,
+    0.03162958721157525,
+    0.636412589592954,
+    0.019031438719961654,
+    0.031015818312360402
+  ],
+  "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": [
+      0.09888716021456875,
+      0.02852231397957182,
+      0.006877159468562143,
+      0.147623932500446,
+      0.03162958721157525,
+      0.636412589592954,
+      0.019031438719961654,
+      0.031015818312360402
+    ],
+    "per_source": [
+      0.3196471036584027,
+      0.016524109438155077,
+      0.0007715730533314212,
+      0.17539928949217498,
+      0.027481458110301667,
+      0.4363205819091585,
+      0.016643531248023022,
+      0.007212353090452595
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.00048686699692031786,
+      "mmd2_mixture": 0.012755679331756475,
+      "mmd2_uniform": 0.09698314484068699,
+      "mmd2_single_source": [
+        0.07244562949164945,
+        0.2568973175367495,
+        0.39290896557558747,
+        0.05287206442937192,
+        0.2649845830320105,
+        0.011934176639413208,
+        0.33925886828153656,
+        0.26844309711822195
+      ],
+      "median_squared_distance": 0.1834397705913593,
+      "gammas": [
+        1.3628451409095685,
+        2.725690281819137,
+        5.451380563638274,
+        10.902761127276548,
+        21.805522254553097
+      ],
+      "max_weight": 0.636412589592954,
+      "entropy": 1.2268812334836157,
+      "effective_sources": 3.4105761415276636,
+      "min_weight": 0.006877159468562143
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.049620160284728154,
+        0.42663804362685775,
+        0.8166050888075048,
+        0.12600028070223912,
+        0.36189803378480956,
+        0.01001954120360149,
+        0.42572157074448214,
+        0.5321462950254623
+      ],
+      "standardized": [
+        -1.1548909896059252,
+        0.32630755724364063,
+        1.8583797643241364,
+        -0.8548147110811022,
+        0.07196204012648774,
+        -1.3104708228948398,
+        0.3227069898393355,
+        0.740820172048267
+      ],
+      "tau": 0.5,
+      "max_weight": 0.4363205819091585,
+      "entropy": 1.3076009419864412,
+      "effective_sources": 3.6972930433455575,
+      "min_weight": 0.0007715730533314212
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+    "seconds": 40.3
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.47571341080584045,
+      0.6329400594773938,
+      0.1419064245794933,
+      0.4669233743357945,
+      0.7079863797747427,
+      0.8450584862058909,
+      0.22902171788078127,
+      0.11473171105409108
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.8488016622672679,
+      "mixture@raw": 0.8468775415451569,
+      "mixture@logit": 0.8498181877900208,
+      "per_source": 0.8430718482924652,
+      "per_source@raw": 0.6192917168334541,
+      "per_source@logit": 0.8479512124278182,
+      "uniform": 0.7183086201637726,
+      "uniform@raw": 0.6771618878909939,
+      "uniform@logit": 0.8065091143369424
+    },
+    "val_patches": 768
+  },
+  "temperatures": [
+    2.575598955154419,
+    4.557764530181885,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    3.547258138656616,
+    4.074584007263184
+  ],
+  "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+  "weight_stats": {
+    "max_weight": 0.636412589592954,
+    "entropy": 1.2268812334836157,
+    "effective_sources": 3.4105761415276636,
+    "min_weight": 0.006877159468562143
+  },
+  "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": 678.1,
+  "round0_val": {
+    "val_dice": 0.4067974008452868,
+    "val_vi": 0.15621365356541697,
+    "val_ece": 0.06549482541855627
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 29,
+      "selection_seconds": 107.5,
+      "round_seconds": 452.9,
+      "teacher_dice_labeled": [
+        0.29568020341690904,
+        0.6193671974528233,
+        0.19588368058977745,
+        0.33035580668548686,
+        0.5508702113809155,
+        0.7308332381787659,
+        0.15051414530012766,
+        0.23402934928630323
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 237130,
+      "evidence_nonempty_patches": 23,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.7237450842918105,
+      "restart_teacher_dice": 0.7308332381787659,
+      "weights": [
+        0.125,
+        0.125,
+        0.125,
+        0.125,
+        0.125,
+        0.125,
+        0.125,
+        0.125
+      ],
+      "ensemble_choice": "uniform",
+      "candidate_dice": {
+        "mixture": 0.7308488562528247,
+        "uniform": 0.7388948429036649,
+        "top1": 0.7308332381787659,
+        "top2": 0.7311335062716425,
+        "top3": 0.7306909902224555
+      },
+      "val_dice": 0.8265384832710779,
+      "val_vi": 0.16094128778513914,
+      "val_ece": 0.04887358321577289
+    },
+    {
+      "round": 2,
+      "n_labeled": 58,
+      "selection_seconds": 25.1,
+      "round_seconds": 97.3,
+      "teacher_dice_labeled": [
+        0.3171883581634619,
+        0.46570797027165556,
+        0.17555021224032277,
+        0.2911379528704814,
+        0.5171020431871733,
+        0.6011370938587045,
+        0.1856036164323916,
+        0.30947977967970813
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 672023,
+      "evidence_nonempty_patches": 51,
+      "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": 0.6008390269849482,
+        "uniform": 0.6162660612905311,
+        "top1": 0.6011370938587045,
+        "top2": 0.6000244930568988,
+        "top3": 0.6001927615324153
+      },
+      "pseudo_student_dice": 0.613930989815908,
+      "pseudo_ensemble_dice": 0.5449893692757639,
+      "pseudo_source": "student",
+      "consistency_scale": 0.38,
+      "val_dice": 0.858755045097654,
+      "val_vi": 0.1513502459658489,
+      "val_ece": 0.012488309354615111
+    },
+    {
+      "round": 3,
+      "n_labeled": 87,
+      "selection_seconds": 24.9,
+      "round_seconds": 394.2,
+      "teacher_dice_labeled": [
+        0.3515194750513133,
+        0.5081483731457119,
+        0.1967517587863077,
+        0.30607281205994263,
+        0.4873790788992553,
+        0.6494763538876254,
+        0.27193648915742474,
+        0.30833822986758297
+      ],
+      "teacher_dice_pool": [
+        0.3713920260153615,
+        0.6156583110479326,
+        0.23892875624458382,
+        0.3647158945021577,
+        0.591553287347347,
+        0.8728288107309928,
+        0.2746848725997957,
+        0.24697999986747887
+      ],
+      "evidence_fg_pixels": 1597523,
+      "evidence_nonempty_patches": 80,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.007602838779626683,
+        0.010501498964006668,
+        0.00011248570107448317,
+        0.0072047729084134095,
+        0.009461503682782231,
+        0.962908280806494,
+        0.000660211341494219,
+        0.0015484078161083674
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.6498703208262191,
+        "uniform": 0.6380122833004724,
+        "top1": 0.6494763538876254,
+        "top2": 0.6498371076706292,
+        "top3": 0.6492208400303191
+      },
+      "pseudo_student_dice": 0.38323891424672846,
+      "pseudo_ensemble_dice": 0.2603137417028411,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.852483095049041,
+      "val_vi": 0.17500004822089713,
+      "val_ece": 0.010542839366299465
+    },
+    {
+      "round": 4,
+      "n_labeled": 116,
+      "selection_seconds": 24.6,
+      "round_seconds": 447.0,
+      "teacher_dice_labeled": [
+        0.37165118582176443,
+        0.5498739512200651,
+        0.19894122455555435,
+        0.32796989413566147,
+        0.49362131291149014,
+        0.6299155752418457,
+        0.29462911835367733,
+        0.31849510251907787
+      ],
+      "teacher_dice_pool": [
+        0.4432668421464532,
+        0.6655030182527801,
+        0.23613268382521105,
+        0.3915591066403852,
+        0.5820384495809867,
+        0.7883949429794211,
+        0.3380038795229039,
+        0.3135432956883544
+      ],
+      "evidence_fg_pixels": 1870770,
+      "evidence_nonempty_patches": 109,
+      "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": 0.6310020695428453,
+        "uniform": 0.6641508814931827,
+        "top1": 0.6299155752418457,
+        "top2": 0.6310825091972322,
+        "top3": 0.6297899101236573
+      },
+      "pseudo_student_dice": 0.7162450297722375,
+      "pseudo_ensemble_dice": 0.6766694366299267,
+      "pseudo_source": "student",
+      "consistency_scale": 0.721,
+      "val_dice": 0.8797632053364045,
+      "val_vi": 0.15423802631351424,
+      "val_ece": 0.007480832876547508
+    },
+    {
+      "round": 5,
+      "n_labeled": 145,
+      "selection_seconds": 24.3,
+      "round_seconds": 472.3,
+      "teacher_dice_labeled": [
+        0.3660794262070859,
+        0.5490828422499876,
+        0.20365563891426552,
+        0.32856908333908263,
+        0.508535981970067,
+        0.670062722012975,
+        0.2709322094932049,
+        0.3038809822055101
+      ],
+      "teacher_dice_pool": [
+        0.41543039333650833,
+        0.6614358260242903,
+        0.24373609609710786,
+        0.384228244423019,
+        0.603001507114574,
+        0.8509662250083222,
+        0.309811235614449,
+        0.2822458495441181
+      ],
+      "evidence_fg_pixels": 2186789,
+      "evidence_nonempty_patches": 138,
+      "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": 0.6707006667666381,
+        "uniform": 0.6809966041879847,
+        "top1": 0.670062722012975,
+        "top2": 0.6708633202970876,
+        "top3": 0.6700126434291589
+      },
+      "pseudo_student_dice": 0.7263052830088944,
+      "pseudo_ensemble_dice": 0.5965506068759152,
+      "pseudo_source": "student",
+      "consistency_scale": 0.754,
+      "val_dice": 0.875244862502354,
+      "val_vi": 0.15694088885158228,
+      "val_ece": 0.008585482330209854
+    },
+    {
+      "round": 6,
+      "n_labeled": 174,
+      "selection_seconds": 24.0,
+      "round_seconds": 499.4,
+      "teacher_dice_labeled": [
+        0.3681232181861549,
+        0.5551674270712343,
+        0.2021990584599616,
+        0.33618850765162767,
+        0.5048303375240749,
+        0.6641611574975141,
+        0.2725573674876931,
+        0.30352943057597503
+      ],
+      "teacher_dice_pool": [
+        0.43428106838072467,
+        0.6680803435489528,
+        0.23994257936146782,
+        0.39529555441652675,
+        0.5914807803652934,
+        0.8240333760711983,
+        0.3198122503561808,
+        0.29933052019728335
+      ],
+      "evidence_fg_pixels": 2478728,
+      "evidence_nonempty_patches": 165,
+      "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": 0.664855616294795,
+        "uniform": 0.6832583651582114,
+        "top1": 0.6641611574975141,
+        "top2": 0.6650440389558555,
+        "top3": 0.6640915000195716
+      },
+      "pseudo_student_dice": 0.7929019284967859,
+      "pseudo_ensemble_dice": 0.5956894105964962,
+      "pseudo_source": "student",
+      "consistency_scale": 0.976,
+      "val_dice": 0.8926858816531681,
+      "val_vi": 0.1439952713975571,
+      "val_ece": 0.007831732653372411
+    },
+    {
+      "round": 7,
+      "n_labeled": 174,
+      "selection_seconds": 0.0,
+      "round_seconds": 222.6,
+      "final_consolidation": true,
+      "val_dice": 0.8886851466384921,
+      "val_vi": 0.1466770562917574,
+      "val_ece": 0.007776746073254894
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": -0.42973054159052243,
+    "auroc_top20_error": 0.01460704062277854
+  },
+  "al_seconds": 2616.4,
+  "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": 0.8644854064227285,
+    "patch_iou": 0.8015775339461436,
+    "patch_precision": 0.8736117259970388,
+    "patch_recall": 0.8942403794858601,
+    "patch_vi": 0.1769803664520788,
+    "n_patches": 768,
+    "cal_ece": 0.01005696782408574,
+    "cal_nll": 0.08044871377833289,
+    "cal_brier": 0.022497044457770853,
+    "cal_mean_confidence": 0.9738861268454938,
+    "cal_accuracy": 0.9699223836263021,
+    "cal_predicted_prevalence": 0.07838185628255208,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.084024411452732
+  },
+  "test_notta": {
+    "patch_dice": 0.8644854064227285,
+    "patch_iou": 0.8015775339461436,
+    "patch_precision": 0.8736117259970388,
+    "patch_recall": 0.8942403794858601,
+    "patch_vi": 0.1769803664520788,
+    "n_patches": 768,
+    "cal_ece": 0.01005696782408574,
+    "cal_nll": 0.08044871377833289,
+    "cal_brier": 0.022497044457770853,
+    "cal_mean_confidence": 0.9738861268454938,
+    "cal_accuracy": 0.9699223836263021,
+    "cal_predicted_prevalence": 0.07838185628255208,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.084024411452732
+  },
+  "test": {
+    "patch_dice": 0.8749669073071713,
+    "patch_iou": 0.8143105444897812,
+    "patch_precision": 0.8867455102594728,
+    "patch_recall": 0.8960972601232834,
+    "patch_vi": 0.17321693095203786,
+    "n_patches": 768,
+    "image_dice": 0.8094618761648809,
+    "image_iou": 0.6866971832851606,
+    "image_precision": 0.781044872770262,
+    "image_recall": 0.8462603726171656,
+    "image_vi": 0.19624323121594725,
+    "n_images": 6,
+    "cal_ece": 0.010036823548981465,
+    "cal_nll": 0.07660645530730685,
+    "cal_brier": 0.021545440108361295,
+    "cal_mean_confidence": 0.9716628545615914,
+    "cal_accuracy": 0.9708366394042969,
+    "cal_predicted_prevalence": 0.07863744099934895,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.08545583541733916
+  },
+  "eval_seconds": 40.5,
+  "total_seconds": 3456.9,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 16.85,
+        "peak_cpu_rss_mb": 2160.3,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 98.94,
+        "peak_cpu_rss_mb": 4183.7,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 678.09,
+        "peak_cpu_rss_mb": 5202.8,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4598.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2616.44,
+        "peak_cpu_rss_mb": 5202.8,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4606.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 40.52,
+        "peak_cpu_rss_mb": 5202.8,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4606.0,
+        "patches": 768,
+        "images": 6,
+        "patches_per_second": 19.0,
+        "seconds_per_image": 6.753
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5202.8,
+    "total_seconds": 3450.8
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace_b5_s1.pth"
+}

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-laplace/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
+1,29,107.5,452.9,"[0.29568020341690904, 0.6193671974528233, 0.19588368058977745, 0.33035580668548686, 0.5508702113809155, 0.7308332381787659, 0.15051414530012766, 0.23402934928630323]",,237130,23,True,kept,0.7237450842918105,0.7308332381787659,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.7308488562528247, 'uniform': 0.7388948429036649, 'top1': 0.7308332381787659, 'top2': 0.7311335062716425, 'top3': 0.7306909902224555}",0.8265384832710779,0.16094128778513914,0.04887358321577289,,,,,
+2,58,25.1,97.3,"[0.3171883581634619, 0.46570797027165556, 0.17555021224032277, 0.2911379528704814, 0.5171020431871733, 0.6011370938587045, 0.1856036164323916, 0.30947977967970813]",,672023,51,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6008390269849482, 'uniform': 0.6162660612905311, 'top1': 0.6011370938587045, 'top2': 0.6000244930568988, 'top3': 0.6001927615324153}",0.858755045097654,0.1513502459658489,0.012488309354615111,0.613930989815908,0.5449893692757639,student,0.38,
+3,87,24.9,394.2,"[0.3515194750513133, 0.5081483731457119, 0.1967517587863077, 0.30607281205994263, 0.4873790788992553, 0.6494763538876254, 0.27193648915742474, 0.30833822986758297]","[0.3713920260153615, 0.6156583110479326, 0.23892875624458382, 0.3647158945021577, 0.591553287347347, 0.8728288107309928, 0.2746848725997957, 0.24697999986747887]",1597523,80,False,,,,"[0.007602838779626683, 0.010501498964006668, 0.00011248570107448317, 0.0072047729084134095, 0.009461503682782231, 0.962908280806494, 0.000660211341494219, 0.0015484078161083674]",mixture,"{'mixture': 0.6498703208262191, 'uniform': 0.6380122833004724, 'top1': 0.6494763538876254, 'top2': 0.6498371076706292, 'top3': 0.6492208400303191}",0.852483095049041,0.17500004822089713,0.010542839366299465,0.38323891424672846,0.2603137417028411,student,0.0,
+4,116,24.6,447.0,"[0.37165118582176443, 0.5498739512200651, 0.19894122455555435, 0.32796989413566147, 0.49362131291149014, 0.6299155752418457, 0.29462911835367733, 0.31849510251907787]","[0.4432668421464532, 0.6655030182527801, 0.23613268382521105, 0.3915591066403852, 0.5820384495809867, 0.7883949429794211, 0.3380038795229039, 0.3135432956883544]",1870770,109,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6310020695428453, 'uniform': 0.6641508814931827, 'top1': 0.6299155752418457, 'top2': 0.6310825091972322, 'top3': 0.6297899101236573}",0.8797632053364045,0.15423802631351424,0.007480832876547508,0.7162450297722375,0.6766694366299267,student,0.721,
+5,145,24.3,472.3,"[0.3660794262070859, 0.5490828422499876, 0.20365563891426552, 0.32856908333908263, 0.508535981970067, 0.670062722012975, 0.2709322094932049, 0.3038809822055101]","[0.41543039333650833, 0.6614358260242903, 0.24373609609710786, 0.384228244423019, 0.603001507114574, 0.8509662250083222, 0.309811235614449, 0.2822458495441181]",2186789,138,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6707006667666381, 'uniform': 0.6809966041879847, 'top1': 0.670062722012975, 'top2': 0.6708633202970876, 'top3': 0.6700126434291589}",0.875244862502354,0.15694088885158228,0.008585482330209854,0.7263052830088944,0.5965506068759152,student,0.754,
+6,174,24.0,499.4,"[0.3681232181861549, 0.5551674270712343, 0.2021990584599616, 0.33618850765162767, 0.5048303375240749, 0.6641611574975141, 0.2725573674876931, 0.30352943057597503]","[0.43428106838072467, 0.6680803435489528, 0.23994257936146782, 0.39529555441652675, 0.5914807803652934, 0.8240333760711983, 0.3198122503561808, 0.29933052019728335]",2478728,165,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.664855616294795, 'uniform': 0.6832583651582114, 'top1': 0.6641611574975141, 'top2': 0.6650440389558555, 'top3': 0.6640915000195716}",0.8926858816531681,0.1439952713975571,0.007831732653372411,0.7929019284967859,0.5956894105964962,student,0.976,
+7,174,0.0,222.6,,,,,,,,,,,,0.8886851466384921,0.1466770562917574,0.007776746073254894,,,,,True

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/log.txt

@@ -0,0 +1,108 @@
+[10:07:32][      0s][START] target=octopus-vulgaris-vertical-lobe-sfltract variant=macs-v11-norestart-acq-mcdropout budget=5.0% seed=2 | {"task": "membrane", "target": "octopus-vulgaris-vertical-lobe-sfltract", "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": 8, "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}
+[10:07:32][      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.
+[10:07:32][      0s][DATA] octopus-vulgaris-vertical-lobe-sfltract: train=3584 val=768 test=768 patches
+[10:07:34][      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, snemi=3.55, mouse-brain=4.07
+[10:07:35][      3s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s
+[10:07:41][      9s][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
+[10:07:41][      9s][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
+[10:08:36][     65s][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
+[10:08:36][     65s][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
+[10:08:36][     65s][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
+[10:08:36][     65s][INIT] student initialised from teacher oct-glia (weight 0.636)
+[10:08:40][     68s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[10:08:40][     68s][KD] distilling the mixture ensemble (prob space) into the student
+[10:11:05][    214s][KD] soft labels for 3584 patches from 8 teachers in 146s
+[10:12:52][    320s][KD] epoch 1/5 | patches=3584 | loss=0.5966 | 107s
+[10:14:39][    427s][KD] epoch 2/5 | patches=3584 | loss=0.4565 | 107s
+[10:16:25][    534s][KD] epoch 3/5 | patches=3584 | loss=0.4027 | 107s
+[10:18:12][    640s][KD] epoch 4/5 | patches=3584 | loss=0.3656 | 107s
+[10:19:58][    747s][KD] epoch 5/5 | patches=3584 | loss=0.3396 | 106s
+[10:20:03][    751s][ROUND0] before any target label: val_dice=0.8449 val_vi=0.1459
+[10:20:03][    751s][AL] patches=3584 (pool 2338 after dropping blank; free blank negatives: True) | budget=179 (5.0%) | rounds=6 | per_round=29 | acquisition=mcdropout | first_round=stratified
+[10:20:17][    765s][AL] round-1 uncertainty quality on 2338 patches: {'spearman_uncertainty_error': 0.4329896382452679, 'auroc_top20_error': 0.6001805384158325}
+[10:21:39][    847s][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
+[10:21:48][    857s][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
+[10:21:49][    857s][WEIGHTS] round 1: student kept (labeled Dice 0.743 vs best teacher oct-glia 0.765); distilling the corrected ensemble into it
+[10:21:49][    857s][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
+[10:23:23][    952s][KD] soft labels for 2309 patches from 8 teachers in 94s
+[10:24:32][   1021s][KD] epoch 1/2 | patches=2309 | loss=0.1381 | 69s
+[10:25:41][   1089s][KD] epoch 2/2 | patches=2309 | loss=0.1326 | 69s
+[10:27:03][   1171s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 82s
+[10:27:08][   1176s][FT] round 1/6 epoch 1/5 | patches=58 | loss=0.8672
+[10:27:12][   1180s][FT] round 1/6 epoch 2/5 | patches=58 | loss=0.5693
+[10:27:17][   1185s][FT] round 1/6 epoch 3/5 | patches=58 | loss=0.5764
+[10:27:21][   1189s][FT] round 1/6 epoch 4/5 | patches=58 | loss=0.4051
+[10:27:25][   1194s][FT] round 1/6 epoch 5/5 | patches=58 | loss=0.4886
+[10:27:30][   1198s][AL] round 1/6 | labeled=29 | selection_seconds=96.1000 round_seconds=442.6000 restart_student_dice=0.7430 restart_teacher_dice=0.7648 val_dice=0.4271 val_vi=0.2099 val_ece=0.0166
+[10:27:44][   1212s][AL] round 2: 288/2309 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[10:28:02][   1230s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.362, uniform=0.381, top1=0.360, top2=0.360, top3=0.359 -> using uniform
+[10:28:02][   1230s][WEIGHTS] round 2: teacher Dice on 58 labeled patches + 58 blank (769883 fg px): c-elegans=0.226, cerebellum=0.317, flywire=0.125, berghia=0.184, microns=0.282, oct-glia=0.360, snemi=0.215, mouse-brain=0.253 | 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
+[10:28:19][   1247s][CONS] round 2: on the 29 new patches student Dice=0.404 vs ensemble 0.189 -> pseudo-labels from the student | consistency x0.00
+[10:28:22][   1251s][FT] round 2/6 epoch 1/5 | patches=116 | loss=1.2214
+[10:28:26][   1255s][FT] round 2/6 epoch 2/5 | patches=116 | loss=0.9194
+[10:28:30][   1259s][FT] round 2/6 epoch 3/5 | patches=116 | loss=1.1992
+[10:28:34][   1262s][FT] round 2/6 epoch 4/5 | patches=116 | loss=1.0262
+[10:28:38][   1266s][FT] round 2/6 epoch 5/5 | patches=116 | loss=0.7869
+[10:28:43][   1271s][AL] round 2/6 | labeled=58 | selection_seconds=14.2000 round_seconds=68.2000 pseudo_student_dice=0.4044 pseudo_ensemble_dice=0.1885 consistency_scale=0.0000 val_dice=0.8393 val_vi=0.1872 val_ece=0.0451
+[10:28:56][   1285s][AL] round 3: 558/2280 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[10:31:44][   1453s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.403, cerebellum=0.620, flywire=0.226, berghia=0.366, microns=0.543, oct-glia=0.767, snemi=0.319, mouse-brain=0.296 | blended with labeled Dice at 0.50
+[10:34:23][   1611s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.603, uniform=0.611, top1=0.603, top2=0.603, top3=0.603 -> using uniform
+[10:34:23][   1611s][WEIGHTS] round 3: teacher Dice on 87 labeled patches + 87 blank (1110404 fg px): c-elegans=0.335, cerebellum=0.500, flywire=0.183, berghia=0.301, microns=0.442, oct-glia=0.603, snemi=0.263, mouse-brain=0.282 | 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
+[10:34:39][   1627s][CONS] round 3: on the 29 new patches student Dice=0.758 vs ensemble 0.613 -> pseudo-labels from the student | consistency x0.86
+[10:34:50][   1638s][FT] round 3/6 epoch 1/5 | patches=174 | loss=0.9883
+[10:35:01][   1649s][FT] round 3/6 epoch 2/5 | patches=174 | loss=0.7900
+[10:35:12][   1660s][FT] round 3/6 epoch 3/5 | patches=174 | loss=0.6876
+[10:35:23][   1672s][FT] round 3/6 epoch 4/5 | patches=174 | loss=0.6380
+[10:35:34][   1683s][FT] round 3/6 epoch 5/5 | patches=174 | loss=0.7518
+[10:35:39][   1687s][AL] round 3/6 | labeled=87 | selection_seconds=14.0000 round_seconds=411.9000 pseudo_student_dice=0.7578 pseudo_ensemble_dice=0.6134 consistency_scale=0.8590 val_dice=0.8606 val_vi=0.1726 val_ece=0.0064
+[10:35:53][   1701s][AL] round 4: 563/2251 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[10:38:45][   1873s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.409, cerebellum=0.642, flywire=0.233, berghia=0.378, microns=0.575, oct-glia=0.810, snemi=0.320, mouse-brain=0.287 | blended with labeled Dice at 0.50
+[10:41:27][   2036s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.638, uniform=0.635, top1=0.637, top2=0.638, top3=0.637 -> using mixture
+[10:41:27][   2036s][WEIGHTS] round 4: teacher Dice on 116 labeled patches + 116 blank (1435427 fg px): c-elegans=0.338, cerebellum=0.505, flywire=0.188, berghia=0.309, microns=0.461, oct-glia=0.637, snemi=0.261, mouse-brain=0.273 | weights: c-elegans=0.008, cerebellum=0.012, flywire=0.000, berghia=0.008, microns=0.008, oct-glia=0.962, snemi=0.001, mouse-brain=0.001
+[10:41:44][   2052s][CONS] round 4: on the 29 new patches student Dice=0.736 vs ensemble 0.545 -> pseudo-labels from the student | consistency x0.79
+[10:41:58][   2066s][FT] round 4/6 epoch 1/5 | patches=232 | loss=0.6580
+[10:42:13][   2081s][FT] round 4/6 epoch 2/5 | patches=232 | loss=0.6007
+[10:42:27][   2095s][FT] round 4/6 epoch 3/5 | patches=232 | loss=0.4883
+[10:42:42][   2110s][FT] round 4/6 epoch 4/5 | patches=232 | loss=0.4529
+[10:42:56][   2124s][FT] round 4/6 epoch 5/5 | patches=232 | loss=0.4728
+[10:43:01][   2129s][AL] round 4/6 | labeled=116 | selection_seconds=13.8000 round_seconds=437.2000 pseudo_student_dice=0.7360 pseudo_ensemble_dice=0.5446 consistency_scale=0.7870 val_dice=0.8463 val_vi=0.1834 val_ece=0.0116
+[10:43:14][   2142s][AL] round 5: 554/2222 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[10:46:10][   2319s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.429, cerebellum=0.644, flywire=0.232, berghia=0.392, microns=0.571, oct-glia=0.789, snemi=0.335, mouse-brain=0.307 | blended with labeled Dice at 0.50
+[10:48:57][   2485s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.643, uniform=0.643, top1=0.643, top2=0.643, top3=0.642 -> using mixture
+[10:48:57][   2485s][WEIGHTS] round 5: teacher Dice on 145 labeled patches + 145 blank (1729181 fg px): c-elegans=0.348, cerebellum=0.509, flywire=0.190, berghia=0.317, microns=0.469, oct-glia=0.643, snemi=0.263, mouse-brain=0.283 | weights: c-elegans=0.008, cerebellum=0.011, flywire=0.000, berghia=0.009, microns=0.008, oct-glia=0.962, snemi=0.001, mouse-brain=0.001
+[10:49:13][   2501s][CONS] round 5: on the 29 new patches student Dice=0.674 vs ensemble 0.635 -> pseudo-labels from the student | consistency x0.58
+[10:49:31][   2519s][FT] round 5/6 epoch 1/5 | patches=290 | loss=0.5204
+[10:49:49][   2537s][FT] round 5/6 epoch 2/5 | patches=290 | loss=0.4699
+[10:50:07][   2555s][FT] round 5/6 epoch 3/5 | patches=290 | loss=0.4563
+[10:50:25][   2573s][FT] round 5/6 epoch 4/5 | patches=290 | loss=0.3912
+[10:50:43][   2591s][FT] round 5/6 epoch 5/5 | patches=290 | loss=0.3980
+[10:50:48][   2596s][AL] round 5/6 | labeled=145 | selection_seconds=13.5000 round_seconds=462.7000 pseudo_student_dice=0.6738 pseudo_ensemble_dice=0.6353 consistency_scale=0.5790 val_dice=0.8727 val_vi=0.1596 val_ece=0.0109
+[10:51:01][   2609s][AL] round 6: 570/2193 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[10:54:03][   2791s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.402, cerebellum=0.642, flywire=0.237, berghia=0.377, microns=0.593, oct-glia=0.844, snemi=0.306, mouse-brain=0.275 | blended with labeled Dice at 0.50
+[10:56:54][   2962s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.681, uniform=0.666, top1=0.680, top2=0.681, top3=0.680 -> using mixture
+[10:56:54][   2962s][WEIGHTS] round 6: teacher Dice on 174 labeled patches + 174 blank (2114988 fg px): c-elegans=0.339, cerebellum=0.523, flywire=0.197, berghia=0.317, microns=0.492, oct-glia=0.680, snemi=0.250, mouse-brain=0.271 | weights: c-elegans=0.005, cerebellum=0.009, flywire=0.000, berghia=0.006, microns=0.007, oct-glia=0.971, snemi=0.000, mouse-brain=0.001
+[10:57:10][   2978s][CONS] round 6: on the 29 new patches student Dice=0.694 vs ensemble 0.595 -> pseudo-labels from the student | consistency x0.65
+[10:57:32][   3000s][FT] round 6/6 epoch 1/5 | patches=348 | loss=0.4019
+[10:57:53][   3021s][FT] round 6/6 epoch 2/5 | patches=348 | loss=0.3831
+[10:58:14][   3043s][FT] round 6/6 epoch 3/5 | patches=348 | loss=0.3692
+[10:58:36][   3064s][FT] round 6/6 epoch 4/5 | patches=348 | loss=0.3469
+[10:58:57][   3085s][FT] round 6/6 epoch 5/5 | patches=348 | loss=0.3359
+[10:59:02][   3090s][AL] round 6/6 | labeled=174 | selection_seconds=13.4000 round_seconds=489.5000 pseudo_student_dice=0.6941 pseudo_ensemble_dice=0.5954 consistency_scale=0.6470 val_dice=0.8811 val_vi=0.1517 val_ece=0.0088
+[10:59:12][   3100s][CONS] final consolidation: pseudo-labels refreshed from the student
+[10:59:33][   3121s][FT] final 1/10 epoch 1/1 | patches=348 | loss=0.3833
+[10:59:54][   3142s][FT] final 2/10 epoch 1/1 | patches=348 | loss=0.3832
+[11:00:16][   3164s][FT] final 3/10 epoch 1/1 | patches=348 | loss=0.3428
+[11:00:37][   3185s][FT] final 4/10 epoch 1/1 | patches=348 | loss=0.3186
+[11:00:58][   3207s][FT] final 5/10 epoch 1/1 | patches=348 | loss=0.3337
+[11:01:20][   3228s][FT] final 6/10 epoch 1/1 | patches=348 | loss=0.3198
+[11:01:41][   3249s][FT] final 7/10 epoch 1/1 | patches=348 | loss=0.3078
+[11:02:02][   3271s][FT] final 8/10 epoch 1/1 | patches=348 | loss=0.3041
+[11:02:24][   3292s][FT] final 9/10 epoch 1/1 | patches=348 | loss=0.3015
+[11:02:45][   3313s][FT] final 10/10 epoch 1/1 | patches=348 | loss=0.3106
+[11:02:50][   3318s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=223.7000 val_dice=0.8877 val_vi=0.1480 val_ece=0.0073
+[11:02:50][   3318s][WEIGHTS] final: c-elegans=0.005, cerebellum=0.009, flywire=0.000, berghia=0.006, microns=0.007, oct-glia=0.971, snemi=0.000, mouse-brain=0.001
+[11:03:26][   3354s][TEST] raw student (no EMA, no TTA): patch_dice=0.8632 patch_vi=0.1811
+[11:03:30][   3359s][TEST] final model without TTA: patch_dice=0.8632 patch_vi=0.1811
+[11:03:30][   3359s][RESOURCES] {"phases": {"setup": {"seconds": 1.48, "peak_cpu_rss_mb": 2182.8, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 65.58, "peak_cpu_rss_mb": 4183.0, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 678.72, "peak_cpu_rss_mb": 5202.5, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4598.0, "kd_epochs": 5}, "active_learning": {"seconds": 2566.86, "peak_cpu_rss_mb": 5202.5, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4606.0, "rounds": 6}, "inference": {"seconds": 40.64, "peak_cpu_rss_mb": 5202.5, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4606.0, "patches": 768, "images": 6, "patches_per_second": 18.9, "seconds_per_image": 6.773}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5202.5, "total_seconds": 3353.3}
+[11:03:30][   3359s][TEST] patch_dice=0.8643 patch_iou=0.8032 patch_precision=0.8908 patch_recall=0.8768 patch_vi=0.1774 image_dice=0.7988 image_iou=0.6722 image_precision=0.7880 image_recall=0.8158 image_vi=0.2006 cal_ece=0.0069 cal_nll=0.0758 cal_brier=0.0221 cal_mean_confidence=0.9769 cal_accuracy=0.9700 cal_predicted_prevalence=0.0760 cal_true_prevalence=0.0718 cal_mean_probability=0.0768
+[11:03:31][   3359s][DONE] octopus-vulgaris-vertical-lobe-sfltract macs-v11-norestart-acq-mcdropout b5 s2 | test patch Dice=0.8643 VI=0.1774 | image Dice=0.7988 VI=0.2006 | 3359s

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-mcdropout/budget-5/seed-2/result.json

@@ -0,0 +1,785 @@
+{
+  "target": "octopus-vulgaris-vertical-lobe-sfltract",
+  "variant": "macs-v11-norestart-acq-mcdropout",
+  "budget": 5.0,
+  "seed": 2,
+  "config": {
+    "task": "membrane",
+    "target": "octopus-vulgaris-vertical-lobe-sfltract",
+    "variant": "macs-v11-norestart-acq-mcdropout",
+    "budget": 5.0,
+    "seed": 2,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "octopus-vulgaris-vertical-lobe-sfltract",
+      "--variants",
+      "macs-v11-norestart-acq-mcdropout",
+      "--budgets",
+      "5",
+      "--seeds",
+      "2",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.09888716021456875,
+    0.02852231397957182,
+    0.006877159468562143,
+    0.147623932500446,
+    0.03162958721157525,
+    0.636412589592954,
+    0.019031438719961654,
+    0.031015818312360402
+  ],
+  "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": [
+      0.09888716021456875,
+      0.02852231397957182,
+      0.006877159468562143,
+      0.147623932500446,
+      0.03162958721157525,
+      0.636412589592954,
+      0.019031438719961654,
+      0.031015818312360402
+    ],
+    "per_source": [
+      0.3196471036584027,
+      0.016524109438155077,
+      0.0007715730533314212,
+      0.17539928949217498,
+      0.027481458110301667,
+      0.4363205819091585,
+      0.016643531248023022,
+      0.007212353090452595
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.00048686699692031786,
+      "mmd2_mixture": 0.012755679331756475,
+      "mmd2_uniform": 0.09698314484068699,
+      "mmd2_single_source": [
+        0.07244562949164945,
+        0.2568973175367495,
+        0.39290896557558747,
+        0.05287206442937192,
+        0.2649845830320105,
+        0.011934176639413208,
+        0.33925886828153656,
+        0.26844309711822195
+      ],
+      "median_squared_distance": 0.1834397705913593,
+      "gammas": [
+        1.3628451409095685,
+        2.725690281819137,
+        5.451380563638274,
+        10.902761127276548,
+        21.805522254553097
+      ],
+      "max_weight": 0.636412589592954,
+      "entropy": 1.2268812334836157,
+      "effective_sources": 3.4105761415276636,
+      "min_weight": 0.006877159468562143
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.049620160284728154,
+        0.42663804362685775,
+        0.8166050888075048,
+        0.12600028070223912,
+        0.36189803378480956,
+        0.01001954120360149,
+        0.42572157074448214,
+        0.5321462950254623
+      ],
+      "standardized": [
+        -1.1548909896059252,
+        0.32630755724364063,
+        1.8583797643241364,
+        -0.8548147110811022,
+        0.07196204012648774,
+        -1.3104708228948398,
+        0.3227069898393355,
+        0.740820172048267
+      ],
+      "tau": 0.5,
+      "max_weight": 0.4363205819091585,
+      "entropy": 1.3076009419864412,
+      "effective_sources": 3.6972930433455575,
+      "min_weight": 0.0007715730533314212
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+    "seconds": 6.6
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.47571341080584045,
+      0.6329400594773938,
+      0.1419064245794933,
+      0.4669233743357945,
+      0.7079863797747427,
+      0.8450584862058909,
+      0.22902171788078127,
+      0.11473171105409108
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.8488016622672679,
+      "mixture@raw": 0.8468775415451569,
+      "mixture@logit": 0.8498181877900208,
+      "per_source": 0.8430718482924652,
+      "per_source@raw": 0.6192917168334541,
+      "per_source@logit": 0.8479512124278182,
+      "uniform": 0.7183086201637726,
+      "uniform@raw": 0.6771618878909939,
+      "uniform@logit": 0.8065091143369424
+    },
+    "val_patches": 768
+  },
+  "temperatures": [
+    2.575598955154419,
+    4.557764530181885,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    3.547258138656616,
+    4.074584007263184
+  ],
+  "top_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+  "weight_stats": {
+    "max_weight": 0.636412589592954,
+    "entropy": 1.2268812334836157,
+    "effective_sources": 3.4105761415276636,
+    "min_weight": 0.006877159468562143
+  },
+  "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": 678.7,
+  "round0_val": {
+    "val_dice": 0.844884651592599,
+    "val_vi": 0.14593690508010287,
+    "val_ece": 0.06297539540092424
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 29,
+      "selection_seconds": 96.1,
+      "round_seconds": 442.6,
+      "teacher_dice_labeled": [
+        0.25717887499142283,
+        0.5690769623050445,
+        0.1686301357941244,
+        0.27579610687630757,
+        0.5811094797160338,
+        0.7648037373241484,
+        0.10985651675729007,
+        0.20342648065706517
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 179039,
+      "evidence_nonempty_patches": 21,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.7429788813082652,
+      "restart_teacher_dice": 0.7648037373241484,
+      "weights": [
+        0.0009536818496459393,
+        0.006223068059955061,
+        2.7359642534432497e-05,
+        0.0017150389707529289,
+        0.0077834094558282,
+        0.9830806325167469,
+        4.206514211543207e-05,
+        0.0001747443624209932
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.7647068027764391,
+        "uniform": 0.7405350046482169,
+        "top1": 0.7648037373241484,
+        "top2": 0.7643927747284575,
+        "top3": 0.7646270894165438
+      },
+      "val_dice": 0.4270724097964022,
+      "val_vi": 0.20986386428805637,
+      "val_ece": 0.016562829469566765
+    },
+    {
+      "round": 2,
+      "n_labeled": 58,
+      "selection_seconds": 14.2,
+      "round_seconds": 68.2,
+      "teacher_dice_labeled": [
+        0.22587745940791537,
+        0.31698070371161113,
+        0.12462178899546054,
+        0.18431241169575352,
+        0.2822639835725904,
+        0.35970436518487914,
+        0.21512327894146918,
+        0.25303882275789313
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 769883,
+      "evidence_nonempty_patches": 47,
+      "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": 0.3621730033525568,
+        "uniform": 0.38051786207576516,
+        "top1": 0.35970436518487914,
+        "top2": 0.36039438680228353,
+        "top3": 0.3593144880664838
+      },
+      "pseudo_student_dice": 0.404397978293183,
+      "pseudo_ensemble_dice": 0.1885413347061179,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.839313244845056,
+      "val_vi": 0.1871619919775931,
+      "val_ece": 0.04511803920160227
+    },
+    {
+      "round": 3,
+      "n_labeled": 87,
+      "selection_seconds": 14.0,
+      "round_seconds": 411.9,
+      "teacher_dice_labeled": [
+        0.3347328518610615,
+        0.5001404973411617,
+        0.18257684052307566,
+        0.301204699930828,
+        0.4424734262397891,
+        0.6025802942919263,
+        0.2628953407838804,
+        0.2819626314895449
+      ],
+      "teacher_dice_pool": [
+        0.4027438950161543,
+        0.6200607333924688,
+        0.22590870742348548,
+        0.366050843102077,
+        0.5429376205791411,
+        0.7671385769259249,
+        0.31892228647137716,
+        0.2960467280637949
+      ],
+      "evidence_fg_pixels": 1110404,
+      "evidence_nonempty_patches": 74,
+      "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": 0.6034687502246806,
+        "uniform": 0.6108538918617112,
+        "top1": 0.6025802942919263,
+        "top2": 0.6033994926595407,
+        "top3": 0.6025226457729906
+      },
+      "pseudo_student_dice": 0.7577983328352536,
+      "pseudo_ensemble_dice": 0.6134312812346644,
+      "pseudo_source": "student",
+      "consistency_scale": 0.859,
+      "val_dice": 0.8606391248002924,
+      "val_vi": 0.17258987468099388,
+      "val_ece": 0.006442501714751189
+    },
+    {
+      "round": 4,
+      "n_labeled": 116,
+      "selection_seconds": 13.8,
+      "round_seconds": 437.2,
+      "teacher_dice_labeled": [
+        0.3381786272232266,
+        0.5052704073606991,
+        0.1878705984148606,
+        0.3089617272275621,
+        0.46140194760202413,
+        0.6371013035682522,
+        0.26058193018573017,
+        0.27253356202443413
+      ],
+      "teacher_dice_pool": [
+        0.4089333685603796,
+        0.6416396488347919,
+        0.23295701412877842,
+        0.3780127247463931,
+        0.5753647964100921,
+        0.8100108079944658,
+        0.3195180987768935,
+        0.2872763217692721
+      ],
+      "evidence_fg_pixels": 1435427,
+      "evidence_nonempty_patches": 100,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.007524908071981864,
+        0.01154028224680597,
+        0.00011641016732088002,
+        0.008387464194102436,
+        0.008252905997431267,
+        0.9622872783077605,
+        0.0006665530822541776,
+        0.0012241979323426642
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.637705710703107,
+        "uniform": 0.6351334535951196,
+        "top1": 0.6371013035682522,
+        "top2": 0.6377429880954093,
+        "top3": 0.636979409689884
+      },
+      "pseudo_student_dice": 0.7359883623178871,
+      "pseudo_ensemble_dice": 0.5445880183283505,
+      "pseudo_source": "student",
+      "consistency_scale": 0.787,
+      "val_dice": 0.8462507575573692,
+      "val_vi": 0.18344553848250167,
+      "val_ece": 0.011602489143117903
+    },
+    {
+      "round": 5,
+      "n_labeled": 145,
+      "selection_seconds": 13.5,
+      "round_seconds": 462.7,
+      "teacher_dice_labeled": [
+        0.34783195263869726,
+        0.5085847810456393,
+        0.18989485557335906,
+        0.31691661905716506,
+        0.46906773359921405,
+        0.6426591476342762,
+        0.2632133787604556,
+        0.2833962106620097
+      ],
+      "teacher_dice_pool": [
+        0.4291246304900841,
+        0.6440484423766484,
+        0.23188019058719783,
+        0.39202517928015146,
+        0.5713678412064529,
+        0.7885259140627896,
+        0.33496846094096766,
+        0.306770538899655
+      ],
+      "evidence_fg_pixels": 1729181,
+      "evidence_nonempty_patches": 128,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.007835715882556655,
+        0.011278833739633114,
+        0.00011231448567228763,
+        0.008586812058593643,
+        0.008424664331596915,
+        0.9618243695229529,
+        0.0006470183518167428,
+        0.0012902716271775594
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.6431960380116553,
+        "uniform": 0.6427129289897711,
+        "top1": 0.6426591476342762,
+        "top2": 0.64328273965317,
+        "top3": 0.6424447067725132
+      },
+      "pseudo_student_dice": 0.6737726910002576,
+      "pseudo_ensemble_dice": 0.6352852470158447,
+      "pseudo_source": "student",
+      "consistency_scale": 0.579,
+      "val_dice": 0.872681189635275,
+      "val_vi": 0.15964571873942823,
+      "val_ece": 0.010866263393352124
+    },
+    {
+      "round": 6,
+      "n_labeled": 174,
+      "selection_seconds": 13.4,
+      "round_seconds": 489.5,
+      "teacher_dice_labeled": [
+        0.3394639924416737,
+        0.5227582892078185,
+        0.197064979552099,
+        0.3167971266642039,
+        0.4920812887171,
+        0.6802133425724319,
+        0.25012603080841567,
+        0.2710178116401448
+      ],
+      "teacher_dice_pool": [
+        0.4019549577525064,
+        0.6424874197113001,
+        0.23732978669094343,
+        0.37713044373411325,
+        0.5927859949030386,
+        0.8444809846026134,
+        0.3055219464803523,
+        0.2746842105078218
+      ],
+      "evidence_fg_pixels": 2114988,
+      "evidence_nonempty_patches": 157,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.004999738768501313,
+        0.00901631717958687,
+        8.371156243438058e-05,
+        0.00595008583835789,
+        0.007357146913473611,
+        0.9714082673176162,
+        0.00039381222300683295,
+        0.0007909201970230076
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.6805360799138476,
+        "uniform": 0.6664409579587651,
+        "top1": 0.6802133425724319,
+        "top2": 0.6806826202966263,
+        "top3": 0.6800856327938514
+      },
+      "pseudo_student_dice": 0.6941139408873892,
+      "pseudo_ensemble_dice": 0.5953877530550494,
+      "pseudo_source": "student",
+      "consistency_scale": 0.647,
+      "val_dice": 0.8810563017493284,
+      "val_vi": 0.1516974844445877,
+      "val_ece": 0.008790176335206875
+    },
+    {
+      "round": 7,
+      "n_labeled": 174,
+      "selection_seconds": 0.0,
+      "round_seconds": 223.7,
+      "final_consolidation": true,
+      "val_dice": 0.8877374793949646,
+      "val_vi": 0.1480138674777313,
+      "val_ece": 0.00734695886139486
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.4329896382452679,
+    "auroc_top20_error": 0.6001805384158325
+  },
+  "al_seconds": 2566.9,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.004999738768501313,
+    0.00901631717958687,
+    8.371156243438058e-05,
+    0.00595008583835789,
+    0.007357146913473611,
+    0.9714082673176162,
+    0.00039381222300683295,
+    0.0007909201970230076
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9714082673176162,
+    "entropy": 0.17327628466338119,
+    "effective_sources": 1.1891946159348459,
+    "min_weight": 8.371156243438058e-05
+  },
+  "top_source_final": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.8631586445206421,
+    "patch_iou": 0.799814331641088,
+    "patch_precision": 0.8821052042572802,
+    "patch_recall": 0.8785268631377646,
+    "patch_vi": 0.18109389859865985,
+    "n_patches": 768,
+    "cal_ece": 0.008786372724529175,
+    "cal_nll": 0.08220366447098529,
+    "cal_brier": 0.022817539634909845,
+    "cal_mean_confidence": 0.9779814207388521,
+    "cal_accuracy": 0.9691950480143229,
+    "cal_predicted_prevalence": 0.07668304443359375,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.0772628882504308
+  },
+  "test_notta": {
+    "patch_dice": 0.8631586445206421,
+    "patch_iou": 0.799814331641088,
+    "patch_precision": 0.8821052042572802,
+    "patch_recall": 0.8785268631377646,
+    "patch_vi": 0.18109389859865985,
+    "n_patches": 768,
+    "cal_ece": 0.008786372724529175,
+    "cal_nll": 0.08220366447098529,
+    "cal_brier": 0.022817539634909845,
+    "cal_mean_confidence": 0.9779814207388521,
+    "cal_accuracy": 0.9691950480143229,
+    "cal_predicted_prevalence": 0.07668304443359375,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.0772628882504308
+  },
+  "test": {
+    "patch_dice": 0.8642838710090276,
+    "patch_iou": 0.8032092821758569,
+    "patch_precision": 0.890816581492523,
+    "patch_recall": 0.8768043999013256,
+    "patch_vi": 0.17736875597624022,
+    "n_patches": 768,
+    "image_dice": 0.7988423190301718,
+    "image_iou": 0.672154603933092,
+    "image_precision": 0.7880281390563937,
+    "image_recall": 0.8158464700761794,
+    "image_vi": 0.2006298433573774,
+    "n_images": 6,
+    "cal_ece": 0.006921095013166664,
+    "cal_nll": 0.07579974673544221,
+    "cal_brier": 0.02208113620039101,
+    "cal_mean_confidence": 0.9768672505378339,
+    "cal_accuracy": 0.9700101216634115,
+    "cal_predicted_prevalence": 0.07602818806966145,
+    "cal_true_prevalence": 0.07175954182942708,
+    "cal_mean_probability": 0.07675483929619768
+  },
+  "eval_seconds": 40.6,
+  "total_seconds": 3358.6,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 1.48,
+        "peak_cpu_rss_mb": 2182.8,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 65.58,
+        "peak_cpu_rss_mb": 4183.0,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 678.72,
+        "peak_cpu_rss_mb": 5202.5,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4598.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2566.86,
+        "peak_cpu_rss_mb": 5202.5,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4606.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 40.64,
+        "peak_cpu_rss_mb": 5202.5,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4606.0,
+        "patches": 768,
+        "images": 6,
+        "patches_per_second": 18.9,
+        "seconds_per_image": 6.773
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5202.5,
+    "total_seconds": 3353.3
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/octopus-vulgaris-vertical-lobe-sfltract/macs-v11-norestart-acq-mcdropout_b5_s2.pth"
+}

macs_v2_r26/runs/membrane/octopus-vulgaris-vertical-lobe-sfltract/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
+1,29,96.1,442.6,"[0.25717887499142283, 0.5690769623050445, 0.1686301357941244, 0.27579610687630757, 0.5811094797160338, 0.7648037373241484, 0.10985651675729007, 0.20342648065706517]",,179039,21,True,kept,0.7429788813082652,0.7648037373241484,"[0.0009536818496459393, 0.006223068059955061, 2.7359642534432497e-05, 0.0017150389707529289, 0.0077834094558282, 0.9830806325167469, 4.206514211543207e-05, 0.0001747443624209932]",mixture,"{'mixture': 0.7647068027764391, 'uniform': 0.7405350046482169, 'top1': 0.7648037373241484, 'top2': 0.7643927747284575, 'top3': 0.7646270894165438}",0.4270724097964022,0.20986386428805637,0.016562829469566765,,,,,
+2,58,14.2,68.2,"[0.22587745940791537, 0.31698070371161113, 0.12462178899546054, 0.18431241169575352, 0.2822639835725904, 0.35970436518487914, 0.21512327894146918, 0.25303882275789313]",,769883,47,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.3621730033525568, 'uniform': 0.38051786207576516, 'top1': 0.35970436518487914, 'top2': 0.36039438680228353, 'top3': 0.3593144880664838}",0.839313244845056,0.1871619919775931,0.04511803920160227,0.404397978293183,0.1885413347061179,student,0.0,
+3,87,14.0,411.9,"[0.3347328518610615, 0.5001404973411617, 0.18257684052307566, 0.301204699930828, 0.4424734262397891, 0.6025802942919263, 0.2628953407838804, 0.2819626314895449]","[0.4027438950161543, 0.6200607333924688, 0.22590870742348548, 0.366050843102077, 0.5429376205791411, 0.7671385769259249, 0.31892228647137716, 0.2960467280637949]",1110404,74,False,,,,"[0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125]",uniform,"{'mixture': 0.6034687502246806, 'uniform': 0.6108538918617112, 'top1': 0.6025802942919263, 'top2': 0.6033994926595407, 'top3': 0.6025226457729906}",0.8606391248002924,0.17258987468099388,0.006442501714751189,0.7577983328352536,0.6134312812346644,student,0.859,
+4,116,13.8,437.2,"[0.3381786272232266, 0.5052704073606991, 0.1878705984148606, 0.3089617272275621, 0.46140194760202413, 0.6371013035682522, 0.26058193018573017, 0.27253356202443413]","[0.4089333685603796, 0.6416396488347919, 0.23295701412877842, 0.3780127247463931, 0.5753647964100921, 0.8100108079944658, 0.3195180987768935, 0.2872763217692721]",1435427,100,False,,,,"[0.007524908071981864, 0.01154028224680597, 0.00011641016732088002, 0.008387464194102436, 0.008252905997431267, 0.9622872783077605, 0.0006665530822541776, 0.0012241979323426642]",mixture,"{'mixture': 0.637705710703107, 'uniform': 0.6351334535951196, 'top1': 0.6371013035682522, 'top2': 0.6377429880954093, 'top3': 0.636979409689884}",0.8462507575573692,0.18344553848250167,0.011602489143117903,0.7359883623178871,0.5445880183283505,student,0.787,
+5,145,13.5,462.7,"[0.34783195263869726, 0.5085847810456393, 0.18989485557335906, 0.31691661905716506, 0.46906773359921405, 0.6426591476342762, 0.2632133787604556, 0.2833962106620097]","[0.4291246304900841, 0.6440484423766484, 0.23188019058719783, 0.39202517928015146, 0.5713678412064529, 0.7885259140627896, 0.33496846094096766, 0.306770538899655]",1729181,128,False,,,,"[0.007835715882556655, 0.011278833739633114, 0.00011231448567228763, 0.008586812058593643, 0.008424664331596915, 0.9618243695229529, 0.0006470183518167428, 0.0012902716271775594]",mixture,"{'mixture': 0.6431960380116553, 'uniform': 0.6427129289897711, 'top1': 0.6426591476342762, 'top2': 0.64328273965317, 'top3': 0.6424447067725132}",0.872681189635275,0.15964571873942823,0.010866263393352124,0.6737726910002576,0.6352852470158447,student,0.579,
+6,174,13.4,489.5,"[0.3394639924416737, 0.5227582892078185, 0.197064979552099, 0.3167971266642039, 0.4920812887171, 0.6802133425724319, 0.25012603080841567, 0.2710178116401448]","[0.4019549577525064, 0.6424874197113001, 0.23732978669094343, 0.37713044373411325, 0.5927859949030386, 0.8444809846026134, 0.3055219464803523, 0.2746842105078218]",2114988,157,False,,,,"[0.004999738768501313, 0.00901631717958687, 8.371156243438058e-05, 0.00595008583835789, 0.007357146913473611, 0.9714082673176162, 0.00039381222300683295, 0.0007909201970230076]",mixture,"{'mixture': 0.6805360799138476, 'uniform': 0.6664409579587651, 'top1': 0.6802133425724319, 'top2': 0.6806826202966263, 'top3': 0.6800856327938514}",0.8810563017493284,0.1516974844445877,0.008790176335206875,0.6941139408873892,0.5953877530550494,student,0.647,
+7,174,0.0,223.7,,,,,,,,,,,,0.8877374793949646,0.1480138674777313,0.00734695886139486,,,,,True

macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt

@@ -0,0 +1,104 @@
+[09:14:20][      0s][START] target=snemi3d variant=macs-v11-norestart-acq-laplace budget=5.0% seed=2 | {"task": "membrane", "target": "snemi3d", "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": 8, "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}
+[09:14:20][      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.
+[09:14:20][      0s][DATA] snemi3d: train=3430 val=735 test=735 patches
+[09:14:22][      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=4.07
+[09:14:22][      3s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s
+[09:14:29][      9s][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
+[09:14:29][      9s][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
+[09:15:22][     62s][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
+[09:15:22][     62s][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
+[09:15:22][     62s][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
+[09:15:22][     62s][INIT] student initialised from teacher microns (weight 0.477)
+[09:15:25][     66s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[09:15:25][     66s][KD] distilling the mixture ensemble (prob space) into the student
+[09:17:45][    206s][KD] soft labels for 3430 patches from 8 teachers in 140s
+[09:19:27][    308s][KD] epoch 1/5 | patches=3430 | loss=0.4414 | 102s
+[09:21:09][    410s][KD] epoch 2/5 | patches=3430 | loss=0.3718 | 102s
+[09:22:51][    512s][KD] epoch 3/5 | patches=3430 | loss=0.3565 | 102s
+[09:24:33][    614s][KD] epoch 4/5 | patches=3430 | loss=0.3515 | 102s
+[09:26:15][    715s][KD] epoch 5/5 | patches=3430 | loss=0.3483 | 102s
+[09:26:19][    720s][ROUND0] before any target label: val_dice=0.3835 val_vi=0.6426
+[09:26:19][    720s][AL] patches=3430 (pool 3430 after dropping blank; free blank negatives: True) | budget=171 (5.0%) | rounds=6 | per_round=28 | acquisition=laplace | first_round=stratified
+[09:26:57][    757s][AL] round-1 uncertainty quality on 3430 patches: {'spearman_uncertainty_error': 0.5072424045387947, 'auroc_top20_error': 0.8403184472456204}
+[09:28:19][    839s][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
+[09:28:24][    844s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.521, uniform=0.380, top1=0.522, top2=0.522, top3=0.522 -> using mixture
+[09:28:24][    845s][WEIGHTS] round 1: student kept (labeled Dice 0.412 vs best teacher microns 0.522); distilling the corrected ensemble into it
+[09:28:24][    845s][WEIGHTS] round 1: teacher Dice on 28 labeled patches + 0 blank (232196 fg px): c-elegans=0.309, cerebellum=0.401, flywire=0.213, berghia=0.349, microns=0.522, oct-glia=0.163, oct-sfl=0.203, mouse-brain=0.230 | weights: c-elegans=0.001, cerebellum=0.033, flywire=0.035, berghia=0.009, microns=0.918, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.002
+[09:30:43][    984s][KD] soft labels for 3402 patches from 8 teachers in 139s
+[09:32:24][   1085s][KD] epoch 1/2 | patches=3402 | loss=0.2505 | 101s
+[09:34:05][   1186s][KD] epoch 2/2 | patches=3402 | loss=0.2471 | 101s
+[09:35:27][   1268s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 82s
+[09:35:30][   1270s][FT] round 1/6 epoch 1/5 | patches=28 | loss=1.0468
+[09:35:32][   1272s][FT] round 1/6 epoch 2/5 | patches=28 | loss=0.9720
+[09:35:35][   1275s][FT] round 1/6 epoch 3/5 | patches=28 | loss=0.9385
+[09:35:37][   1277s][FT] round 1/6 epoch 4/5 | patches=28 | loss=0.9014
+[09:35:40][   1280s][FT] round 1/6 epoch 5/5 | patches=28 | loss=0.8928
+[09:35:44][   1284s][AL] round 1/6 | labeled=28 | selection_seconds=119.3000 round_seconds=560.4000 restart_student_dice=0.4118 restart_teacher_dice=0.5217 val_dice=0.7175 val_vi=0.5862 val_ece=0.0104
+[09:36:21][   1321s][AL] round 2: 3/3402 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[09:36:30][   1331s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.482, uniform=0.311, top1=0.486, top2=0.487, top3=0.487 -> using top3
+[09:36:30][   1331s][WEIGHTS] round 2: teacher Dice on 56 labeled patches + 0 blank (551979 fg px): c-elegans=0.330, cerebellum=0.390, flywire=0.196, berghia=0.293, microns=0.486, oct-glia=0.129, oct-sfl=0.220, mouse-brain=0.271 | weights: c-elegans=0.002, cerebellum=0.044, flywire=0.000, berghia=0.000, microns=0.954, oct-glia=0.000, oct-sfl=0.000, mouse-brain=0.000
+[09:36:44][   1344s][CONS] round 2: on the 28 new patches student Dice=0.609 vs ensemble 0.454 -> pseudo-labels from the student | consistency x0.36
+[09:36:48][   1348s][FT] round 2/6 epoch 1/5 | patches=56 | loss=0.8503
+[09:36:52][   1352s][FT] round 2/6 epoch 2/5 | patches=56 | loss=0.8056
+[09:36:56][   1356s][FT] round 2/6 epoch 3/5 | patches=56 | loss=0.7764
+[09:37:00][   1360s][FT] round 2/6 epoch 4/5 | patches=56 | loss=0.7538
+[09:37:04][   1364s][FT] round 2/6 epoch 5/5 | patches=56 | loss=0.7283
+[09:37:08][   1369s][AL] round 2/6 | labeled=56 | selection_seconds=37.0000 round_seconds=80.0000 pseudo_student_dice=0.6088 pseudo_ensemble_dice=0.4538 consistency_scale=0.3630 val_dice=0.7391 val_vi=0.5686 val_ece=0.0112
+[09:40:27][   1567s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.390, cerebellum=0.476, flywire=0.239, berghia=0.389, microns=0.668, oct-glia=0.198, oct-sfl=0.276, mouse-brain=0.309 | blended with labeled Dice at 0.50
+[09:42:59][   1719s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.575, uniform=0.366, top1=0.576, top2=0.578, top3=0.578 -> using mixture
+[09:42:59][   1719s][WEIGHTS] round 3: teacher Dice on 84 labeled patches + 0 blank (913408 fg px): c-elegans=0.371, cerebellum=0.445, flywire=0.207, berghia=0.335, microns=0.576, oct-glia=0.155, oct-sfl=0.237, mouse-brain=0.298 | weights: c-elegans=0.001, cerebellum=0.031, flywire=0.019, berghia=0.005, microns=0.940, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[09:43:13][   1733s][CONS] round 3: on the 28 new patches student Dice=0.636 vs ensemble 0.480 -> pseudo-labels from the student | consistency x0.45
+[09:43:18][   1739s][FT] round 3/6 epoch 1/5 | patches=84 | loss=0.7969
+[09:43:24][   1745s][FT] round 3/6 epoch 2/5 | patches=84 | loss=0.7781
+[09:43:30][   1750s][FT] round 3/6 epoch 3/5 | patches=84 | loss=0.7675
+[09:43:36][   1756s][FT] round 3/6 epoch 4/5 | patches=84 | loss=0.7291
+[09:43:42][   1762s][FT] round 3/6 epoch 5/5 | patches=84 | loss=0.7268
+[09:43:46][   1766s][AL] round 3/6 | labeled=84 | selection_seconds=36.7000 round_seconds=393.4000 pseudo_student_dice=0.6362 pseudo_ensemble_dice=0.4801 consistency_scale=0.4540 val_dice=0.7496 val_vi=0.5489 val_ece=0.0118
+[09:47:06][   1966s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.400, cerebellum=0.482, flywire=0.235, berghia=0.379, microns=0.649, oct-glia=0.197, oct-sfl=0.269, mouse-brain=0.319 | blended with labeled Dice at 0.50
+[09:49:40][   2120s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.565, uniform=0.360, top1=0.566, top2=0.567, top3=0.567 -> using mixture
+[09:49:40][   2120s][WEIGHTS] round 4: teacher Dice on 112 labeled patches + 0 blank (1179147 fg px): c-elegans=0.373, cerebellum=0.443, flywire=0.204, berghia=0.328, microns=0.566, oct-glia=0.155, oct-sfl=0.243, mouse-brain=0.300 | weights: c-elegans=0.002, cerebellum=0.033, flywire=0.021, berghia=0.005, microns=0.936, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[09:49:54][   2134s][CONS] round 4: on the 28 new patches student Dice=0.665 vs ensemble 0.476 -> pseudo-labels from the student | consistency x0.55
+[09:50:01][   2142s][FT] round 4/6 epoch 1/5 | patches=112 | loss=0.7404
+[09:50:08][   2149s][FT] round 4/6 epoch 2/5 | patches=112 | loss=0.7197
+[09:50:16][   2156s][FT] round 4/6 epoch 3/5 | patches=112 | loss=0.7162
+[09:50:23][   2163s][FT] round 4/6 epoch 4/5 | patches=112 | loss=0.6985
+[09:50:30][   2171s][FT] round 4/6 epoch 5/5 | patches=112 | loss=0.6919
+[09:50:35][   2175s][AL] round 4/6 | labeled=112 | selection_seconds=36.4000 round_seconds=404.4000 pseudo_student_dice=0.6648 pseudo_ensemble_dice=0.4758 consistency_scale=0.5490 val_dice=0.7550 val_vi=0.5303 val_ece=0.0146
+[09:53:57][   2377s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.379, cerebellum=0.471, flywire=0.242, berghia=0.378, microns=0.642, oct-glia=0.196, oct-sfl=0.279, mouse-brain=0.302 | blended with labeled Dice at 0.50
+[09:56:33][   2533s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.558, uniform=0.358, top1=0.559, top2=0.560, top3=0.560 -> using mixture
+[09:56:33][   2533s][WEIGHTS] round 5: teacher Dice on 140 labeled patches + 0 blank (1474014 fg px): c-elegans=0.363, cerebellum=0.436, flywire=0.203, berghia=0.330, microns=0.559, oct-glia=0.156, oct-sfl=0.240, mouse-brain=0.291 | weights: c-elegans=0.001, cerebellum=0.033, flywire=0.022, berghia=0.005, microns=0.934, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[09:56:47][   2547s][CONS] round 5: on the 28 new patches student Dice=0.644 vs ensemble 0.449 -> pseudo-labels from the student | consistency x0.48
+[09:56:56][   2556s][FT] round 5/6 epoch 1/5 | patches=140 | loss=0.7162
+[09:57:05][   2565s][FT] round 5/6 epoch 2/5 | patches=140 | loss=0.6915
+[09:57:14][   2574s][FT] round 5/6 epoch 3/5 | patches=140 | loss=0.7144
+[09:57:23][   2583s][FT] round 5/6 epoch 4/5 | patches=140 | loss=0.7015
+[09:57:32][   2593s][FT] round 5/6 epoch 5/5 | patches=140 | loss=0.6746
+[09:57:36][   2597s][AL] round 5/6 | labeled=140 | selection_seconds=36.2000 round_seconds=417.5000 pseudo_student_dice=0.6435 pseudo_ensemble_dice=0.4493 consistency_scale=0.4780 val_dice=0.7550 val_vi=0.5302 val_ece=0.0132
+[10:01:00][   2801s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.384, cerebellum=0.468, flywire=0.235, berghia=0.383, microns=0.635, oct-glia=0.192, oct-sfl=0.273, mouse-brain=0.307 | blended with labeled Dice at 0.50
+[10:03:38][   2959s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.557, uniform=0.359, top1=0.558, top2=0.560, top3=0.560 -> using mixture
+[10:03:38][   2959s][WEIGHTS] round 6: teacher Dice on 168 labeled patches + 0 blank (1744619 fg px): c-elegans=0.364, cerebellum=0.435, flywire=0.201, berghia=0.333, microns=0.558, oct-glia=0.157, oct-sfl=0.239, mouse-brain=0.293 | weights: c-elegans=0.001, cerebellum=0.033, flywire=0.022, berghia=0.006, microns=0.934, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[10:03:52][   2972s][CONS] round 6: on the 28 new patches student Dice=0.682 vs ensemble 0.509 -> pseudo-labels from the student | consistency x0.61
+[10:04:02][   2983s][FT] round 6/6 epoch 1/5 | patches=168 | loss=0.6951
+[10:04:13][   2993s][FT] round 6/6 epoch 2/5 | patches=168 | loss=0.6870
+[10:04:24][   3004s][FT] round 6/6 epoch 3/5 | patches=168 | loss=0.6925
+[10:04:34][   3014s][FT] round 6/6 epoch 4/5 | patches=168 | loss=0.6733
+[10:04:45][   3025s][FT] round 6/6 epoch 5/5 | patches=168 | loss=0.6769
+[10:04:49][   3030s][AL] round 6/6 | labeled=168 | selection_seconds=35.9000 round_seconds=428.2000 pseudo_student_dice=0.6815 pseudo_ensemble_dice=0.5090 consistency_scale=0.6050 val_dice=0.7581 val_vi=0.5190 val_ece=0.0157
+[10:04:59][   3040s][CONS] final consolidation: pseudo-labels refreshed from the student
+[10:05:10][   3050s][FT] final 1/10 epoch 1/1 | patches=168 | loss=0.6609
+[10:05:20][   3061s][FT] final 2/10 epoch 1/1 | patches=168 | loss=0.6543
+[10:05:31][   3071s][FT] final 3/10 epoch 1/1 | patches=168 | loss=0.6464
+[10:05:42][   3082s][FT] final 4/10 epoch 1/1 | patches=168 | loss=0.6473
+[10:05:52][   3093s][FT] final 5/10 epoch 1/1 | patches=168 | loss=0.6407
+[10:06:03][   3103s][FT] final 6/10 epoch 1/1 | patches=168 | loss=0.6366
+[10:06:13][   3114s][FT] final 7/10 epoch 1/1 | patches=168 | loss=0.6301
+[10:06:24][   3124s][FT] final 8/10 epoch 1/1 | patches=168 | loss=0.6261
+[10:06:34][   3135s][FT] final 9/10 epoch 1/1 | patches=168 | loss=0.6188
+[10:06:45][   3145s][FT] final 10/10 epoch 1/1 | patches=168 | loss=0.6215
+[10:06:49][   3150s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=115.9000 val_dice=0.7586 val_vi=0.5180 val_ece=0.0218
+[10:06:49][   3150s][WEIGHTS] final: c-elegans=0.001, cerebellum=0.033, flywire=0.022, berghia=0.006, microns=0.934, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[10:07:24][   3184s][TEST] raw student (no EMA, no TTA): patch_dice=0.7503 patch_vi=0.5264
+[10:07:28][   3189s][TEST] final model without TTA: patch_dice=0.7503 patch_vi=0.5264
+[10:07:28][   3189s][RESOURCES] {"phases": {"setup": {"seconds": 1.48, "peak_cpu_rss_mb": 2191.8, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 63.32, "peak_cpu_rss_mb": 4181.8, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 649.83, "peak_cpu_rss_mb": 5156.0, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 2430.01, "peak_cpu_rss_mb": 5156.0, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4646.0, "rounds": 6}, "inference": {"seconds": 38.92, "peak_cpu_rss_mb": 5156.0, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4646.0, "patches": 735, "images": 15, "patches_per_second": 18.9, "seconds_per_image": 2.595}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5156.0, "total_seconds": 3183.6}
+[10:07:28][   3189s][TEST] patch_dice=0.7592 patch_iou=0.6215 patch_precision=0.7384 patch_recall=0.7841 patch_vi=0.5102 image_dice=0.8083 image_iou=0.6787 image_precision=0.7898 image_recall=0.8282 image_vi=0.5167 cal_ece=0.0124 cal_nll=0.1814 cal_brier=0.0572 cal_mean_confidence=0.9298 cal_accuracy=0.9174 cal_predicted_prevalence=0.2178 cal_true_prevalence=0.2059 cal_mean_probability=0.2169
+[10:07:29][   3189s][DONE] snemi3d macs-v11-norestart-acq-laplace b5 s2 | test patch Dice=0.7592 VI=0.5102 | image Dice=0.8083 VI=0.5167 | 3189s

macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json

@@ -0,0 +1,785 @@
+{
+  "target": "snemi3d",
+  "variant": "macs-v11-norestart-acq-laplace",
+  "budget": 5.0,
+  "seed": 2,
+  "config": {
+    "task": "membrane",
+    "target": "snemi3d",
+    "variant": "macs-v11-norestart-acq-laplace",
+    "budget": 5.0,
+    "seed": 2,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "snemi3d",
+      "--variants",
+      "macs-v11-norestart-acq-laplace",
+      "--budgets",
+      "5",
+      "--seeds",
+      "2",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.005265163985947525,
+    0.05775356661687405,
+    0.3937117170688725,
+    0.02742108315870487,
+    0.4770168284742763,
+    0.006835511403011292,
+    0.011200087973044965,
+    0.020796041319268533
+  ],
+  "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": [
+      0.005265163985947525,
+      0.05775356661687405,
+      0.3937117170688725,
+      0.02742108315870487,
+      0.4770168284742763,
+      0.006835511403011292,
+      0.011200087973044965,
+      0.020796041319268533
+    ],
+    "per_source": [
+      0.0030099407416697968,
+      0.30826237672231105,
+      0.030874093815323564,
+      0.005310674836551046,
+      0.6243418715179362,
+      0.02131301441017302,
+      0.0048036204646279375,
+      0.002084407491407262
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.048363233996627016,
+      "mmd2_mixture": 0.060122682064390254,
+      "mmd2_uniform": 0.15285688303099165,
+      "mmd2_single_source": [
+        0.4522663898122707,
+        0.14701298584193184,
+        0.12241687119492284,
+        0.2726521753211397,
+        0.08995170742180114,
+        0.3744254300178685,
+        0.33957846092746485,
+        0.23688038103280662
+      ],
+      "median_squared_distance": 0.18662990120119327,
+      "gammas": [
+        1.3395495490858758,
+        2.6790990981717515,
+        5.358198196343503,
+        10.716396392687006,
+        21.432792785374012
+      ],
+      "max_weight": 0.4770168284742763,
+      "entropy": 1.1759448067763238,
+      "effective_sources": 3.24120380896937,
+      "min_weight": 0.005265163985947525
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.4624401356050084,
+        0.12211205990556462,
+        0.29128492533416295,
+        0.4206953767024899,
+        0.07022529260621624,
+        0.3185315424880186,
+        0.42807307387367177,
+        0.489454149021297
+      ],
+      "standardized": [
+        0.9323134642607491,
+        -1.3822019864377946,
+        -0.23168506223542185,
+        0.6484142024470035,
+        -1.735075384714815,
+        -0.04638528568228213,
+        0.6985887114356643,
+        1.1160313409268967
+      ],
+      "tau": 0.5,
+      "max_weight": 0.6243418715179362,
+      "entropy": 0.9300659298676801,
+      "effective_sources": 2.534676282981126,
+      "min_weight": 0.002084407491407262
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "microns_basil",
+    "seconds": 6.7
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.41247125221353,
+      0.47321264983034456,
+      0.19639403256049165,
+      0.31013587127331976,
+      0.5894646206275136,
+      0.13744302537529526,
+      0.3054434345563225,
+      0.35012697230796425
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.4205061683646676,
+      "mixture@raw": 0.462918877571213,
+      "mixture@logit": 0.3665782427199844,
+      "per_source": 0.5892228765147073,
+      "per_source@raw": 0.5905348345941427,
+      "per_source@logit": 0.5781053080039771,
+      "uniform": 0.41124230184319877,
+      "uniform@raw": 0.4669572710990906,
+      "uniform@logit": 0.22782834015245554
+    },
+    "val_patches": 735
+  },
+  "temperatures": [
+    2.575598955154419,
+    4.557764530181885,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    2.642308473587036,
+    4.074584007263184
+  ],
+  "top_source": "microns_basil",
+  "weight_stats": {
+    "max_weight": 0.4770168284742763,
+    "entropy": 1.1759448067763238,
+    "effective_sources": 3.24120380896937,
+    "min_weight": 0.005265163985947525
+  },
+  "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": 649.8,
+  "round0_val": {
+    "val_dice": 0.38345661880027987,
+    "val_vi": 0.6425668875430072,
+    "val_ece": 0.07046950570811931
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 28,
+      "selection_seconds": 119.3,
+      "round_seconds": 560.4,
+      "teacher_dice_labeled": [
+        0.30923963036625823,
+        0.40141865169099034,
+        0.21298253591750957,
+        0.34862235687346954,
+        0.5216517007952947,
+        0.16310498252910807,
+        0.20307595354559912,
+        0.2304380164604082
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 232196,
+      "evidence_nonempty_patches": 28,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.4117972424112869,
+      "restart_teacher_dice": 0.5216517007952947,
+      "weights": [
+        0.0012112506663100886,
+        0.03339868283529253,
+        0.0345908486511191,
+        0.009352845080749347,
+        0.9180148960941229,
+        0.0003647022371434577,
+        0.000891211066458353,
+        0.0021755633688042347
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5209970401005555,
+        "uniform": 0.3803190293127032,
+        "top1": 0.5216517007952947,
+        "top2": 0.5224545359489741,
+        "top3": 0.5224050337353666
+      },
+      "val_dice": 0.7174647731793253,
+      "val_vi": 0.5862075152354411,
+      "val_ece": 0.01040988329108115
+    },
+    {
+      "round": 2,
+      "n_labeled": 56,
+      "selection_seconds": 37.0,
+      "round_seconds": 80.0,
+      "teacher_dice_labeled": [
+        0.32968286739321734,
+        0.3899514887912725,
+        0.1956341759576112,
+        0.29252135560018844,
+        0.4860608454924249,
+        0.12947791408419423,
+        0.21962028025532235,
+        0.2707968487830147
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 551979,
+      "evidence_nonempty_patches": 56,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0022035355669607175,
+        0.044160122494152035,
+        0.0,
+        0.0,
+        0.9536363419388872,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top3",
+      "candidate_dice": {
+        "mixture": 0.4822414733525932,
+        "uniform": 0.31097768281455124,
+        "top1": 0.4860608454924249,
+        "top2": 0.4872193497639249,
+        "top3": 0.4872618953534774
+      },
+      "pseudo_student_dice": 0.6088462828961946,
+      "pseudo_ensemble_dice": 0.45378946490070243,
+      "pseudo_source": "student",
+      "consistency_scale": 0.363,
+      "val_dice": 0.7390804504089595,
+      "val_vi": 0.5685614483112967,
+      "val_ece": 0.011165779298286696
+    },
+    {
+      "round": 3,
+      "n_labeled": 84,
+      "selection_seconds": 36.7,
+      "round_seconds": 393.4,
+      "teacher_dice_labeled": [
+        0.37086031388801677,
+        0.4453667316802009,
+        0.20723350644101518,
+        0.33534482250576847,
+        0.5762288381413482,
+        0.15528764228810432,
+        0.23703538953587494,
+        0.2977320132507578
+      ],
+      "teacher_dice_pool": [
+        0.3898740464376021,
+        0.4755429928578754,
+        0.23910686384852367,
+        0.3886655657749072,
+        0.668424476613459,
+        0.19800193940952124,
+        0.2762308341946166,
+        0.30910087439506295
+      ],
+      "evidence_fg_pixels": 913408,
+      "evidence_nonempty_patches": 84,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0013310497510279384,
+        0.03075660939664637,
+        0.019379319848684722,
+        0.004859875828552802,
+        0.9402000187080747,
+        0.00020013967092648373,
+        0.0007426933897470523,
+        0.002530293406339934
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5754038750517916,
+        "uniform": 0.36575840911571345,
+        "top1": 0.5762288381413482,
+        "top2": 0.5777510078628051,
+        "top3": 0.5778078801028542
+      },
+      "pseudo_student_dice": 0.6361807161340302,
+      "pseudo_ensemble_dice": 0.480113758300175,
+      "pseudo_source": "student",
+      "consistency_scale": 0.454,
+      "val_dice": 0.7495523539931895,
+      "val_vi": 0.5489239394094233,
+      "val_ece": 0.011783477941583785
+    },
+    {
+      "round": 4,
+      "n_labeled": 112,
+      "selection_seconds": 36.4,
+      "round_seconds": 404.4,
+      "teacher_dice_labeled": [
+        0.37321174857677786,
+        0.44278635107161046,
+        0.20425591711599464,
+        0.327772126479294,
+        0.5657931122865841,
+        0.1549070230922441,
+        0.24313737842154926,
+        0.29954958369556517
+      ],
+      "teacher_dice_pool": [
+        0.4001746244878322,
+        0.48220145867073366,
+        0.23499525796249365,
+        0.3786014216751024,
+        0.6490690617814941,
+        0.19743975469322234,
+        0.26903872285765806,
+        0.31888259615125375
+      ],
+      "evidence_fg_pixels": 1179147,
+      "evidence_nonempty_patches": 112,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0015053565082109753,
+        0.033110438433428105,
+        0.020779732351032795,
+        0.004977030440218871,
+        0.9356887305472527,
+        0.00022024857037611335,
+        0.0008720534096111227,
+        0.0028464097398693314
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5647967946865813,
+        "uniform": 0.360234486572068,
+        "top1": 0.5657931122865841,
+        "top2": 0.5673055047680104,
+        "top3": 0.5673725014471637
+      },
+      "pseudo_student_dice": 0.6648006350571386,
+      "pseudo_ensemble_dice": 0.4757541945192298,
+      "pseudo_source": "student",
+      "consistency_scale": 0.549,
+      "val_dice": 0.7550029429579439,
+      "val_vi": 0.5302964007298193,
+      "val_ece": 0.014572060405993651
+    },
+    {
+      "round": 5,
+      "n_labeled": 140,
+      "selection_seconds": 36.2,
+      "round_seconds": 417.5,
+      "teacher_dice_labeled": [
+        0.3625410938591591,
+        0.4355784897218964,
+        0.20264899618503762,
+        0.3295045734408726,
+        0.5588364198985812,
+        0.15605754418445766,
+        0.2401934049537906,
+        0.2913797382341989
+      ],
+      "teacher_dice_pool": [
+        0.3786420858421469,
+        0.47081271613508735,
+        0.24180065149346994,
+        0.3775880634035507,
+        0.6415367429953798,
+        0.19634054315397,
+        0.2792974265186966,
+        0.30249360274341297
+      ],
+      "evidence_fg_pixels": 1474014,
+      "evidence_nonempty_patches": 140,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0014483361718081902,
+        0.03297871007616184,
+        0.021889369773742826,
+        0.005420836341058693,
+        0.9343098877686563,
+        0.0002384963984002584,
+        0.0009064203503226087,
+        0.002807943119849266
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.557823971764361,
+        "uniform": 0.3578505232668598,
+        "top1": 0.5588364198985812,
+        "top2": 0.5602451647369182,
+        "top3": 0.5603049948006693
+      },
+      "pseudo_student_dice": 0.643506272736545,
+      "pseudo_ensemble_dice": 0.44930986227259595,
+      "pseudo_source": "student",
+      "consistency_scale": 0.478,
+      "val_dice": 0.7549671911288235,
+      "val_vi": 0.5302006794911862,
+      "val_ece": 0.013241020018444667
+    },
+    {
+      "round": 6,
+      "n_labeled": 168,
+      "selection_seconds": 35.9,
+      "round_seconds": 428.2,
+      "teacher_dice_labeled": [
+        0.36353511631327795,
+        0.43497302604973287,
+        0.20074485294527286,
+        0.3327935590239828,
+        0.5582466822186675,
+        0.1567819919136381,
+        0.23909282753480066,
+        0.2925195195546896
+      ],
+      "teacher_dice_pool": [
+        0.3839850647413449,
+        0.46831841356244397,
+        0.23537658668363787,
+        0.3826950837371619,
+        0.6349460257289151,
+        0.19188588782082197,
+        0.27303537783430637,
+        0.30658306254260204
+      ],
+      "evidence_fg_pixels": 1744619,
+      "evidence_nonempty_patches": 168,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0014714657544328564,
+        0.03297372159029093,
+        0.02160366658468838,
+        0.005635260229233188,
+        0.9343154793074689,
+        0.00024165281514750797,
+        0.0009018071986931946,
+        0.0028569465200449728
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.557289318573895,
+        "uniform": 0.3591946978714461,
+        "top1": 0.5582466822186675,
+        "top2": 0.5595871272722308,
+        "top3": 0.5596502336834291
+      },
+      "pseudo_student_dice": 0.6815412134626851,
+      "pseudo_ensemble_dice": 0.5090477033045854,
+      "pseudo_source": "student",
+      "consistency_scale": 0.605,
+      "val_dice": 0.7581036709465326,
+      "val_vi": 0.5190172698185712,
+      "val_ece": 0.01572609371733461
+    },
+    {
+      "round": 7,
+      "n_labeled": 168,
+      "selection_seconds": 0.0,
+      "round_seconds": 115.9,
+      "final_consolidation": true,
+      "val_dice": 0.7585554059155132,
+      "val_vi": 0.5180210703137048,
+      "val_ece": 0.0218110641494762
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.5072424045387947,
+    "auroc_top20_error": 0.8403184472456204
+  },
+  "al_seconds": 2430.0,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.0014714657544328564,
+    0.03297372159029093,
+    0.02160366658468838,
+    0.005635260229233188,
+    0.9343154793074689,
+    0.00024165281514750797,
+    0.0009018071986931946,
+    0.0028569465200449728
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9343154793074689,
+    "entropy": 0.322684680800098,
+    "effective_sources": 1.3808298800800942,
+    "min_weight": 0.00024165281514750797
+  },
+  "top_source_final": "microns_basil",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.7502993844021068,
+    "patch_iou": 0.610265425863319,
+    "patch_precision": 0.726259786758238,
+    "patch_recall": 0.779129192487549,
+    "patch_vi": 0.5263768898094677,
+    "n_patches": 735,
+    "cal_ece": 0.020970448656017486,
+    "cal_nll": 0.19405148446924358,
+    "cal_brier": 0.060640539511800845,
+    "cal_mean_confidence": 0.934816377652616,
+    "cal_accuracy": 0.9138459289965987,
+    "cal_predicted_prevalence": 0.2200547406462585,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.21815898913369627
+  },
+  "test_notta": {
+    "patch_dice": 0.7502993844021068,
+    "patch_iou": 0.610265425863319,
+    "patch_precision": 0.726259786758238,
+    "patch_recall": 0.779129192487549,
+    "patch_vi": 0.5263768898094677,
+    "n_patches": 735,
+    "cal_ece": 0.020970448656017486,
+    "cal_nll": 0.19405148446924358,
+    "cal_brier": 0.060640539511800845,
+    "cal_mean_confidence": 0.934816377652616,
+    "cal_accuracy": 0.9138459289965987,
+    "cal_predicted_prevalence": 0.2200547406462585,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.21815898913369627
+  },
+  "test": {
+    "patch_dice": 0.7591598943652398,
+    "patch_iou": 0.6215398307814628,
+    "patch_precision": 0.7383584817490859,
+    "patch_recall": 0.7840826714217108,
+    "patch_vi": 0.5102328361065348,
+    "n_patches": 735,
+    "image_dice": 0.8083227144183348,
+    "image_iou": 0.6787329371214111,
+    "image_precision": 0.7897767967423324,
+    "image_recall": 0.828165408514625,
+    "image_vi": 0.5166640790459083,
+    "n_images": 15,
+    "cal_ece": 0.012392638944076592,
+    "cal_nll": 0.1813798769940655,
+    "cal_brier": 0.05722207631886344,
+    "cal_mean_confidence": 0.9298206257638044,
+    "cal_accuracy": 0.9174279868197279,
+    "cal_predicted_prevalence": 0.21780665391156462,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.21693074734329712
+  },
+  "eval_seconds": 38.9,
+  "total_seconds": 3188.7,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 1.48,
+        "peak_cpu_rss_mb": 2191.8,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 63.32,
+        "peak_cpu_rss_mb": 4181.8,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 649.83,
+        "peak_cpu_rss_mb": 5156.0,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4638.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2430.01,
+        "peak_cpu_rss_mb": 5156.0,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4646.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 38.92,
+        "peak_cpu_rss_mb": 5156.0,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4646.0,
+        "patches": 735,
+        "images": 15,
+        "patches_per_second": 18.9,
+        "seconds_per_image": 2.595
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5156.0,
+    "total_seconds": 3183.6
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/snemi3d/macs-v11-norestart-acq-laplace_b5_s2.pth"
+}

macs_v2_r26/runs/membrane/snemi3d/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
+1,28,119.3,560.4,"[0.30923963036625823, 0.40141865169099034, 0.21298253591750957, 0.34862235687346954, 0.5216517007952947, 0.16310498252910807, 0.20307595354559912, 0.2304380164604082]",,232196,28,True,kept,0.4117972424112869,0.5216517007952947,"[0.0012112506663100886, 0.03339868283529253, 0.0345908486511191, 0.009352845080749347, 0.9180148960941229, 0.0003647022371434577, 0.000891211066458353, 0.0021755633688042347]",mixture,"{'mixture': 0.5209970401005555, 'uniform': 0.3803190293127032, 'top1': 0.5216517007952947, 'top2': 0.5224545359489741, 'top3': 0.5224050337353666}",0.7174647731793253,0.5862075152354411,0.01040988329108115,,,,,
+2,56,37.0,80.0,"[0.32968286739321734, 0.3899514887912725, 0.1956341759576112, 0.29252135560018844, 0.4860608454924249, 0.12947791408419423, 0.21962028025532235, 0.2707968487830147]",,551979,56,False,,,,"[0.0022035355669607175, 0.044160122494152035, 0.0, 0.0, 0.9536363419388872, 0.0, 0.0, 0.0]",top3,"{'mixture': 0.4822414733525932, 'uniform': 0.31097768281455124, 'top1': 0.4860608454924249, 'top2': 0.4872193497639249, 'top3': 0.4872618953534774}",0.7390804504089595,0.5685614483112967,0.011165779298286696,0.6088462828961946,0.45378946490070243,student,0.363,
+3,84,36.7,393.4,"[0.37086031388801677, 0.4453667316802009, 0.20723350644101518, 0.33534482250576847, 0.5762288381413482, 0.15528764228810432, 0.23703538953587494, 0.2977320132507578]","[0.3898740464376021, 0.4755429928578754, 0.23910686384852367, 0.3886655657749072, 0.668424476613459, 0.19800193940952124, 0.2762308341946166, 0.30910087439506295]",913408,84,False,,,,"[0.0013310497510279384, 0.03075660939664637, 0.019379319848684722, 0.004859875828552802, 0.9402000187080747, 0.00020013967092648373, 0.0007426933897470523, 0.002530293406339934]",mixture,"{'mixture': 0.5754038750517916, 'uniform': 0.36575840911571345, 'top1': 0.5762288381413482, 'top2': 0.5777510078628051, 'top3': 0.5778078801028542}",0.7495523539931895,0.5489239394094233,0.011783477941583785,0.6361807161340302,0.480113758300175,student,0.454,
+4,112,36.4,404.4,"[0.37321174857677786, 0.44278635107161046, 0.20425591711599464, 0.327772126479294, 0.5657931122865841, 0.1549070230922441, 0.24313737842154926, 0.29954958369556517]","[0.4001746244878322, 0.48220145867073366, 0.23499525796249365, 0.3786014216751024, 0.6490690617814941, 0.19743975469322234, 0.26903872285765806, 0.31888259615125375]",1179147,112,False,,,,"[0.0015053565082109753, 0.033110438433428105, 0.020779732351032795, 0.004977030440218871, 0.9356887305472527, 0.00022024857037611335, 0.0008720534096111227, 0.0028464097398693314]",mixture,"{'mixture': 0.5647967946865813, 'uniform': 0.360234486572068, 'top1': 0.5657931122865841, 'top2': 0.5673055047680104, 'top3': 0.5673725014471637}",0.7550029429579439,0.5302964007298193,0.014572060405993651,0.6648006350571386,0.4757541945192298,student,0.549,
+5,140,36.2,417.5,"[0.3625410938591591, 0.4355784897218964, 0.20264899618503762, 0.3295045734408726, 0.5588364198985812, 0.15605754418445766, 0.2401934049537906, 0.2913797382341989]","[0.3786420858421469, 0.47081271613508735, 0.24180065149346994, 0.3775880634035507, 0.6415367429953798, 0.19634054315397, 0.2792974265186966, 0.30249360274341297]",1474014,140,False,,,,"[0.0014483361718081902, 0.03297871007616184, 0.021889369773742826, 0.005420836341058693, 0.9343098877686563, 0.0002384963984002584, 0.0009064203503226087, 0.002807943119849266]",mixture,"{'mixture': 0.557823971764361, 'uniform': 0.3578505232668598, 'top1': 0.5588364198985812, 'top2': 0.5602451647369182, 'top3': 0.5603049948006693}",0.7549671911288235,0.5302006794911862,0.013241020018444667,0.643506272736545,0.44930986227259595,student,0.478,
+6,168,35.9,428.2,"[0.36353511631327795, 0.43497302604973287, 0.20074485294527286, 0.3327935590239828, 0.5582466822186675, 0.1567819919136381, 0.23909282753480066, 0.2925195195546896]","[0.3839850647413449, 0.46831841356244397, 0.23537658668363787, 0.3826950837371619, 0.6349460257289151, 0.19188588782082197, 0.27303537783430637, 0.30658306254260204]",1744619,168,False,,,,"[0.0014714657544328564, 0.03297372159029093, 0.02160366658468838, 0.005635260229233188, 0.9343154793074689, 0.00024165281514750797, 0.0009018071986931946, 0.0028569465200449728]",mixture,"{'mixture': 0.557289318573895, 'uniform': 0.3591946978714461, 'top1': 0.5582466822186675, 'top2': 0.5595871272722308, 'top3': 0.5596502336834291}",0.7581036709465326,0.5190172698185712,0.01572609371733461,0.6815412134626851,0.5090477033045854,student,0.605,
+7,168,0.0,115.9,,,,,,,,,,,,0.7585554059155132,0.5180210703137048,0.0218110641494762,,,,,True

macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-1/log.txt

@@ -0,0 +1,103 @@
+[06:00:54][      0s][START] target=snemi3d variant=macs-v11-norestart-acq-margin budget=5.0% seed=1 | {"task": "membrane", "target": "snemi3d", "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": 8, "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}
+[06:00:54][      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.
+[06:00:54][      0s][DATA] snemi3d: train=3430 val=735 test=735 patches
+[06:00:56][      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=4.07
+[06:00:57][      3s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s
+[06:01:03][      9s][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
+[06:01:03][      9s][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
+[06:01:56][     62s][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
+[06:01:56][     62s][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
+[06:01:56][     62s][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
+[06:01:56][     62s][INIT] student initialised from teacher microns (weight 0.477)
+[06:01:59][     66s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[06:01:59][     66s][KD] distilling the mixture ensemble (prob space) into the student
+[06:04:19][    205s][KD] soft labels for 3430 patches from 8 teachers in 140s
+[06:06:01][    307s][KD] epoch 1/5 | patches=3430 | loss=0.4461 | 102s
+[06:07:43][    409s][KD] epoch 2/5 | patches=3430 | loss=0.3728 | 102s
+[06:09:25][    511s][KD] epoch 3/5 | patches=3430 | loss=0.3563 | 102s
+[06:11:07][    613s][KD] epoch 4/5 | patches=3430 | loss=0.3506 | 102s
+[06:12:49][    715s][KD] epoch 5/5 | patches=3430 | loss=0.3475 | 102s
+[06:12:53][    719s][ROUND0] before any target label: val_dice=0.3968 val_vi=0.6469
+[06:12:53][    719s][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
+[06:13:11][    737s][AL] round-1 uncertainty quality on 3430 patches: {'spearman_uncertainty_error': 0.3088075353000357, 'auroc_top20_error': 0.7218027777541671}
+[06:14:33][    819s][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
+[06:14:38][    824s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.543, uniform=0.427, top1=0.543, top2=0.545, top3=0.545 -> using mixture
+[06:14:39][    825s][WEIGHTS] round 1: student kept (labeled Dice 0.433 vs best teacher microns 0.543); distilling the corrected ensemble into it
+[06:14:39][    825s][WEIGHTS] round 1: teacher Dice on 28 labeled patches + 0 blank (279577 fg px): c-elegans=0.358, cerebellum=0.442, flywire=0.213, berghia=0.348, microns=0.543, oct-glia=0.243, oct-sfl=0.282, mouse-brain=0.275 | weights: c-elegans=0.002, cerebellum=0.040, flywire=0.028, berghia=0.008, microns=0.917, oct-glia=0.001, oct-sfl=0.002, mouse-brain=0.003
+[06:16:57][    964s][KD] soft labels for 3402 patches from 8 teachers in 139s
+[06:18:38][   1065s][KD] epoch 1/2 | patches=3402 | loss=0.2525 | 101s
+[06:20:19][   1166s][KD] epoch 2/2 | patches=3402 | loss=0.2504 | 101s
+[06:21:41][   1247s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 82s
+[06:21:44][   1250s][FT] round 1/6 epoch 1/5 | patches=28 | loss=1.0453
+[06:21:46][   1252s][FT] round 1/6 epoch 2/5 | patches=28 | loss=0.9823
+[06:21:49][   1255s][FT] round 1/6 epoch 3/5 | patches=28 | loss=0.8996
+[06:21:51][   1257s][FT] round 1/6 epoch 4/5 | patches=28 | loss=0.8659
+[06:21:54][   1260s][FT] round 1/6 epoch 5/5 | patches=28 | loss=0.8378
+[06:21:58][   1264s][AL] round 1/6 | labeled=28 | selection_seconds=100.1000 round_seconds=540.4000 restart_student_dice=0.4330 restart_teacher_dice=0.5428 val_dice=0.7295 val_vi=0.5884 val_ece=0.0197
+[06:22:25][   1291s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.413, uniform=0.307, top1=0.416, top2=0.422, top3=0.422 -> using top3
+[06:22:25][   1291s][WEIGHTS] round 2: teacher Dice on 56 labeled patches + 0 blank (566444 fg px): c-elegans=0.321, cerebellum=0.413, flywire=0.204, berghia=0.296, microns=0.416, oct-glia=0.179, oct-sfl=0.203, mouse-brain=0.273 | weights: c-elegans=0.004, cerebellum=0.104, flywire=0.000, berghia=0.000, microns=0.892, oct-glia=0.000, oct-sfl=0.000, mouse-brain=0.000
+[06:22:39][   1305s][CONS] round 2: on the 28 new patches student Dice=0.611 vs ensemble 0.235 -> pseudo-labels from the student | consistency x0.37
+[06:22:43][   1309s][FT] round 2/6 epoch 1/5 | patches=56 | loss=0.8161
+[06:22:47][   1313s][FT] round 2/6 epoch 2/5 | patches=56 | loss=0.7693
+[06:22:50][   1317s][FT] round 2/6 epoch 3/5 | patches=56 | loss=0.7324
+[06:22:54][   1321s][FT] round 2/6 epoch 4/5 | patches=56 | loss=0.7172
+[06:22:58][   1325s][FT] round 2/6 epoch 5/5 | patches=56 | loss=0.7192
+[06:23:03][   1329s][AL] round 2/6 | labeled=56 | selection_seconds=18.1000 round_seconds=60.7000 pseudo_student_dice=0.6111 pseudo_ensemble_dice=0.2345 consistency_scale=0.3700 val_dice=0.7433 val_vi=0.5467 val_ece=0.0149
+[06:26:02][   1508s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.390, cerebellum=0.476, flywire=0.236, berghia=0.386, microns=0.659, oct-glia=0.187, oct-sfl=0.263, mouse-brain=0.311 | blended with labeled Dice at 0.50
+[06:28:33][   1660s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.539, uniform=0.366, top1=0.540, top2=0.542, top3=0.542 -> using mixture
+[06:28:33][   1660s][WEIGHTS] round 3: teacher Dice on 84 labeled patches + 0 blank (865230 fg px): c-elegans=0.363, cerebellum=0.437, flywire=0.211, berghia=0.339, microns=0.540, oct-glia=0.176, oct-sfl=0.225, mouse-brain=0.293 | weights: c-elegans=0.002, cerebellum=0.040, flywire=0.028, berghia=0.007, microns=0.919, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[06:28:47][   1673s][CONS] round 3: on the 28 new patches student Dice=0.625 vs ensemble 0.429 -> pseudo-labels from the student | consistency x0.42
+[06:28:53][   1679s][FT] round 3/6 epoch 1/5 | patches=84 | loss=0.7737
+[06:28:59][   1685s][FT] round 3/6 epoch 2/5 | patches=84 | loss=0.7402
+[06:29:04][   1690s][FT] round 3/6 epoch 3/5 | patches=84 | loss=0.7220
+[06:29:11][   1697s][FT] round 3/6 epoch 4/5 | patches=84 | loss=0.7066
+[06:29:16][   1702s][FT] round 3/6 epoch 5/5 | patches=84 | loss=0.6914
+[06:29:21][   1707s][AL] round 3/6 | labeled=84 | selection_seconds=18.0000 round_seconds=373.7000 pseudo_student_dice=0.6252 pseudo_ensemble_dice=0.4286 consistency_scale=0.4170 val_dice=0.7527 val_vi=0.5420 val_ece=0.0168
+[06:32:22][   1888s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.408, cerebellum=0.489, flywire=0.238, berghia=0.386, microns=0.657, oct-glia=0.193, oct-sfl=0.272, mouse-brain=0.322 | blended with labeled Dice at 0.50
+[06:34:55][   2041s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.549, uniform=0.372, top1=0.550, top2=0.551, top3=0.551 -> using mixture
+[06:34:55][   2041s][WEIGHTS] round 4: teacher Dice on 112 labeled patches + 0 blank (1141567 fg px): c-elegans=0.375, cerebellum=0.441, flywire=0.210, berghia=0.339, microns=0.550, oct-glia=0.175, oct-sfl=0.235, mouse-brain=0.299 | weights: c-elegans=0.002, cerebellum=0.038, flywire=0.025, berghia=0.006, microns=0.924, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[06:35:09][   2055s][CONS] round 4: on the 28 new patches student Dice=0.663 vs ensemble 0.499 -> pseudo-labels from the student | consistency x0.54
+[06:35:16][   2062s][FT] round 4/6 epoch 1/5 | patches=112 | loss=0.7368
+[06:35:23][   2069s][FT] round 4/6 epoch 2/5 | patches=112 | loss=0.7219
+[06:35:31][   2077s][FT] round 4/6 epoch 3/5 | patches=112 | loss=0.7088
+[06:35:38][   2084s][FT] round 4/6 epoch 4/5 | patches=112 | loss=0.6966
+[06:35:45][   2091s][FT] round 4/6 epoch 5/5 | patches=112 | loss=0.6849
+[06:35:49][   2096s][AL] round 4/6 | labeled=112 | selection_seconds=17.8000 round_seconds=384.6000 pseudo_student_dice=0.6629 pseudo_ensemble_dice=0.4986 consistency_scale=0.5430 val_dice=0.7510 val_vi=0.5316 val_ece=0.0170
+[06:38:52][   2279s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.382, cerebellum=0.468, flywire=0.238, berghia=0.375, microns=0.650, oct-glia=0.198, oct-sfl=0.290, mouse-brain=0.304 | blended with labeled Dice at 0.50
+[06:41:28][   2434s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.553, uniform=0.381, top1=0.554, top2=0.555, top3=0.555 -> using mixture
+[06:41:28][   2434s][WEIGHTS] round 5: teacher Dice on 140 labeled patches + 0 blank (1422513 fg px): c-elegans=0.364, cerebellum=0.434, flywire=0.211, berghia=0.335, microns=0.554, oct-glia=0.178, oct-sfl=0.249, mouse-brain=0.291 | weights: c-elegans=0.002, cerebellum=0.034, flywire=0.025, berghia=0.006, microns=0.929, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[06:41:42][   2448s][CONS] round 5: on the 28 new patches student Dice=0.661 vs ensemble 0.511 -> pseudo-labels from the student | consistency x0.54
+[06:41:51][   2457s][FT] round 5/6 epoch 1/5 | patches=140 | loss=0.6910
+[06:42:00][   2466s][FT] round 5/6 epoch 2/5 | patches=140 | loss=0.6891
+[06:42:09][   2475s][FT] round 5/6 epoch 3/5 | patches=140 | loss=0.6933
+[06:42:18][   2484s][FT] round 5/6 epoch 4/5 | patches=140 | loss=0.6768
+[06:42:27][   2493s][FT] round 5/6 epoch 5/5 | patches=140 | loss=0.6743
+[06:42:31][   2497s][AL] round 5/6 | labeled=140 | selection_seconds=17.7000 round_seconds=397.3000 pseudo_student_dice=0.6609 pseudo_ensemble_dice=0.5111 consistency_scale=0.5360 val_dice=0.7545 val_vi=0.5338 val_ece=0.0179
+[06:45:36][   2682s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.394, cerebellum=0.475, flywire=0.242, berghia=0.381, microns=0.646, oct-glia=0.202, oct-sfl=0.277, mouse-brain=0.310 | blended with labeled Dice at 0.50
+[06:48:13][   2839s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.552, uniform=0.374, top1=0.553, top2=0.554, top3=0.554 -> using mixture
+[06:48:13][   2839s][WEIGHTS] round 6: teacher Dice on 168 labeled patches + 0 blank (1683796 fg px): c-elegans=0.368, cerebellum=0.434, flywire=0.209, berghia=0.336, microns=0.553, oct-glia=0.177, oct-sfl=0.237, mouse-brain=0.292 | weights: c-elegans=0.002, cerebellum=0.034, flywire=0.025, berghia=0.006, microns=0.930, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[06:48:26][   2853s][CONS] round 6: on the 28 new patches student Dice=0.643 vs ensemble 0.475 -> pseudo-labels from the student | consistency x0.48
+[06:48:37][   2863s][FT] round 6/6 epoch 1/5 | patches=168 | loss=0.6883
+[06:48:47][   2874s][FT] round 6/6 epoch 2/5 | patches=168 | loss=0.6804
+[06:48:58][   2884s][FT] round 6/6 epoch 3/5 | patches=168 | loss=0.6762
+[06:49:08][   2895s][FT] round 6/6 epoch 4/5 | patches=168 | loss=0.6747
+[06:49:19][   2905s][FT] round 6/6 epoch 5/5 | patches=168 | loss=0.6631
+[06:49:23][   2909s][AL] round 6/6 | labeled=168 | selection_seconds=17.5000 round_seconds=408.0000 pseudo_student_dice=0.6425 pseudo_ensemble_dice=0.4749 consistency_scale=0.4750 val_dice=0.7557 val_vi=0.5299 val_ece=0.0180
+[06:49:33][   2920s][CONS] final consolidation: pseudo-labels refreshed from the student
+[06:49:44][   2930s][FT] final 1/10 epoch 1/1 | patches=168 | loss=0.6598
+[06:49:54][   2940s][FT] final 2/10 epoch 1/1 | patches=168 | loss=0.6540
+[06:50:05][   2951s][FT] final 3/10 epoch 1/1 | patches=168 | loss=0.6513
+[06:50:15][   2962s][FT] final 4/10 epoch 1/1 | patches=168 | loss=0.6429
+[06:50:26][   2972s][FT] final 5/10 epoch 1/1 | patches=168 | loss=0.6299
+[06:50:37][   2983s][FT] final 6/10 epoch 1/1 | patches=168 | loss=0.6251
+[06:50:47][   2993s][FT] final 7/10 epoch 1/1 | patches=168 | loss=0.6197
+[06:50:58][   3004s][FT] final 8/10 epoch 1/1 | patches=168 | loss=0.6161
+[06:51:08][   3014s][FT] final 9/10 epoch 1/1 | patches=168 | loss=0.6184
+[06:51:19][   3025s][FT] final 10/10 epoch 1/1 | patches=168 | loss=0.6177
+[06:51:23][   3029s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=115.3000 val_dice=0.7554 val_vi=0.5248 val_ece=0.0231
+[06:51:23][   3029s][WEIGHTS] final: c-elegans=0.002, cerebellum=0.034, flywire=0.025, berghia=0.006, microns=0.930, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[06:51:57][   3064s][TEST] raw student (no EMA, no TTA): patch_dice=0.7491 patch_vi=0.5299
+[06:52:02][   3068s][TEST] final model without TTA: patch_dice=0.7491 patch_vi=0.5299
+[06:52:02][   3068s][RESOURCES] {"phases": {"setup": {"seconds": 1.48, "peak_cpu_rss_mb": 2192.2, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 63.24, "peak_cpu_rss_mb": 4183.8, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 649.47, "peak_cpu_rss_mb": 5161.3, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 2309.76, "peak_cpu_rss_mb": 5161.3, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4646.0, "rounds": 6}, "inference": {"seconds": 38.77, "peak_cpu_rss_mb": 5161.3, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4646.0, "patches": 735, "images": 15, "patches_per_second": 19.0, "seconds_per_image": 2.585}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5161.3, "total_seconds": 3062.7}
+[06:52:02][   3068s][TEST] patch_dice=0.7606 patch_iou=0.6231 patch_precision=0.7390 patch_recall=0.7867 patch_vi=0.5100 image_dice=0.8081 image_iou=0.6785 image_precision=0.7903 image_recall=0.8273 image_vi=0.5165 cal_ece=0.0104 cal_nll=0.1794 cal_brier=0.0567 cal_mean_confidence=0.9284 cal_accuracy=0.9180 cal_predicted_prevalence=0.2157 cal_true_prevalence=0.2059 cal_mean_probability=0.2157
+[06:52:02][   3068s][DONE] snemi3d macs-v11-norestart-acq-margin b5 s1 | test patch Dice=0.7606 VI=0.5100 | image Dice=0.8081 VI=0.5165 | 3068s

macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-margin/budget-5/seed-1/result.json

@@ -0,0 +1,785 @@
+{
+  "target": "snemi3d",
+  "variant": "macs-v11-norestart-acq-margin",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "membrane",
+    "target": "snemi3d",
+    "variant": "macs-v11-norestart-acq-margin",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "snemi3d",
+      "--variants",
+      "macs-v11-norestart-acq-margin",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.005265163985947525,
+    0.05775356661687405,
+    0.3937117170688725,
+    0.02742108315870487,
+    0.4770168284742763,
+    0.006835511403011292,
+    0.011200087973044965,
+    0.020796041319268533
+  ],
+  "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": [
+      0.005265163985947525,
+      0.05775356661687405,
+      0.3937117170688725,
+      0.02742108315870487,
+      0.4770168284742763,
+      0.006835511403011292,
+      0.011200087973044965,
+      0.020796041319268533
+    ],
+    "per_source": [
+      0.0030099407416697968,
+      0.30826237672231105,
+      0.030874093815323564,
+      0.005310674836551046,
+      0.6243418715179362,
+      0.02131301441017302,
+      0.0048036204646279375,
+      0.002084407491407262
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.048363233996627016,
+      "mmd2_mixture": 0.060122682064390254,
+      "mmd2_uniform": 0.15285688303099165,
+      "mmd2_single_source": [
+        0.4522663898122707,
+        0.14701298584193184,
+        0.12241687119492284,
+        0.2726521753211397,
+        0.08995170742180114,
+        0.3744254300178685,
+        0.33957846092746485,
+        0.23688038103280662
+      ],
+      "median_squared_distance": 0.18662990120119327,
+      "gammas": [
+        1.3395495490858758,
+        2.6790990981717515,
+        5.358198196343503,
+        10.716396392687006,
+        21.432792785374012
+      ],
+      "max_weight": 0.4770168284742763,
+      "entropy": 1.1759448067763238,
+      "effective_sources": 3.24120380896937,
+      "min_weight": 0.005265163985947525
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.4624401356050084,
+        0.12211205990556462,
+        0.29128492533416295,
+        0.4206953767024899,
+        0.07022529260621624,
+        0.3185315424880186,
+        0.42807307387367177,
+        0.489454149021297
+      ],
+      "standardized": [
+        0.9323134642607491,
+        -1.3822019864377946,
+        -0.23168506223542185,
+        0.6484142024470035,
+        -1.735075384714815,
+        -0.04638528568228213,
+        0.6985887114356643,
+        1.1160313409268967
+      ],
+      "tau": 0.5,
+      "max_weight": 0.6243418715179362,
+      "entropy": 0.9300659298676801,
+      "effective_sources": 2.534676282981126,
+      "min_weight": 0.002084407491407262
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "microns_basil",
+    "seconds": 6.7
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.41247125221353,
+      0.47321264983034456,
+      0.19639403256049165,
+      0.31013587127331976,
+      0.5894646206275136,
+      0.13744302537529526,
+      0.3054434345563225,
+      0.35012697230796425
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.4205061683646676,
+      "mixture@raw": 0.462918877571213,
+      "mixture@logit": 0.3665782427199844,
+      "per_source": 0.5892228765147073,
+      "per_source@raw": 0.5905348345941427,
+      "per_source@logit": 0.5781053080039771,
+      "uniform": 0.41124230184319877,
+      "uniform@raw": 0.4669572710990906,
+      "uniform@logit": 0.22782834015245554
+    },
+    "val_patches": 735
+  },
+  "temperatures": [
+    2.575598955154419,
+    4.557764530181885,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    2.642308473587036,
+    4.074584007263184
+  ],
+  "top_source": "microns_basil",
+  "weight_stats": {
+    "max_weight": 0.4770168284742763,
+    "entropy": 1.1759448067763238,
+    "effective_sources": 3.24120380896937,
+    "min_weight": 0.005265163985947525
+  },
+  "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": 649.5,
+  "round0_val": {
+    "val_dice": 0.39676821516355265,
+    "val_vi": 0.6468693296003585,
+    "val_ece": 0.07077332120435081
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 28,
+      "selection_seconds": 100.1,
+      "round_seconds": 540.4,
+      "teacher_dice_labeled": [
+        0.35791764681874827,
+        0.4417052610943664,
+        0.21306314427094966,
+        0.3478407216013448,
+        0.5427570406835327,
+        0.2427000034404966,
+        0.2815004375801934,
+        0.2748319195941523
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 279577,
+      "evidence_nonempty_patches": 28,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.4329726063476451,
+      "restart_teacher_dice": 0.5427570406835327,
+      "weights": [
+        0.0015948525493022832,
+        0.04043639495708322,
+        0.0280150797666998,
+        0.007509823162345696,
+        0.9174652491168103,
+        0.0006541777664380059,
+        0.001579990027762249,
+        0.002744432653558335
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.543427983950943,
+        "uniform": 0.42744369108351205,
+        "top1": 0.5427570406835327,
+        "top2": 0.5446170820822631,
+        "top3": 0.5446748258576916
+      },
+      "val_dice": 0.7295256186222652,
+      "val_vi": 0.5884044823108097,
+      "val_ece": 0.019701315203793958
+    },
+    {
+      "round": 2,
+      "n_labeled": 56,
+      "selection_seconds": 18.1,
+      "round_seconds": 60.7,
+      "teacher_dice_labeled": [
+        0.3208247601267692,
+        0.41290094315652537,
+        0.2040200567361768,
+        0.2958021771680537,
+        0.41646418759010906,
+        0.1794556776225178,
+        0.20312701542316977,
+        0.2733950553226702
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 566444,
+      "evidence_nonempty_patches": 56,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.003783436799557813,
+        0.10421618553102502,
+        0.0,
+        0.0,
+        0.8920003776694172,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top3",
+      "candidate_dice": {
+        "mixture": 0.4129965036621742,
+        "uniform": 0.30746055861849897,
+        "top1": 0.41646418759010906,
+        "top2": 0.4222413401292082,
+        "top3": 0.4224928778934572
+      },
+      "pseudo_student_dice": 0.6110748083498765,
+      "pseudo_ensemble_dice": 0.23453104616733697,
+      "pseudo_source": "student",
+      "consistency_scale": 0.37,
+      "val_dice": 0.7432631786091862,
+      "val_vi": 0.5466669819950293,
+      "val_ece": 0.01494093341048987
+    },
+    {
+      "round": 3,
+      "n_labeled": 84,
+      "selection_seconds": 18.0,
+      "round_seconds": 373.7,
+      "teacher_dice_labeled": [
+        0.36273121427086064,
+        0.4372980514434087,
+        0.21105590198097918,
+        0.3393942140665256,
+        0.540192314721962,
+        0.17578747616669918,
+        0.22484020938853538,
+        0.29315816733536
+      ],
+      "teacher_dice_pool": [
+        0.39034661919349994,
+        0.4761150428010431,
+        0.23573876902414095,
+        0.38585503964914447,
+        0.659109861832459,
+        0.18712604425893262,
+        0.2631012063047301,
+        0.3108627311160762
+      ],
+      "evidence_fg_pixels": 865230,
+      "evidence_nonempty_patches": 84,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0017190749968959006,
+        0.03974672870367697,
+        0.02820614601955311,
+        0.007089510065337752,
+        0.9185871992481591,
+        0.0003441657073025338,
+        0.0009209823989749294,
+        0.0033861928600996084
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5393467185906318,
+        "uniform": 0.3655772808350024,
+        "top1": 0.540192314721962,
+        "top2": 0.541732206975406,
+        "top3": 0.541822389111206
+      },
+      "pseudo_student_dice": 0.6251532581237091,
+      "pseudo_ensemble_dice": 0.4286494839036211,
+      "pseudo_source": "student",
+      "consistency_scale": 0.417,
+      "val_dice": 0.7527323068124526,
+      "val_vi": 0.541985946096611,
+      "val_ece": 0.01677360838441982
+    },
+    {
+      "round": 4,
+      "n_labeled": 112,
+      "selection_seconds": 17.8,
+      "round_seconds": 384.6,
+      "teacher_dice_labeled": [
+        0.3749811196798546,
+        0.44113104576281453,
+        0.20960975570483995,
+        0.3387416433056273,
+        0.5495413048885877,
+        0.17535012944458866,
+        0.23520326785288623,
+        0.2988056113727459
+      ],
+      "teacher_dice_pool": [
+        0.40815563458449466,
+        0.4890113030690118,
+        0.23786569305882985,
+        0.3857946049443244,
+        0.6569114664585771,
+        0.1934436301469982,
+        0.2718589618428649,
+        0.3221797503419022
+      ],
+      "evidence_fg_pixels": 1141567,
+      "evidence_nonempty_patches": 112,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.001779979044505536,
+        0.037832694697906344,
+        0.02546728326344037,
+        0.006452106793159957,
+        0.9239361080593134,
+        0.00031389733227876727,
+        0.0009357868195762976,
+        0.003282143989819337
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5488637066017804,
+        "uniform": 0.37210821529489313,
+        "top1": 0.5495413048885877,
+        "top2": 0.5509644238645611,
+        "top3": 0.5510551274892082
+      },
+      "pseudo_student_dice": 0.6628713841310659,
+      "pseudo_ensemble_dice": 0.49864015261866945,
+      "pseudo_source": "student",
+      "consistency_scale": 0.543,
+      "val_dice": 0.7509571821136031,
+      "val_vi": 0.5315801650431149,
+      "val_ece": 0.01703855312358479
+    },
+    {
+      "round": 5,
+      "n_labeled": 140,
+      "selection_seconds": 17.7,
+      "round_seconds": 397.3,
+      "teacher_dice_labeled": [
+        0.3642759867907599,
+        0.43358117185195766,
+        0.21068374327687786,
+        0.33467724413782185,
+        0.5535326066210269,
+        0.17798854274162795,
+        0.24864286910367464,
+        0.2906701335016776
+      ],
+      "teacher_dice_pool": [
+        0.38218506328160307,
+        0.4682752407941175,
+        0.23816118972250283,
+        0.3750151946437794,
+        0.6498135108202829,
+        0.19764850889095936,
+        0.2903312054949055,
+        0.30404861986044174
+      ],
+      "evidence_fg_pixels": 1422513,
+      "evidence_nonempty_patches": 140,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.001545796995137486,
+        0.03390845176712642,
+        0.02488148279719892,
+        0.005987967802871813,
+        0.9294056619302986,
+        0.0003115125004738827,
+        0.0010346030187799481,
+        0.002924523188112882
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5528844985001343,
+        "uniform": 0.3808982550980504,
+        "top1": 0.5535326066210269,
+        "top2": 0.5547303978453834,
+        "top3": 0.554809582158575
+      },
+      "pseudo_student_dice": 0.6609389649034607,
+      "pseudo_ensemble_dice": 0.5110580295086186,
+      "pseudo_source": "student",
+      "consistency_scale": 0.536,
+      "val_dice": 0.7545000122039486,
+      "val_vi": 0.533773990628012,
+      "val_ece": 0.017922561350632537
+    },
+    {
+      "round": 6,
+      "n_labeled": 168,
+      "selection_seconds": 17.5,
+      "round_seconds": 408.0,
+      "teacher_dice_labeled": [
+        0.3679188297764999,
+        0.4335570192451118,
+        0.20895861407370306,
+        0.33643354064960024,
+        0.5531799194234047,
+        0.17692680619249224,
+        0.2365003677000348,
+        0.2917085664928225
+      ],
+      "teacher_dice_pool": [
+        0.394125303581231,
+        0.47494787946458666,
+        0.24170386806015567,
+        0.3806315415340604,
+        0.6459687009021247,
+        0.20181318421776986,
+        0.2769592359329296,
+        0.3104454867226209
+      ],
+      "evidence_fg_pixels": 1683796,
+      "evidence_nonempty_patches": 168,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0016089924285309985,
+        0.034023888217738066,
+        0.02454511057925963,
+        0.00611628664709167,
+        0.9295109032521134,
+        0.0003093465602107464,
+        0.0009196454009503332,
+        0.0029658269141052424
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5524438317045043,
+        "uniform": 0.37409291675858214,
+        "top1": 0.5531799194234047,
+        "top2": 0.5542681111280099,
+        "top3": 0.5543604600633237
+      },
+      "pseudo_student_dice": 0.642503799192999,
+      "pseudo_ensemble_dice": 0.47485483653762367,
+      "pseudo_source": "student",
+      "consistency_scale": 0.475,
+      "val_dice": 0.7556781192564479,
+      "val_vi": 0.5299056030507209,
+      "val_ece": 0.01800360725531013
+    },
+    {
+      "round": 7,
+      "n_labeled": 168,
+      "selection_seconds": 0.0,
+      "round_seconds": 115.3,
+      "final_consolidation": true,
+      "val_dice": 0.7553752308778334,
+      "val_vi": 0.5248065360487096,
+      "val_ece": 0.023144835746207494
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.3088075353000357,
+    "auroc_top20_error": 0.7218027777541671
+  },
+  "al_seconds": 2309.8,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.0016089924285309985,
+    0.034023888217738066,
+    0.02454511057925963,
+    0.00611628664709167,
+    0.9295109032521134,
+    0.0003093465602107464,
+    0.0009196454009503332,
+    0.0029658269141052424
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9295109032521134,
+    "entropy": 0.3416784214280678,
+    "effective_sources": 1.4073076647504048,
+    "min_weight": 0.0003093465602107464
+  },
+  "top_source_final": "microns_basil",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.7490979386466647,
+    "patch_iou": 0.6085798192529274,
+    "patch_precision": 0.7248434080358238,
+    "patch_recall": 0.7785118795775513,
+    "patch_vi": 0.5298728975152815,
+    "n_patches": 735,
+    "cal_ece": 0.020920834522142353,
+    "cal_nll": 0.1941572906083464,
+    "cal_brier": 0.060966210654075155,
+    "cal_mean_confidence": 0.9338938362228226,
+    "cal_accuracy": 0.9129730017006803,
+    "cal_predicted_prevalence": 0.2180125956632653,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.21629047139314697
+  },
+  "test_notta": {
+    "patch_dice": 0.7490979386466647,
+    "patch_iou": 0.6085798192529274,
+    "patch_precision": 0.7248434080358238,
+    "patch_recall": 0.7785118795775513,
+    "patch_vi": 0.5298728975152815,
+    "n_patches": 735,
+    "cal_ece": 0.020920834522142353,
+    "cal_nll": 0.1941572906083464,
+    "cal_brier": 0.060966210654075155,
+    "cal_mean_confidence": 0.9338938362228226,
+    "cal_accuracy": 0.9129730017006803,
+    "cal_predicted_prevalence": 0.2180125956632653,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.21629047139314697
+  },
+  "test": {
+    "patch_dice": 0.7605704968659796,
+    "patch_iou": 0.6231444504129582,
+    "patch_precision": 0.7389570238203815,
+    "patch_recall": 0.7866511074140025,
+    "patch_vi": 0.5100490645122637,
+    "n_patches": 735,
+    "image_dice": 0.8081044889984601,
+    "image_iou": 0.6784548862789703,
+    "image_precision": 0.790270861952727,
+    "image_recall": 0.8272684535945679,
+    "image_vi": 0.516524529058358,
+    "n_images": 15,
+    "cal_ece": 0.010442009413797379,
+    "cal_nll": 0.17936422828495757,
+    "cal_brier": 0.05674318918072776,
+    "cal_mean_confidence": 0.9284094307573348,
+    "cal_accuracy": 0.9179674213435374,
+    "cal_predicted_prevalence": 0.21568877551020407,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.21565880639981094
+  },
+  "eval_seconds": 38.8,
+  "total_seconds": 3067.8,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 1.48,
+        "peak_cpu_rss_mb": 2192.2,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 63.24,
+        "peak_cpu_rss_mb": 4183.8,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 649.47,
+        "peak_cpu_rss_mb": 5161.3,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4638.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2309.76,
+        "peak_cpu_rss_mb": 5161.3,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4646.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 38.77,
+        "peak_cpu_rss_mb": 5161.3,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4646.0,
+        "patches": 735,
+        "images": 15,
+        "patches_per_second": 19.0,
+        "seconds_per_image": 2.585
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5161.3,
+    "total_seconds": 3062.7
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/snemi3d/macs-v11-norestart-acq-margin_b5_s1.pth"
+}

macs_v2_r26/runs/membrane/snemi3d/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
+1,28,100.1,540.4,"[0.35791764681874827, 0.4417052610943664, 0.21306314427094966, 0.3478407216013448, 0.5427570406835327, 0.2427000034404966, 0.2815004375801934, 0.2748319195941523]",,279577,28,True,kept,0.4329726063476451,0.5427570406835327,"[0.0015948525493022832, 0.04043639495708322, 0.0280150797666998, 0.007509823162345696, 0.9174652491168103, 0.0006541777664380059, 0.001579990027762249, 0.002744432653558335]",mixture,"{'mixture': 0.543427983950943, 'uniform': 0.42744369108351205, 'top1': 0.5427570406835327, 'top2': 0.5446170820822631, 'top3': 0.5446748258576916}",0.7295256186222652,0.5884044823108097,0.019701315203793958,,,,,
+2,56,18.1,60.7,"[0.3208247601267692, 0.41290094315652537, 0.2040200567361768, 0.2958021771680537, 0.41646418759010906, 0.1794556776225178, 0.20312701542316977, 0.2733950553226702]",,566444,56,False,,,,"[0.003783436799557813, 0.10421618553102502, 0.0, 0.0, 0.8920003776694172, 0.0, 0.0, 0.0]",top3,"{'mixture': 0.4129965036621742, 'uniform': 0.30746055861849897, 'top1': 0.41646418759010906, 'top2': 0.4222413401292082, 'top3': 0.4224928778934572}",0.7432631786091862,0.5466669819950293,0.01494093341048987,0.6110748083498765,0.23453104616733697,student,0.37,
+3,84,18.0,373.7,"[0.36273121427086064, 0.4372980514434087, 0.21105590198097918, 0.3393942140665256, 0.540192314721962, 0.17578747616669918, 0.22484020938853538, 0.29315816733536]","[0.39034661919349994, 0.4761150428010431, 0.23573876902414095, 0.38585503964914447, 0.659109861832459, 0.18712604425893262, 0.2631012063047301, 0.3108627311160762]",865230,84,False,,,,"[0.0017190749968959006, 0.03974672870367697, 0.02820614601955311, 0.007089510065337752, 0.9185871992481591, 0.0003441657073025338, 0.0009209823989749294, 0.0033861928600996084]",mixture,"{'mixture': 0.5393467185906318, 'uniform': 0.3655772808350024, 'top1': 0.540192314721962, 'top2': 0.541732206975406, 'top3': 0.541822389111206}",0.7527323068124526,0.541985946096611,0.01677360838441982,0.6251532581237091,0.4286494839036211,student,0.417,
+4,112,17.8,384.6,"[0.3749811196798546, 0.44113104576281453, 0.20960975570483995, 0.3387416433056273, 0.5495413048885877, 0.17535012944458866, 0.23520326785288623, 0.2988056113727459]","[0.40815563458449466, 0.4890113030690118, 0.23786569305882985, 0.3857946049443244, 0.6569114664585771, 0.1934436301469982, 0.2718589618428649, 0.3221797503419022]",1141567,112,False,,,,"[0.001779979044505536, 0.037832694697906344, 0.02546728326344037, 0.006452106793159957, 0.9239361080593134, 0.00031389733227876727, 0.0009357868195762976, 0.003282143989819337]",mixture,"{'mixture': 0.5488637066017804, 'uniform': 0.37210821529489313, 'top1': 0.5495413048885877, 'top2': 0.5509644238645611, 'top3': 0.5510551274892082}",0.7509571821136031,0.5315801650431149,0.01703855312358479,0.6628713841310659,0.49864015261866945,student,0.543,
+5,140,17.7,397.3,"[0.3642759867907599, 0.43358117185195766, 0.21068374327687786, 0.33467724413782185, 0.5535326066210269, 0.17798854274162795, 0.24864286910367464, 0.2906701335016776]","[0.38218506328160307, 0.4682752407941175, 0.23816118972250283, 0.3750151946437794, 0.6498135108202829, 0.19764850889095936, 0.2903312054949055, 0.30404861986044174]",1422513,140,False,,,,"[0.001545796995137486, 0.03390845176712642, 0.02488148279719892, 0.005987967802871813, 0.9294056619302986, 0.0003115125004738827, 0.0010346030187799481, 0.002924523188112882]",mixture,"{'mixture': 0.5528844985001343, 'uniform': 0.3808982550980504, 'top1': 0.5535326066210269, 'top2': 0.5547303978453834, 'top3': 0.554809582158575}",0.7545000122039486,0.533773990628012,0.017922561350632537,0.6609389649034607,0.5110580295086186,student,0.536,
+6,168,17.5,408.0,"[0.3679188297764999, 0.4335570192451118, 0.20895861407370306, 0.33643354064960024, 0.5531799194234047, 0.17692680619249224, 0.2365003677000348, 0.2917085664928225]","[0.394125303581231, 0.47494787946458666, 0.24170386806015567, 0.3806315415340604, 0.6459687009021247, 0.20181318421776986, 0.2769592359329296, 0.3104454867226209]",1683796,168,False,,,,"[0.0016089924285309985, 0.034023888217738066, 0.02454511057925963, 0.00611628664709167, 0.9295109032521134, 0.0003093465602107464, 0.0009196454009503332, 0.0029658269141052424]",mixture,"{'mixture': 0.5524438317045043, 'uniform': 0.37409291675858214, 'top1': 0.5531799194234047, 'top2': 0.5542681111280099, 'top3': 0.5543604600633237}",0.7556781192564479,0.5299056030507209,0.01800360725531013,0.642503799192999,0.47485483653762367,student,0.475,
+7,168,0.0,115.3,,,,,,,,,,,,0.7553752308778334,0.5248065360487096,0.023144835746207494,,,,,True

macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/log.txt

@@ -0,0 +1,103 @@
+[01:53:44][      0s][START] target=snemi3d variant=macs-v11-norestart-acq-mcdropout budget=5.0% seed=1 | {"task": "membrane", "target": "snemi3d", "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": 8, "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}
+[01:53:44][      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.
+[01:53:44][      1s][DATA] snemi3d: train=3430 val=735 test=735 patches
+[01:53:50][      6s][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=4.07
+[01:53:51][      7s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 1s
+[01:53:57][     13s][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
+[01:53:57][     13s][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
+[01:54:51][     67s][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
+[01:54:51][     67s][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
+[01:54:51][     67s][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
+[01:54:51][     67s][INIT] student initialised from teacher microns (weight 0.477)
+[01:54:56][     73s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[01:54:56][     73s][KD] distilling the mixture ensemble (prob space) into the student
+[01:57:16][    213s][KD] soft labels for 3430 patches from 8 teachers in 140s
+[01:58:59][    315s][KD] epoch 1/5 | patches=3430 | loss=0.4461 | 102s
+[02:00:41][    417s][KD] epoch 2/5 | patches=3430 | loss=0.3728 | 102s
+[02:02:23][    519s][KD] epoch 3/5 | patches=3430 | loss=0.3563 | 102s
+[02:04:05][    621s][KD] epoch 4/5 | patches=3430 | loss=0.3563 | 102s
+[02:05:46][    723s][KD] epoch 5/5 | patches=3430 | loss=0.3474 | 102s
+[02:05:51][    727s][ROUND0] before any target label: val_dice=0.3756 val_vi=0.6391
+[02:05:51][    727s][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
+[02:06:11][    748s][AL] round-1 uncertainty quality on 3430 patches: {'spearman_uncertainty_error': 0.4788310711035088, 'auroc_top20_error': 0.8198074888014347}
+[02:07:33][    830s][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
+[02:07:38][    835s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.543, uniform=0.427, top1=0.543, top2=0.545, top3=0.545 -> using mixture
+[02:07:39][    835s][WEIGHTS] round 1: student kept (labeled Dice 0.418 vs best teacher microns 0.543); distilling the corrected ensemble into it
+[02:07:39][    835s][WEIGHTS] round 1: teacher Dice on 28 labeled patches + 0 blank (279577 fg px): c-elegans=0.358, cerebellum=0.442, flywire=0.213, berghia=0.348, microns=0.543, oct-glia=0.243, oct-sfl=0.282, mouse-brain=0.275 | weights: c-elegans=0.002, cerebellum=0.040, flywire=0.028, berghia=0.008, microns=0.917, oct-glia=0.001, oct-sfl=0.002, mouse-brain=0.003
+[02:09:58][    974s][KD] soft labels for 3402 patches from 8 teachers in 139s
+[02:11:39][   1076s][KD] epoch 1/2 | patches=3402 | loss=0.2532 | 101s
+[02:13:21][   1177s][KD] epoch 2/2 | patches=3402 | loss=0.2504 | 101s
+[02:14:43][   1259s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 82s
+[02:14:45][   1261s][FT] round 1/6 epoch 1/5 | patches=28 | loss=1.0414
+[02:14:48][   1264s][FT] round 1/6 epoch 2/5 | patches=28 | loss=0.9800
+[02:14:50][   1266s][FT] round 1/6 epoch 3/5 | patches=28 | loss=0.9003
+[02:14:52][   1269s][FT] round 1/6 epoch 4/5 | patches=28 | loss=0.8646
+[02:14:55][   1271s][FT] round 1/6 epoch 5/5 | patches=28 | loss=0.8398
+[02:14:59][   1275s][AL] round 1/6 | labeled=28 | selection_seconds=102.6000 round_seconds=544.0000 restart_student_dice=0.4183 restart_teacher_dice=0.5428 val_dice=0.7278 val_vi=0.5919 val_ece=0.0210
+[02:15:29][   1305s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.422, uniform=0.320, top1=0.424, top2=0.430, top3=0.430 -> using top3
+[02:15:29][   1305s][WEIGHTS] round 2: teacher Dice on 56 labeled patches + 0 blank (542568 fg px): c-elegans=0.312, cerebellum=0.412, flywire=0.206, berghia=0.270, microns=0.424, oct-glia=0.186, oct-sfl=0.216, mouse-brain=0.263 | weights: c-elegans=0.003, cerebellum=0.097, flywire=0.000, berghia=0.000, microns=0.900, oct-glia=0.000, oct-sfl=0.000, mouse-brain=0.000
+[02:15:42][   1318s][CONS] round 2: on the 28 new patches student Dice=0.595 vs ensemble 0.236 -> pseudo-labels from the student | consistency x0.32
+[02:15:46][   1322s][FT] round 2/6 epoch 1/5 | patches=56 | loss=0.8159
+[02:15:50][   1326s][FT] round 2/6 epoch 2/5 | patches=56 | loss=0.7590
+[02:15:54][   1330s][FT] round 2/6 epoch 3/5 | patches=56 | loss=0.7268
+[02:15:58][   1334s][FT] round 2/6 epoch 4/5 | patches=56 | loss=0.7150
+[02:16:01][   1338s][FT] round 2/6 epoch 5/5 | patches=56 | loss=0.7119
+[02:16:06][   1342s][AL] round 2/6 | labeled=56 | selection_seconds=20.5000 round_seconds=62.5000 pseudo_student_dice=0.5949 pseudo_ensemble_dice=0.2362 consistency_scale=0.3160 val_dice=0.7417 val_vi=0.5611 val_ece=0.0185
+[02:19:08][   1524s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.412, cerebellum=0.490, flywire=0.240, berghia=0.396, microns=0.664, oct-glia=0.193, oct-sfl=0.265, mouse-brain=0.328 | blended with labeled Dice at 0.50
+[02:21:40][   1676s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.534, uniform=0.357, top1=0.536, top2=0.537, top3=0.537 -> using mixture
+[02:21:40][   1676s][WEIGHTS] round 3: teacher Dice on 84 labeled patches + 0 blank (821274 fg px): c-elegans=0.361, cerebellum=0.441, flywire=0.214, berghia=0.334, microns=0.536, oct-glia=0.174, oct-sfl=0.224, mouse-brain=0.294 | weights: c-elegans=0.002, cerebellum=0.043, flywire=0.030, berghia=0.007, microns=0.913, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.004
+[02:21:53][   1689s][CONS] round 3: on the 28 new patches student Dice=0.647 vs ensemble 0.369 -> pseudo-labels from the student | consistency x0.49
+[02:21:59][   1695s][FT] round 3/6 epoch 1/5 | patches=84 | loss=0.7596
+[02:22:05][   1701s][FT] round 3/6 epoch 2/5 | patches=84 | loss=0.7431
+[02:22:10][   1706s][FT] round 3/6 epoch 3/5 | patches=84 | loss=0.7259
+[02:22:16][   1712s][FT] round 3/6 epoch 4/5 | patches=84 | loss=0.6955
+[02:22:22][   1718s][FT] round 3/6 epoch 5/5 | patches=84 | loss=0.6926
+[02:22:26][   1722s][AL] round 3/6 | labeled=84 | selection_seconds=20.2000 round_seconds=375.8000 pseudo_student_dice=0.6475 pseudo_ensemble_dice=0.3694 consistency_scale=0.4920 val_dice=0.7496 val_vi=0.5421 val_ece=0.0175
+[02:25:30][   1906s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.391, cerebellum=0.475, flywire=0.235, berghia=0.386, microns=0.652, oct-glia=0.185, oct-sfl=0.267, mouse-brain=0.310 | blended with labeled Dice at 0.50
+[02:28:03][   2060s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.543, uniform=0.359, top1=0.544, top2=0.545, top3=0.545 -> using mixture
+[02:28:03][   2060s][WEIGHTS] round 4: teacher Dice on 112 labeled patches + 0 blank (1131999 fg px): c-elegans=0.358, cerebellum=0.430, flywire=0.206, berghia=0.329, microns=0.544, oct-glia=0.165, oct-sfl=0.231, mouse-brain=0.290 | weights: c-elegans=0.002, cerebellum=0.036, flywire=0.026, berghia=0.006, microns=0.926, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[02:28:17][   2073s][CONS] round 4: on the 28 new patches student Dice=0.661 vs ensemble 0.499 -> pseudo-labels from the student | consistency x0.54
+[02:28:24][   2080s][FT] round 4/6 epoch 1/5 | patches=112 | loss=0.7447
+[02:28:31][   2088s][FT] round 4/6 epoch 2/5 | patches=112 | loss=0.7579
+[02:28:39][   2095s][FT] round 4/6 epoch 3/5 | patches=112 | loss=0.7504
+[02:28:46][   2102s][FT] round 4/6 epoch 4/5 | patches=112 | loss=0.7336
+[02:28:53][   2109s][FT] round 4/6 epoch 5/5 | patches=112 | loss=0.7056
+[02:28:57][   2113s][AL] round 4/6 | labeled=112 | selection_seconds=20.2000 round_seconds=387.2000 pseudo_student_dice=0.6605 pseudo_ensemble_dice=0.4986 consistency_scale=0.5350 val_dice=0.7432 val_vi=0.5390 val_ece=0.0128
+[02:32:03][   2299s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.367, cerebellum=0.452, flywire=0.235, berghia=0.373, microns=0.640, oct-glia=0.192, oct-sfl=0.282, mouse-brain=0.291 | blended with labeled Dice at 0.50
+[02:34:39][   2455s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.549, uniform=0.365, top1=0.550, top2=0.551, top3=0.551 -> using mixture
+[02:34:39][   2455s][WEIGHTS] round 5: teacher Dice on 140 labeled patches + 0 blank (1433291 fg px): c-elegans=0.352, cerebellum=0.422, flywire=0.208, berghia=0.323, microns=0.550, oct-glia=0.166, oct-sfl=0.241, mouse-brain=0.283 | weights: c-elegans=0.001, cerebellum=0.031, flywire=0.025, berghia=0.006, microns=0.933, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[02:34:52][   2468s][CONS] round 5: on the 28 new patches student Dice=0.692 vs ensemble 0.540 -> pseudo-labels from the student | consistency x0.64
+[02:35:01][   2477s][FT] round 5/6 epoch 1/5 | patches=140 | loss=0.7161
+[02:35:10][   2486s][FT] round 5/6 epoch 2/5 | patches=140 | loss=0.7050
+[02:35:19][   2495s][FT] round 5/6 epoch 3/5 | patches=140 | loss=0.6953
+[02:35:28][   2504s][FT] round 5/6 epoch 4/5 | patches=140 | loss=0.6930
+[02:35:37][   2513s][FT] round 5/6 epoch 5/5 | patches=140 | loss=0.6856
+[02:35:41][   2517s][AL] round 5/6 | labeled=140 | selection_seconds=19.9000 round_seconds=399.7000 pseudo_student_dice=0.6920 pseudo_ensemble_dice=0.5397 consistency_scale=0.6400 val_dice=0.7580 val_vi=0.5310 val_ece=0.0156
+[02:38:49][   2705s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.395, cerebellum=0.477, flywire=0.236, berghia=0.376, microns=0.639, oct-glia=0.197, oct-sfl=0.282, mouse-brain=0.316 | blended with labeled Dice at 0.50
+[02:41:27][   2863s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.558, uniform=0.364, top1=0.559, top2=0.560, top3=0.560 -> using mixture
+[02:41:27][   2863s][WEIGHTS] round 6: teacher Dice on 168 labeled patches + 0 blank (1730353 fg px): c-elegans=0.369, cerebellum=0.435, flywire=0.207, berghia=0.325, microns=0.559, oct-glia=0.162, oct-sfl=0.245, mouse-brain=0.297 | weights: c-elegans=0.002, cerebellum=0.033, flywire=0.023, berghia=0.005, microns=0.933, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[02:41:40][   2876s][CONS] round 6: on the 28 new patches student Dice=0.684 vs ensemble 0.552 -> pseudo-labels from the student | consistency x0.61
+[02:41:51][   2887s][FT] round 6/6 epoch 1/5 | patches=168 | loss=0.6902
+[02:42:01][   2897s][FT] round 6/6 epoch 2/5 | patches=168 | loss=0.6889
+[02:42:12][   2908s][FT] round 6/6 epoch 3/5 | patches=168 | loss=0.6774
+[02:42:22][   2918s][FT] round 6/6 epoch 4/5 | patches=168 | loss=0.6739
+[02:42:32][   2929s][FT] round 6/6 epoch 5/5 | patches=168 | loss=0.6641
+[02:42:37][   2933s][AL] round 6/6 | labeled=168 | selection_seconds=19.9000 round_seconds=411.2000 pseudo_student_dice=0.6845 pseudo_ensemble_dice=0.5515 consistency_scale=0.6150 val_dice=0.7619 val_vi=0.5293 val_ece=0.0177
+[02:42:47][   2943s][CONS] final consolidation: pseudo-labels refreshed from the student
+[02:42:57][   2953s][FT] final 1/10 epoch 1/1 | patches=168 | loss=0.6622
+[02:43:08][   2964s][FT] final 2/10 epoch 1/1 | patches=168 | loss=0.6623
+[02:43:18][   2974s][FT] final 3/10 epoch 1/1 | patches=168 | loss=0.6467
+[02:43:29][   2985s][FT] final 4/10 epoch 1/1 | patches=168 | loss=0.6459
+[02:43:39][   2995s][FT] final 5/10 epoch 1/1 | patches=168 | loss=0.6379
+[02:43:50][   3006s][FT] final 6/10 epoch 1/1 | patches=168 | loss=0.6291
+[02:44:00][   3016s][FT] final 7/10 epoch 1/1 | patches=168 | loss=0.6209
+[02:44:11][   3027s][FT] final 8/10 epoch 1/1 | patches=168 | loss=0.6170
+[02:44:21][   3037s][FT] final 9/10 epoch 1/1 | patches=168 | loss=0.6202
+[02:44:32][   3048s][FT] final 10/10 epoch 1/1 | patches=168 | loss=0.6203
+[02:44:36][   3052s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=114.9000 val_dice=0.7586 val_vi=0.5257 val_ece=0.0234
+[02:44:36][   3052s][WEIGHTS] final: c-elegans=0.002, cerebellum=0.033, flywire=0.023, berghia=0.005, microns=0.933, oct-glia=0.000, oct-sfl=0.001, mouse-brain=0.003
+[02:45:11][   3087s][TEST] raw student (no EMA, no TTA): patch_dice=0.7514 patch_vi=0.5301
+[02:45:15][   3092s][TEST] final model without TTA: patch_dice=0.7514 patch_vi=0.5301
+[02:45:15][   3092s][RESOURCES] {"phases": {"setup": {"seconds": 4.3, "peak_cpu_rss_mb": 2193.5, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 66.88, "peak_cpu_rss_mb": 4186.1, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 650.11, "peak_cpu_rss_mb": 5163.3, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4638.0, "kd_epochs": 5}, "active_learning": {"seconds": 2325.13, "peak_cpu_rss_mb": 5163.3, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4646.0, "rounds": 6}, "inference": {"seconds": 39.44, "peak_cpu_rss_mb": 5163.3, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4646.0, "patches": 735, "images": 15, "patches_per_second": 18.6, "seconds_per_image": 2.629}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5163.3, "total_seconds": 3085.9}
+[02:45:15][   3092s][TEST] patch_dice=0.7617 patch_iou=0.6247 patch_precision=0.7320 patch_recall=0.7970 patch_vi=0.5112 image_dice=0.8086 image_iou=0.6791 image_precision=0.7826 image_recall=0.8367 image_vi=0.5191 cal_ece=0.0115 cal_nll=0.1807 cal_brier=0.0572 cal_mean_confidence=0.9288 cal_accuracy=0.9173 cal_predicted_prevalence=0.2214 cal_true_prevalence=0.2059 cal_mean_probability=0.2207
+[02:45:16][   3092s][DONE] snemi3d macs-v11-norestart-acq-mcdropout b5 s1 | test patch Dice=0.7617 VI=0.5112 | image Dice=0.8086 VI=0.5191 | 3092s

macs_v2_r26/runs/membrane/snemi3d/macs-v11-norestart-acq-mcdropout/budget-5/seed-1/result.json

@@ -0,0 +1,785 @@
+{
+  "target": "snemi3d",
+  "variant": "macs-v11-norestart-acq-mcdropout",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "membrane",
+    "target": "snemi3d",
+    "variant": "macs-v11-norestart-acq-mcdropout",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "snemi3d",
+      "--variants",
+      "macs-v11-norestart-acq-mcdropout",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.005265163985947525,
+    0.05775356661687405,
+    0.3937117170688725,
+    0.02742108315870487,
+    0.4770168284742763,
+    0.006835511403011292,
+    0.011200087973044965,
+    0.020796041319268533
+  ],
+  "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": [
+      0.005265163985947525,
+      0.05775356661687405,
+      0.3937117170688725,
+      0.02742108315870487,
+      0.4770168284742763,
+      0.006835511403011292,
+      0.011200087973044965,
+      0.020796041319268533
+    ],
+    "per_source": [
+      0.0030099407416697968,
+      0.30826237672231105,
+      0.030874093815323564,
+      0.005310674836551046,
+      0.6243418715179362,
+      0.02131301441017302,
+      0.0048036204646279375,
+      0.002084407491407262
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.048363233996627016,
+      "mmd2_mixture": 0.060122682064390254,
+      "mmd2_uniform": 0.15285688303099165,
+      "mmd2_single_source": [
+        0.4522663898122707,
+        0.14701298584193184,
+        0.12241687119492284,
+        0.2726521753211397,
+        0.08995170742180114,
+        0.3744254300178685,
+        0.33957846092746485,
+        0.23688038103280662
+      ],
+      "median_squared_distance": 0.18662990120119327,
+      "gammas": [
+        1.3395495490858758,
+        2.6790990981717515,
+        5.358198196343503,
+        10.716396392687006,
+        21.432792785374012
+      ],
+      "max_weight": 0.4770168284742763,
+      "entropy": 1.1759448067763238,
+      "effective_sources": 3.24120380896937,
+      "min_weight": 0.005265163985947525
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.4624401356050084,
+        0.12211205990556462,
+        0.29128492533416295,
+        0.4206953767024899,
+        0.07022529260621624,
+        0.3185315424880186,
+        0.42807307387367177,
+        0.489454149021297
+      ],
+      "standardized": [
+        0.9323134642607491,
+        -1.3822019864377946,
+        -0.23168506223542185,
+        0.6484142024470035,
+        -1.735075384714815,
+        -0.04638528568228213,
+        0.6985887114356643,
+        1.1160313409268967
+      ],
+      "tau": 0.5,
+      "max_weight": 0.6243418715179362,
+      "entropy": 0.9300659298676801,
+      "effective_sources": 2.534676282981126,
+      "min_weight": 0.002084407491407262
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "microns_basil",
+    "seconds": 7.4
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.41247125221353,
+      0.47321264983034456,
+      0.19639403256049165,
+      0.31013587127331976,
+      0.5894646206275136,
+      0.13744302537529526,
+      0.3054434345563225,
+      0.35012697230796425
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.4205061683646676,
+      "mixture@raw": 0.462918877571213,
+      "mixture@logit": 0.3665782427199844,
+      "per_source": 0.5892228765147073,
+      "per_source@raw": 0.5905348345941427,
+      "per_source@logit": 0.5781053080039771,
+      "uniform": 0.41124230184319877,
+      "uniform@raw": 0.4669572710990906,
+      "uniform@logit": 0.22782834015245554
+    },
+    "val_patches": 735
+  },
+  "temperatures": [
+    2.575598955154419,
+    4.557764530181885,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    2.642308473587036,
+    4.074584007263184
+  ],
+  "top_source": "microns_basil",
+  "weight_stats": {
+    "max_weight": 0.4770168284742763,
+    "entropy": 1.1759448067763238,
+    "effective_sources": 3.24120380896937,
+    "min_weight": 0.005265163985947525
+  },
+  "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": 650.1,
+  "round0_val": {
+    "val_dice": 0.3755735946634418,
+    "val_vi": 0.6390912588283757,
+    "val_ece": 0.07344130447917224
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 28,
+      "selection_seconds": 102.6,
+      "round_seconds": 544.0,
+      "teacher_dice_labeled": [
+        0.35791764681874827,
+        0.4417052610943664,
+        0.21306314427094966,
+        0.3478407216013448,
+        0.5427570406835327,
+        0.2427000034404966,
+        0.2815004375801934,
+        0.2748319195941523
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 279577,
+      "evidence_nonempty_patches": 28,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.41833134967864793,
+      "restart_teacher_dice": 0.5427570406835327,
+      "weights": [
+        0.0015948525493022832,
+        0.04043639495708322,
+        0.0280150797666998,
+        0.007509823162345696,
+        0.9174652491168103,
+        0.0006541777664380059,
+        0.001579990027762249,
+        0.002744432653558335
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.543427983950943,
+        "uniform": 0.42744369108351205,
+        "top1": 0.5427570406835327,
+        "top2": 0.5446170820822631,
+        "top3": 0.5446748258576916
+      },
+      "val_dice": 0.7278339915577575,
+      "val_vi": 0.5919187624963894,
+      "val_ece": 0.020966535856338883
+    },
+    {
+      "round": 2,
+      "n_labeled": 56,
+      "selection_seconds": 20.5,
+      "round_seconds": 62.5,
+      "teacher_dice_labeled": [
+        0.31181172477305286,
+        0.4123490463356484,
+        0.20608301371020069,
+        0.27008440705749637,
+        0.4244959608987849,
+        0.18641832734596342,
+        0.21613912931956966,
+        0.26276946418977837
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 542568,
+      "evidence_nonempty_patches": 56,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.003220035189592252,
+        0.09652850169918203,
+        0.0,
+        0.0,
+        0.9002514631112257,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top3",
+      "candidate_dice": {
+        "mixture": 0.4215898139097271,
+        "uniform": 0.3200273826153171,
+        "top1": 0.4244959608987849,
+        "top2": 0.43002979418608317,
+        "top3": 0.4302272878082905
+      },
+      "pseudo_student_dice": 0.5949284088746447,
+      "pseudo_ensemble_dice": 0.23617475681690378,
+      "pseudo_source": "student",
+      "consistency_scale": 0.316,
+      "val_dice": 0.7416816901534429,
+      "val_vi": 0.5611395139148813,
+      "val_ece": 0.018502803413247136
+    },
+    {
+      "round": 3,
+      "n_labeled": 84,
+      "selection_seconds": 20.2,
+      "round_seconds": 375.8,
+      "teacher_dice_labeled": [
+        0.36133371836338635,
+        0.44085686194906737,
+        0.2136680721683852,
+        0.3340398695594378,
+        0.535725225724645,
+        0.17392592764506856,
+        0.2235620921091399,
+        0.2938632487285832
+      ],
+      "teacher_dice_pool": [
+        0.41194979078777727,
+        0.49042473385394364,
+        0.240436043260484,
+        0.3960085279038956,
+        0.6642957966312557,
+        0.19332109643781695,
+        0.2651361815473013,
+        0.3280756406570733
+      ],
+      "evidence_fg_pixels": 821274,
+      "evidence_nonempty_patches": 84,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0017627896179064318,
+        0.04282841635643022,
+        0.030106700948834032,
+        0.0069877524086444905,
+        0.913471445141928,
+        0.0003512836206721712,
+        0.0009455302888582306,
+        0.003546081616726441
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5344386028800578,
+        "uniform": 0.35706953542833364,
+        "top1": 0.535725225724645,
+        "top2": 0.5373175326220334,
+        "top3": 0.5374325155195703
+      },
+      "pseudo_student_dice": 0.647461004157173,
+      "pseudo_ensemble_dice": 0.3694070799234942,
+      "pseudo_source": "student",
+      "consistency_scale": 0.492,
+      "val_dice": 0.749626075653042,
+      "val_vi": 0.5421390132818664,
+      "val_ece": 0.017509178777303697
+    },
+    {
+      "round": 4,
+      "n_labeled": 112,
+      "selection_seconds": 20.2,
+      "round_seconds": 387.2,
+      "teacher_dice_labeled": [
+        0.3576869817133024,
+        0.4298228509106173,
+        0.20622688258575914,
+        0.3286969851255738,
+        0.5436391225695805,
+        0.16475371617754514,
+        0.23146357713103805,
+        0.2901076096672102
+      ],
+      "teacher_dice_pool": [
+        0.39138772527845306,
+        0.475016594588615,
+        0.23501316215553472,
+        0.38595151149100176,
+        0.6518267028414817,
+        0.1851465791692955,
+        0.266669253417049,
+        0.31029983774319086
+      ],
+      "evidence_fg_pixels": 1131999,
+      "evidence_nonempty_patches": 112,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0015913184560729483,
+        0.03590923115993884,
+        0.02616620169627435,
+        0.006201944108146959,
+        0.9256755334459202,
+        0.00030006661828817027,
+        0.0009580437690906246,
+        0.0031976607462679196
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.5425422974902394,
+        "uniform": 0.35911726422428564,
+        "top1": 0.5436391225695805,
+        "top2": 0.5449977421327727,
+        "top3": 0.5450780283194936
+      },
+      "pseudo_student_dice": 0.6605078248292389,
+      "pseudo_ensemble_dice": 0.4986002301881694,
+      "pseudo_source": "student",
+      "consistency_scale": 0.535,
+      "val_dice": 0.743164012252046,
+      "val_vi": 0.5390013919900811,
+      "val_ece": 0.012813062188756805
+    },
+    {
+      "round": 5,
+      "n_labeled": 140,
+      "selection_seconds": 19.9,
+      "round_seconds": 399.7,
+      "teacher_dice_labeled": [
+        0.3518329705798253,
+        0.4218277901383832,
+        0.2075243638051535,
+        0.3233491742376384,
+        0.5503626085755353,
+        0.1655991382860516,
+        0.2414640614809438,
+        0.2833941818452663
+      ],
+      "teacher_dice_pool": [
+        0.36696379474936547,
+        0.45218328641072336,
+        0.23535295439566672,
+        0.3729089636121604,
+        0.6402919292608433,
+        0.1915812146114833,
+        0.28248284990582206,
+        0.29116450464463495
+      ],
+      "evidence_fg_pixels": 1433291,
+      "evidence_nonempty_patches": 140,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0014139596753337835,
+        0.03123112082644807,
+        0.024973484663534776,
+        0.005538678696212871,
+        0.9327431865927268,
+        0.0002850971507548484,
+        0.0009975173861996782,
+        0.0028169550087891832
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.549249434136135,
+        "uniform": 0.3652362373751127,
+        "top1": 0.5503626085755353,
+        "top2": 0.5514334084307191,
+        "top3": 0.5514877157094626
+      },
+      "pseudo_student_dice": 0.6920095778498158,
+      "pseudo_ensemble_dice": 0.5397081235622629,
+      "pseudo_source": "student",
+      "consistency_scale": 0.64,
+      "val_dice": 0.7580254046092908,
+      "val_vi": 0.5310221427433142,
+      "val_ece": 0.015614619822543528
+    },
+    {
+      "round": 6,
+      "n_labeled": 168,
+      "selection_seconds": 19.9,
+      "round_seconds": 411.2,
+      "teacher_dice_labeled": [
+        0.368547271203964,
+        0.43520792181378254,
+        0.20708955846823568,
+        0.3250520172516318,
+        0.5587925243587022,
+        0.1624849748096609,
+        0.24510524644786275,
+        0.2974032896168033
+      ],
+      "teacher_dice_pool": [
+        0.39540179798971437,
+        0.476869730677141,
+        0.2360398325311521,
+        0.3762109930294652,
+        0.6388893327362402,
+        0.19723726041841333,
+        0.28174786281265013,
+        0.31583038741705616
+      ],
+      "evidence_fg_pixels": 1730353,
+      "evidence_nonempty_patches": 168,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0015371438732314031,
+        0.032838624156987614,
+        0.02287067693233035,
+        0.005181908431598925,
+        0.9333852928032684,
+        0.0002541888205423604,
+        0.0009515292617998961,
+        0.0029806357202409647
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.557801610985773,
+        "uniform": 0.3639750335187959,
+        "top1": 0.5587925243587022,
+        "top2": 0.5598847509339682,
+        "top3": 0.559953710155138
+      },
+      "pseudo_student_dice": 0.6844702929054793,
+      "pseudo_ensemble_dice": 0.5515304120552174,
+      "pseudo_source": "student",
+      "consistency_scale": 0.615,
+      "val_dice": 0.7619157981028477,
+      "val_vi": 0.5293165228256188,
+      "val_ece": 0.0177300918492753
+    },
+    {
+      "round": 7,
+      "n_labeled": 168,
+      "selection_seconds": 0.0,
+      "round_seconds": 114.9,
+      "final_consolidation": true,
+      "val_dice": 0.7586223312419698,
+      "val_vi": 0.5257261950571073,
+      "val_ece": 0.02341865030215247
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.4788310711035088,
+    "auroc_top20_error": 0.8198074888014347
+  },
+  "al_seconds": 2325.1,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.0015371438732314031,
+    0.032838624156987614,
+    0.02287067693233035,
+    0.005181908431598925,
+    0.9333852928032684,
+    0.0002541888205423604,
+    0.0009515292617998961,
+    0.0029806357202409647
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9333852928032684,
+    "entropy": 0.3262158082697485,
+    "effective_sources": 1.3857143852303648,
+    "min_weight": 0.0002541888205423604
+  },
+  "top_source_final": "microns_basil",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.7513668328802934,
+    "patch_iou": 0.611391144333655,
+    "patch_precision": 0.7196353193099454,
+    "patch_recall": 0.7894107777903655,
+    "patch_vi": 0.5301334119821381,
+    "n_patches": 735,
+    "cal_ece": 0.02146211620065379,
+    "cal_nll": 0.1954418500392805,
+    "cal_brier": 0.06123293992316221,
+    "cal_mean_confidence": 0.9340989678162999,
+    "cal_accuracy": 0.9126368516156462,
+    "cal_predicted_prevalence": 0.22335246598639455,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.22105475044292994
+  },
+  "test_notta": {
+    "patch_dice": 0.7513668328802934,
+    "patch_iou": 0.611391144333655,
+    "patch_precision": 0.7196353193099454,
+    "patch_recall": 0.7894107777903655,
+    "patch_vi": 0.5301334119821381,
+    "n_patches": 735,
+    "cal_ece": 0.02146211620065379,
+    "cal_nll": 0.1954418500392805,
+    "cal_brier": 0.06123293992316221,
+    "cal_mean_confidence": 0.9340989678162999,
+    "cal_accuracy": 0.9126368516156462,
+    "cal_predicted_prevalence": 0.22335246598639455,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.22105475044292994
+  },
+  "test": {
+    "patch_dice": 0.7617087743524753,
+    "patch_iou": 0.6246963651828682,
+    "patch_precision": 0.7320438250707431,
+    "patch_recall": 0.7970178665759515,
+    "patch_vi": 0.5112018400107708,
+    "n_patches": 735,
+    "image_dice": 0.8085869204312826,
+    "image_iou": 0.67914010219418,
+    "image_precision": 0.7826366296591518,
+    "image_recall": 0.8366991090311863,
+    "image_vi": 0.5190660044759824,
+    "n_images": 15,
+    "cal_ece": 0.011528958416880475,
+    "cal_nll": 0.18065925569353553,
+    "cal_brier": 0.057158908950022694,
+    "cal_mean_confidence": 0.9287868772093976,
+    "cal_accuracy": 0.9172579187925171,
+    "cal_predicted_prevalence": 0.22142591411564627,
+    "cal_true_prevalence": 0.20588727678571428,
+    "cal_mean_probability": 0.22065802468645127
+  },
+  "eval_seconds": 39.4,
+  "total_seconds": 3091.5,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 4.3,
+        "peak_cpu_rss_mb": 2193.5,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 66.88,
+        "peak_cpu_rss_mb": 4186.1,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 650.11,
+        "peak_cpu_rss_mb": 5163.3,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4638.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2325.13,
+        "peak_cpu_rss_mb": 5163.3,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4646.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 39.44,
+        "peak_cpu_rss_mb": 5163.3,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4646.0,
+        "patches": 735,
+        "images": 15,
+        "patches_per_second": 18.6,
+        "seconds_per_image": 2.629
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5163.3,
+    "total_seconds": 3085.9
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/snemi3d/macs-v11-norestart-acq-mcdropout_b5_s1.pth"
+}

macs_v2_r26/runs/membrane/snemi3d/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
+1,28,102.6,544.0,"[0.35791764681874827, 0.4417052610943664, 0.21306314427094966, 0.3478407216013448, 0.5427570406835327, 0.2427000034404966, 0.2815004375801934, 0.2748319195941523]",,279577,28,True,kept,0.41833134967864793,0.5427570406835327,"[0.0015948525493022832, 0.04043639495708322, 0.0280150797666998, 0.007509823162345696, 0.9174652491168103, 0.0006541777664380059, 0.001579990027762249, 0.002744432653558335]",mixture,"{'mixture': 0.543427983950943, 'uniform': 0.42744369108351205, 'top1': 0.5427570406835327, 'top2': 0.5446170820822631, 'top3': 0.5446748258576916}",0.7278339915577575,0.5919187624963894,0.020966535856338883,,,,,
+2,56,20.5,62.5,"[0.31181172477305286, 0.4123490463356484, 0.20608301371020069, 0.27008440705749637, 0.4244959608987849, 0.18641832734596342, 0.21613912931956966, 0.26276946418977837]",,542568,56,False,,,,"[0.003220035189592252, 0.09652850169918203, 0.0, 0.0, 0.9002514631112257, 0.0, 0.0, 0.0]",top3,"{'mixture': 0.4215898139097271, 'uniform': 0.3200273826153171, 'top1': 0.4244959608987849, 'top2': 0.43002979418608317, 'top3': 0.4302272878082905}",0.7416816901534429,0.5611395139148813,0.018502803413247136,0.5949284088746447,0.23617475681690378,student,0.316,
+3,84,20.2,375.8,"[0.36133371836338635, 0.44085686194906737, 0.2136680721683852, 0.3340398695594378, 0.535725225724645, 0.17392592764506856, 0.2235620921091399, 0.2938632487285832]","[0.41194979078777727, 0.49042473385394364, 0.240436043260484, 0.3960085279038956, 0.6642957966312557, 0.19332109643781695, 0.2651361815473013, 0.3280756406570733]",821274,84,False,,,,"[0.0017627896179064318, 0.04282841635643022, 0.030106700948834032, 0.0069877524086444905, 0.913471445141928, 0.0003512836206721712, 0.0009455302888582306, 0.003546081616726441]",mixture,"{'mixture': 0.5344386028800578, 'uniform': 0.35706953542833364, 'top1': 0.535725225724645, 'top2': 0.5373175326220334, 'top3': 0.5374325155195703}",0.749626075653042,0.5421390132818664,0.017509178777303697,0.647461004157173,0.3694070799234942,student,0.492,
+4,112,20.2,387.2,"[0.3576869817133024, 0.4298228509106173, 0.20622688258575914, 0.3286969851255738, 0.5436391225695805, 0.16475371617754514, 0.23146357713103805, 0.2901076096672102]","[0.39138772527845306, 0.475016594588615, 0.23501316215553472, 0.38595151149100176, 0.6518267028414817, 0.1851465791692955, 0.266669253417049, 0.31029983774319086]",1131999,112,False,,,,"[0.0015913184560729483, 0.03590923115993884, 0.02616620169627435, 0.006201944108146959, 0.9256755334459202, 0.00030006661828817027, 0.0009580437690906246, 0.0031976607462679196]",mixture,"{'mixture': 0.5425422974902394, 'uniform': 0.35911726422428564, 'top1': 0.5436391225695805, 'top2': 0.5449977421327727, 'top3': 0.5450780283194936}",0.743164012252046,0.5390013919900811,0.012813062188756805,0.6605078248292389,0.4986002301881694,student,0.535,
+5,140,19.9,399.7,"[0.3518329705798253, 0.4218277901383832, 0.2075243638051535, 0.3233491742376384, 0.5503626085755353, 0.1655991382860516, 0.2414640614809438, 0.2833941818452663]","[0.36696379474936547, 0.45218328641072336, 0.23535295439566672, 0.3729089636121604, 0.6402919292608433, 0.1915812146114833, 0.28248284990582206, 0.29116450464463495]",1433291,140,False,,,,"[0.0014139596753337835, 0.03123112082644807, 0.024973484663534776, 0.005538678696212871, 0.9327431865927268, 0.0002850971507548484, 0.0009975173861996782, 0.0028169550087891832]",mixture,"{'mixture': 0.549249434136135, 'uniform': 0.3652362373751127, 'top1': 0.5503626085755353, 'top2': 0.5514334084307191, 'top3': 0.5514877157094626}",0.7580254046092908,0.5310221427433142,0.015614619822543528,0.6920095778498158,0.5397081235622629,student,0.64,
+6,168,19.9,411.2,"[0.368547271203964, 0.43520792181378254, 0.20708955846823568, 0.3250520172516318, 0.5587925243587022, 0.1624849748096609, 0.24510524644786275, 0.2974032896168033]","[0.39540179798971437, 0.476869730677141, 0.2360398325311521, 0.3762109930294652, 0.6388893327362402, 0.19723726041841333, 0.28174786281265013, 0.31583038741705616]",1730353,168,False,,,,"[0.0015371438732314031, 0.032838624156987614, 0.02287067693233035, 0.005181908431598925, 0.9333852928032684, 0.0002541888205423604, 0.0009515292617998961, 0.0029806357202409647]",mixture,"{'mixture': 0.557801610985773, 'uniform': 0.3639750335187959, 'top1': 0.5587925243587022, 'top2': 0.5598847509339682, 'top3': 0.559953710155138}",0.7619157981028477,0.5293165228256188,0.0177300918492753,0.6844702929054793,0.5515304120552174,student,0.615,
+7,168,0.0,114.9,,,,,,,,,,,,0.7586223312419698,0.5257261950571073,0.02341865030215247,,,,,True

macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-2/log.txt

@@ -0,0 +1,108 @@
+[11:03:34][      0s][START] target=whole-mouse-brain variant=macs-v11-norestart-acq-entropy budget=5.0% seed=2 | {"task": "membrane", "target": "whole-mouse-brain", "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": 8, "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}
+[11:03:34][      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.
+[11:03:34][      0s][DATA] whole-mouse-brain: train=5925 val=1250 test=1300 patches
+[11:03:36][      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, snemi=3.55
+[11:03:37][      3s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s
+[11:03:43][      9s][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
+[11:03:43][      9s][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
+[11:04:57][     83s][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
+[11:04:57][     83s][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
+[11:04:57][     83s][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
+[11:04:57][     83s][INIT] student initialised from teacher microns (weight 0.376)
+[11:05:00][     86s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[11:05:00][     86s][KD] distilling the mixture ensemble (prob space) into the student
+[11:09:02][    328s][KD] soft labels for 5925 patches from 8 teachers in 242s
+[11:11:59][    505s][KD] epoch 1/5 | patches=5925 | loss=0.6655 | 176s
+[11:14:55][    681s][KD] epoch 2/5 | patches=5925 | loss=0.4946 | 176s
+[11:17:51][    857s][KD] epoch 3/5 | patches=5925 | loss=0.4649 | 176s
+[11:20:47][   1033s][KD] epoch 4/5 | patches=5925 | loss=0.4584 | 176s
+[11:23:43][   1209s][KD] epoch 5/5 | patches=5925 | loss=0.4562 | 176s
+[11:23:49][   1215s][ROUND0] before any target label: val_dice=0.3932 val_vi=0.5071
+[11:23:49][   1215s][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
+[11:24:21][   1247s][AL] round-1 uncertainty quality on 5925 patches: {'spearman_uncertainty_error': -0.2505878322752116, 'auroc_top20_error': 0.3924166090497315}
+[11:25:43][   1329s][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
+[11:25:51][   1337s][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
+[11:25:52][   1338s][WEIGHTS] round 1: student kept (labeled Dice 0.292 vs best teacher oct-sfl 0.453); distilling the corrected ensemble into it
+[11:25:52][   1338s][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
+[11:26:52][   1398s][KD] soft labels for 5876 patches from 8 teachers in 61s
+[11:29:47][   1573s][KD] epoch 1/2 | patches=5876 | loss=0.2658 | 175s
+[11:32:43][   1749s][KD] epoch 2/2 | patches=5876 | loss=0.2580 | 175s
+[11:33:04][   1770s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 21s
+[11:33:07][   1773s][FT] round 1/6 epoch 1/5 | patches=49 | loss=1.4617
+[11:33:11][   1777s][FT] round 1/6 epoch 2/5 | patches=49 | loss=1.2396
+[11:33:15][   1781s][FT] round 1/6 epoch 3/5 | patches=49 | loss=1.1040
+[11:33:19][   1785s][FT] round 1/6 epoch 4/5 | patches=49 | loss=0.9489
+[11:33:23][   1789s][FT] round 1/6 epoch 5/5 | patches=49 | loss=1.1202
+[11:33:28][   1794s][AL] round 1/6 | labeled=49 | selection_seconds=113.6000 round_seconds=573.4000 restart_student_dice=0.2920 restart_teacher_dice=0.4531 val_dice=0.6300 val_vi=0.5107 val_ece=0.0330
+[11:33:59][   1825s][AL] round 2: 474/5876 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[11:34:15][   1841s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.302, uniform=0.301, top1=0.358, top2=0.360, top3=0.325 -> using top2
+[11:34:15][   1841s][WEIGHTS] round 2: teacher Dice on 98 labeled patches + 0 blank (1155835 fg px): c-elegans=0.325, cerebellum=0.262, flywire=0.116, berghia=0.268, microns=0.190, oct-glia=0.316, oct-sfl=0.358, snemi=0.269 | weights: c-elegans=0.054, cerebellum=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.946, snemi=0.000
+[11:34:30][   1856s][CONS] round 2: on the 49 new patches student Dice=0.338 vs ensemble 0.125 -> pseudo-labels from the student | consistency x0.00
+[11:34:33][   1859s][FT] round 2/6 epoch 1/5 | patches=98 | loss=1.0522
+[11:34:37][   1863s][FT] round 2/6 epoch 2/5 | patches=98 | loss=1.0992
+[11:34:40][   1866s][FT] round 2/6 epoch 3/5 | patches=98 | loss=0.9450
+[11:34:43][   1869s][FT] round 2/6 epoch 4/5 | patches=98 | loss=0.9605
+[11:34:47][   1873s][FT] round 2/6 epoch 5/5 | patches=98 | loss=0.9465
+[11:34:53][   1879s][AL] round 2/6 | labeled=98 | selection_seconds=31.1000 round_seconds=78.3000 pseudo_student_dice=0.3380 pseudo_ensemble_dice=0.1250 consistency_scale=0.0000 val_dice=0.6671 val_vi=0.4213 val_ece=0.0160
+[11:35:23][   1909s][AL] round 3: 768/5827 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[11:38:11][   2077s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.383, cerebellum=0.374, flywire=0.154, berghia=0.264, microns=0.332, oct-glia=0.446, oct-sfl=0.555, snemi=0.293 | blended with labeled Dice at 0.50
+[11:40:48][   2234s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.365, uniform=0.342, top1=0.415, top2=0.381, top3=0.383 -> using top1
+[11:40:48][   2234s][WEIGHTS] round 3: teacher Dice on 147 labeled patches + 0 blank (1901791 fg px): c-elegans=0.336, cerebellum=0.330, flywire=0.116, berghia=0.257, microns=0.231, oct-glia=0.343, oct-sfl=0.415, snemi=0.299 | 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
+[11:41:04][   2250s][CONS] round 3: on the 49 new patches student Dice=0.490 vs ensemble 0.116 -> pseudo-labels from the student | consistency x0.00
+[11:41:08][   2254s][FT] round 3/6 epoch 1/5 | patches=147 | loss=1.1947
+[11:41:13][   2259s][FT] round 3/6 epoch 2/5 | patches=147 | loss=1.0835
+[11:41:18][   2264s][FT] round 3/6 epoch 3/5 | patches=147 | loss=1.1484
+[11:41:23][   2269s][FT] round 3/6 epoch 4/5 | patches=147 | loss=1.0334
+[11:41:27][   2273s][FT] round 3/6 epoch 5/5 | patches=147 | loss=0.9493
+[11:41:33][   2279s][AL] round 3/6 | labeled=147 | selection_seconds=30.8000 round_seconds=394.7000 pseudo_student_dice=0.4902 pseudo_ensemble_dice=0.1158 consistency_scale=0.0000 val_dice=0.6500 val_vi=0.4666 val_ece=0.0266
+[11:42:04][   2310s][AL] round 4: 544/5778 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[11:44:54][   2480s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.412, cerebellum=0.380, flywire=0.153, berghia=0.304, microns=0.302, oct-glia=0.433, oct-sfl=0.529, snemi=0.329 | blended with labeled Dice at 0.50
+[11:47:34][   2640s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.364, uniform=0.322, top1=0.395, top2=0.389, top3=0.386 -> using top1
+[11:47:34][   2640s][WEIGHTS] round 4: teacher Dice on 196 labeled patches + 0 blank (2568324 fg px): c-elegans=0.352, cerebellum=0.355, flywire=0.110, berghia=0.253, microns=0.219, oct-glia=0.324, oct-sfl=0.395, snemi=0.321 | 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
+[11:47:49][   2655s][CONS] round 4: on the 49 new patches student Dice=0.536 vs ensemble 0.226 -> pseudo-labels from the student | consistency x0.12
+[11:48:02][   2668s][FT] round 4/6 epoch 1/5 | patches=196 | loss=1.1507
+[11:48:14][   2680s][FT] round 4/6 epoch 2/5 | patches=196 | loss=1.0430
+[11:48:27][   2693s][FT] round 4/6 epoch 3/5 | patches=196 | loss=1.0931
+[11:48:39][   2705s][FT] round 4/6 epoch 4/5 | patches=196 | loss=1.0568
+[11:48:51][   2717s][FT] round 4/6 epoch 5/5 | patches=196 | loss=0.9960
+[11:48:57][   2723s][AL] round 4/6 | labeled=196 | selection_seconds=30.6000 round_seconds=438.1000 pseudo_student_dice=0.5360 pseudo_ensemble_dice=0.2257 consistency_scale=0.1200 val_dice=0.6867 val_vi=0.4335 val_ece=0.0067
+[11:49:27][   2753s][AL] round 5: 766/5729 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[11:52:21][   2927s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.388, cerebellum=0.400, flywire=0.151, berghia=0.267, microns=0.312, oct-glia=0.430, oct-sfl=0.528, snemi=0.320 | blended with labeled Dice at 0.50
+[11:55:05][   3091s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.391, uniform=0.319, top1=0.394, top2=0.406, top3=0.405 -> using top3
+[11:55:05][   3091s][WEIGHTS] round 5: teacher Dice on 245 labeled patches + 0 blank (3582198 fg px): c-elegans=0.358, cerebellum=0.381, flywire=0.111, berghia=0.244, microns=0.218, oct-glia=0.316, oct-sfl=0.394, snemi=0.332 | weights: c-elegans=0.016, cerebellum=0.689, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.294, snemi=0.000
+[11:55:20][   3106s][CONS] round 5: on the 49 new patches student Dice=0.611 vs ensemble 0.448 -> pseudo-labels from the student | consistency x0.37
+[11:55:35][   3121s][FT] round 5/6 epoch 1/5 | patches=245 | loss=1.1906
+[11:55:50][   3136s][FT] round 5/6 epoch 2/5 | patches=245 | loss=1.1607
+[11:56:05][   3151s][FT] round 5/6 epoch 3/5 | patches=245 | loss=1.1352
+[11:56:21][   3167s][FT] round 5/6 epoch 4/5 | patches=245 | loss=1.0207
+[11:56:36][   3182s][FT] round 5/6 epoch 5/5 | patches=245 | loss=1.0371
+[11:56:42][   3188s][AL] round 5/6 | labeled=245 | selection_seconds=30.3000 round_seconds=458.7000 pseudo_student_dice=0.6111 pseudo_ensemble_dice=0.4482 consistency_scale=0.3700 val_dice=0.7010 val_vi=0.4148 val_ece=0.0096
+[11:57:12][   3218s][AL] round 6: 701/5680 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:00:09][   3395s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.400, cerebellum=0.392, flywire=0.148, berghia=0.280, microns=0.298, oct-glia=0.433, oct-sfl=0.526, snemi=0.334 | blended with labeled Dice at 0.50
+[12:02:57][   3563s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.409, uniform=0.308, top1=0.393, top2=0.411, top3=0.411 -> using mixture
+[12:02:57][   3563s][WEIGHTS] round 6: teacher Dice on 294 labeled patches + 0 blank (4504610 fg px): c-elegans=0.378, cerebellum=0.393, flywire=0.108, berghia=0.247, microns=0.210, oct-glia=0.315, oct-sfl=0.385, snemi=0.347 | weights: c-elegans=0.013, cerebellum=0.509, flywire=0.004, berghia=0.035, microns=0.120, oct-glia=0.123, oct-sfl=0.176, snemi=0.020
+[12:03:13][   3579s][CONS] round 6: on the 49 new patches student Dice=0.650 vs ensemble 0.298 -> pseudo-labels from the student | consistency x0.50
+[12:03:30][   3596s][FT] round 6/6 epoch 1/5 | patches=294 | loss=1.1132
+[12:03:48][   3614s][FT] round 6/6 epoch 2/5 | patches=294 | loss=1.0773
+[12:04:07][   3633s][FT] round 6/6 epoch 3/5 | patches=294 | loss=1.1183
+[12:04:25][   3651s][FT] round 6/6 epoch 4/5 | patches=294 | loss=1.0493
+[12:04:43][   3669s][FT] round 6/6 epoch 5/5 | patches=294 | loss=0.9931
+[12:04:48][   3674s][AL] round 6/6 | labeled=294 | selection_seconds=30.1000 round_seconds=481.0000 pseudo_student_dice=0.6501 pseudo_ensemble_dice=0.2981 consistency_scale=0.5000 val_dice=0.7047 val_vi=0.3934 val_ece=0.0079
+[12:04:59][   3685s][CONS] final consolidation: pseudo-labels refreshed from the student
+[12:05:17][   3703s][FT] final 1/10 epoch 1/1 | patches=294 | loss=0.9656
+[12:05:35][   3721s][FT] final 2/10 epoch 1/1 | patches=294 | loss=0.9334
+[12:05:53][   3739s][FT] final 3/10 epoch 1/1 | patches=294 | loss=0.9946
+[12:06:11][   3757s][FT] final 4/10 epoch 1/1 | patches=294 | loss=0.9482
+[12:06:29][   3775s][FT] final 5/10 epoch 1/1 | patches=294 | loss=0.9485
+[12:06:47][   3793s][FT] final 6/10 epoch 1/1 | patches=294 | loss=0.9475
+[12:07:05][   3811s][FT] final 7/10 epoch 1/1 | patches=294 | loss=0.8823
+[12:07:23][   3829s][FT] final 8/10 epoch 1/1 | patches=294 | loss=0.8645
+[12:07:41][   3847s][FT] final 9/10 epoch 1/1 | patches=294 | loss=0.8566
+[12:07:59][   3865s][FT] final 10/10 epoch 1/1 | patches=294 | loss=0.8279
+[12:08:05][   3871s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=190.4000 val_dice=0.7156 val_vi=0.3905 val_ece=0.0081
+[12:08:05][   3871s][WEIGHTS] final: c-elegans=0.013, cerebellum=0.509, flywire=0.004, berghia=0.035, microns=0.120, oct-glia=0.123, oct-sfl=0.176, snemi=0.020
+[12:09:06][   3932s][TEST] raw student (no EMA, no TTA): patch_dice=0.7055 patch_vi=0.3509
+[12:09:13][   3939s][TEST] final model without TTA: patch_dice=0.7055 patch_vi=0.3509
+[12:09:13][   3939s][RESOURCES] {"phases": {"setup": {"seconds": 1.48, "peak_cpu_rss_mb": 2192.2, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 84.15, "peak_cpu_rss_mb": 4183.1, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 1122.94, "peak_cpu_rss_mb": 5940.4, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 2655.43, "peak_cpu_rss_mb": 5940.4, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4626.0, "rounds": 6}, "inference": {"seconds": 68.47, "peak_cpu_rss_mb": 5940.4, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4626.0, "patches": 1300, "images": 52, "patches_per_second": 19.0, "seconds_per_image": 1.317}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5940.4, "total_seconds": 3932.5}
+[12:09:13][   3939s][TEST] patch_dice=0.7241 patch_iou=0.6435 patch_precision=0.7535 patch_recall=0.8265 patch_vi=0.3296 image_dice=0.8312 image_iou=0.7270 image_precision=0.8134 image_recall=0.8774 image_vi=0.3823 cal_ece=0.0033 cal_nll=0.1784 cal_brier=0.0514 cal_mean_confidence=0.9305 cal_accuracy=0.9315 cal_predicted_prevalence=0.2275 cal_true_prevalence=0.2108 cal_mean_probability=0.2430
+[12:09:14][   3940s][DONE] whole-mouse-brain macs-v11-norestart-acq-entropy b5 s2 | test patch Dice=0.7241 VI=0.3296 | image Dice=0.8312 VI=0.3823 | 3939s

macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-entropy/budget-5/seed-2/result.json

@@ -0,0 +1,785 @@
+{
+  "target": "whole-mouse-brain",
+  "variant": "macs-v11-norestart-acq-entropy",
+  "budget": 5.0,
+  "seed": 2,
+  "config": {
+    "task": "membrane",
+    "target": "whole-mouse-brain",
+    "variant": "macs-v11-norestart-acq-entropy",
+    "budget": 5.0,
+    "seed": 2,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "whole-mouse-brain",
+      "--variants",
+      "macs-v11-norestart-acq-entropy",
+      "--budgets",
+      "5",
+      "--seeds",
+      "2",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.00761798355588039,
+    0.2559331647259477,
+    0.03865987551029595,
+    0.0753340114213908,
+    0.3764144201177104,
+    0.13449626142769927,
+    0.09581294628140101,
+    0.015731336959674627
+  ],
+  "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": [
+      0.00761798355588039,
+      0.2559331647259477,
+      0.03865987551029595,
+      0.0753340114213908,
+      0.3764144201177104,
+      0.13449626142769927,
+      0.09581294628140101,
+      0.015731336959674627
+    ],
+    "per_source": [
+      0.14507434834977073,
+      0.041950586009985594,
+      0.0008814746660836073,
+      0.14183397467001313,
+      0.02035313133415202,
+      0.40237139544038975,
+      0.24338615668507596,
+      0.0041489328445293286
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.11311692308315584,
+      "mmd2_mixture": 0.12945845805399875,
+      "mmd2_uniform": 0.15528887887137932,
+      "mmd2_single_source": [
+        0.38071558003618083,
+        0.17697908713268734,
+        0.3194072222380283,
+        0.24037631453841346,
+        0.16266629855891124,
+        0.27345204518121097,
+        0.2563983777819251,
+        0.2660430684822974
+      ],
+      "median_squared_distance": 0.1859244048719758,
+      "gammas": [
+        1.3446325143390696,
+        2.689265028678139,
+        5.378530057356278,
+        10.757060114712557,
+        21.514120229425114
+      ],
+      "max_weight": 0.3764144201177104,
+      "entropy": 1.6341534970843021,
+      "effective_sources": 5.125117733158912,
+      "min_weight": 0.00761798355588039
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.3322006978705003,
+        0.46442575425639476,
+        0.8760620010344873,
+        0.3346079883794868,
+        0.5415020997962443,
+        0.22348725249801393,
+        0.277061938603162,
+        0.710986599609967
+      ],
+      "standardized": [
+        -0.6467264485566062,
+        -0.026349470463598257,
+        1.9049762391239466,
+        -0.6354318599803043,
+        0.33527934358594014,
+        -1.1567910351159896,
+        -0.9054279215228022,
+        1.130471152929414
+      ],
+      "tau": 0.5,
+      "max_weight": 0.40237139544038975,
+      "entropy": 1.508582930948667,
+      "effective_sources": 4.5203206470852155,
+      "min_weight": 0.0008814746660836073
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+    "seconds": 6.6
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.3857342957712184,
+      0.3164308639401021,
+      0.10997194512386699,
+      0.2588511091009309,
+      0.32881765823766074,
+      0.40721108437406883,
+      0.43978952859529485,
+      0.30535964718503905
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.3845120898706078,
+      "mixture@raw": 0.3998895285882327,
+      "mixture@logit": 0.38020955262748224,
+      "per_source": 0.4199047055822174,
+      "per_source@raw": 0.42186208777273215,
+      "per_source@logit": 0.41929804644233215,
+      "uniform": 0.41282903446624,
+      "uniform@raw": 0.3877116419741835,
+      "uniform@logit": 0.4054966159612228
+    },
+    "val_patches": 1024
+  },
+  "temperatures": [
+    2.575598955154419,
+    4.557764530181885,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    2.642308473587036,
+    3.547258138656616
+  ],
+  "top_source": "microns_basil",
+  "weight_stats": {
+    "max_weight": 0.3764144201177104,
+    "entropy": 1.6341534970843021,
+    "effective_sources": 5.125117733158912,
+    "min_weight": 0.00761798355588039
+  },
+  "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": 1122.9,
+  "round0_val": {
+    "val_dice": 0.39316518885694673,
+    "val_vi": 0.5071037992390497,
+    "val_ece": 0.13929538678771358
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 49,
+      "selection_seconds": 113.6,
+      "round_seconds": 573.4,
+      "teacher_dice_labeled": [
+        0.41667508794746294,
+        0.3592800679911643,
+        0.15312328284483975,
+        0.3102883118187754,
+        0.24828297584066997,
+        0.40443558048329065,
+        0.45309956020060516,
+        0.3151859425557686
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 745730,
+      "evidence_nonempty_patches": 43,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.2919523581210136,
+      "restart_teacher_dice": 0.45309956020060516,
+      "weights": [
+        0.052345161843924994,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.947654838156075,
+        0.0
+      ],
+      "ensemble_choice": "top2",
+      "candidate_dice": {
+        "mixture": 0.41707512159255195,
+        "uniform": 0.39628150854804123,
+        "top1": 0.45309956020060516,
+        "top2": 0.4560005583432433,
+        "top3": 0.42428559646215114
+      },
+      "val_dice": 0.6299640712246265,
+      "val_vi": 0.5106772900495546,
+      "val_ece": 0.03300236752641396
+    },
+    {
+      "round": 2,
+      "n_labeled": 98,
+      "selection_seconds": 31.1,
+      "round_seconds": 78.3,
+      "teacher_dice_labeled": [
+        0.3249594717244935,
+        0.2618002400057593,
+        0.11626361986650778,
+        0.2682125036278144,
+        0.19005829309477,
+        0.3157203697046871,
+        0.357571670301185,
+        0.26891050108232195
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1155835,
+      "evidence_nonempty_patches": 80,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0542688419808356,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.9457311580191643,
+        0.0
+      ],
+      "ensemble_choice": "top2",
+      "candidate_dice": {
+        "mixture": 0.30164075804164814,
+        "uniform": 0.30147095589245054,
+        "top1": 0.357571670301185,
+        "top2": 0.35967682146094637,
+        "top3": 0.32455408079063386
+      },
+      "pseudo_student_dice": 0.33799498785878834,
+      "pseudo_ensemble_dice": 0.12497048992244641,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.6671426429611568,
+      "val_vi": 0.42133914138201256,
+      "val_ece": 0.015966444291345954
+    },
+    {
+      "round": 3,
+      "n_labeled": 147,
+      "selection_seconds": 30.8,
+      "round_seconds": 394.7,
+      "teacher_dice_labeled": [
+        0.3362206587862502,
+        0.33035776451296633,
+        0.11570615054827438,
+        0.25699455697450335,
+        0.23065137965115518,
+        0.34275950768617824,
+        0.4148087923555059,
+        0.29939804919802215
+      ],
+      "teacher_dice_pool": [
+        0.38312200607089864,
+        0.37357687854150773,
+        0.1536548607926944,
+        0.2644659774961156,
+        0.33231233949642547,
+        0.44608789466873955,
+        0.5548695967709021,
+        0.29333183491515874
+      ],
+      "evidence_fg_pixels": 1901791,
+      "evidence_nonempty_patches": 106,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        1.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.3647942449410265,
+        "uniform": 0.3424460851717929,
+        "top1": 0.4148087923555059,
+        "top2": 0.3811854830612246,
+        "top3": 0.3826582850005112
+      },
+      "pseudo_student_dice": 0.4902302504374955,
+      "pseudo_ensemble_dice": 0.11578052428171628,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.6500437319549233,
+      "val_vi": 0.46661771310585776,
+      "val_ece": 0.026616772063627447
+    },
+    {
+      "round": 4,
+      "n_labeled": 196,
+      "selection_seconds": 30.6,
+      "round_seconds": 438.1,
+      "teacher_dice_labeled": [
+        0.3519483730790272,
+        0.3550619667303971,
+        0.11032225058098022,
+        0.2531933417507679,
+        0.2190875017687,
+        0.3241839148029759,
+        0.3952944285319471,
+        0.32148791585466907
+      ],
+      "teacher_dice_pool": [
+        0.4121643222363875,
+        0.37964937638880697,
+        0.15275562810191143,
+        0.3040249259871205,
+        0.3021430411408676,
+        0.4332669406483501,
+        0.5293982426240719,
+        0.3288904127260634
+      ],
+      "evidence_fg_pixels": 2568324,
+      "evidence_nonempty_patches": 139,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        1.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.36402664300994514,
+        "uniform": 0.3222085755550682,
+        "top1": 0.3952944285319471,
+        "top2": 0.3887335829311641,
+        "top3": 0.3864764612355511
+      },
+      "pseudo_student_dice": 0.53600076651133,
+      "pseudo_ensemble_dice": 0.22566554473495054,
+      "pseudo_source": "student",
+      "consistency_scale": 0.12,
+      "val_dice": 0.6867063671616198,
+      "val_vi": 0.4334551684265923,
+      "val_ece": 0.006732057576624399
+    },
+    {
+      "round": 5,
+      "n_labeled": 245,
+      "selection_seconds": 30.3,
+      "round_seconds": 458.7,
+      "teacher_dice_labeled": [
+        0.3584230095199944,
+        0.38096205524839705,
+        0.11117752910496465,
+        0.2439638336867353,
+        0.21809780395525596,
+        0.31570369170261414,
+        0.3941231445083987,
+        0.33228651701690143
+      ],
+      "teacher_dice_pool": [
+        0.3875614767122367,
+        0.400028641768668,
+        0.15084672589212084,
+        0.2667770048170303,
+        0.31216759610745726,
+        0.4302824637590289,
+        0.5282474877153657,
+        0.31952189062440434
+      ],
+      "evidence_fg_pixels": 3582198,
+      "evidence_nonempty_patches": 178,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.01637661851566227,
+        0.6892814651914534,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.29434191629288425,
+        0.0
+      ],
+      "ensemble_choice": "top3",
+      "candidate_dice": {
+        "mixture": 0.3906889930951955,
+        "uniform": 0.31898344193918804,
+        "top1": 0.3941231445083987,
+        "top2": 0.40592475485919916,
+        "top3": 0.40537019629518667
+      },
+      "pseudo_student_dice": 0.6111341036273602,
+      "pseudo_ensemble_dice": 0.448249058234845,
+      "pseudo_source": "student",
+      "consistency_scale": 0.37,
+      "val_dice": 0.7010087760930706,
+      "val_vi": 0.4148022188492324,
+      "val_ece": 0.009600339954623855
+    },
+    {
+      "round": 6,
+      "n_labeled": 294,
+      "selection_seconds": 30.1,
+      "round_seconds": 481.0,
+      "teacher_dice_labeled": [
+        0.37813906679059617,
+        0.3929658426455666,
+        0.10759623232507429,
+        0.2470899709358254,
+        0.2097623786143264,
+        0.3154036114080012,
+        0.3847248817593051,
+        0.3467007071990453
+      ],
+      "teacher_dice_pool": [
+        0.4000509789602617,
+        0.39176571714136516,
+        0.1482235973977921,
+        0.27994162679950685,
+        0.2977878407705155,
+        0.4326847309976392,
+        0.5262451617634437,
+        0.3340916004034753
+      ],
+      "evidence_fg_pixels": 4504610,
+      "evidence_nonempty_patches": 224,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.013069321687079801,
+        0.5092504873046682,
+        0.004433236159290681,
+        0.03485500253066859,
+        0.11990260091834273,
+        0.1232158588333893,
+        0.1755654455535038,
+        0.01970804701305701
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.40943043039142457,
+        "uniform": 0.30811234119371467,
+        "top1": 0.3929658426455666,
+        "top2": 0.41077618458479637,
+        "top3": 0.41068862692388175
+      },
+      "pseudo_student_dice": 0.6500897437229997,
+      "pseudo_ensemble_dice": 0.2980986095351606,
+      "pseudo_source": "student",
+      "consistency_scale": 0.5,
+      "val_dice": 0.7046566133521246,
+      "val_vi": 0.39338360372380154,
+      "val_ece": 0.007922761299779176
+    },
+    {
+      "round": 7,
+      "n_labeled": 294,
+      "selection_seconds": 0.0,
+      "round_seconds": 190.4,
+      "final_consolidation": true,
+      "val_dice": 0.715600054385459,
+      "val_vi": 0.3905030362479474,
+      "val_ece": 0.0081281918504199
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": -0.2505878322752116,
+    "auroc_top20_error": 0.3924166090497315
+  },
+  "al_seconds": 2655.4,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.013069321687079801,
+    0.5092504873046682,
+    0.004433236159290681,
+    0.03485500253066859,
+    0.11990260091834273,
+    0.1232158588333893,
+    0.1755654455535038,
+    0.01970804701305701
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.5092504873046682,
+    "entropy": 1.4364938946277095,
+    "effective_sources": 4.205923524457596,
+    "min_weight": 0.004433236159290681
+  },
+  "top_source_final": "dhanyasi-P14-mouse-cerebellum",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.7054563808799655,
+    "patch_iou": 0.6213611844092362,
+    "patch_precision": 0.7248983941572255,
+    "patch_recall": 0.8271769452900526,
+    "patch_vi": 0.35090675928415266,
+    "n_patches": 1300,
+    "cal_ece": 0.00821402980845419,
+    "cal_nll": 0.19145299976349267,
+    "cal_brier": 0.05514656907793308,
+    "cal_mean_confidence": 0.9346074646634128,
+    "cal_accuracy": 0.9264100060096154,
+    "cal_predicted_prevalence": 0.23219651442307693,
+    "cal_true_prevalence": 0.21078049879807692,
+    "cal_mean_probability": 0.244326443916351
+  },
+  "test_notta": {
+    "patch_dice": 0.7054563808799655,
+    "patch_iou": 0.6213611844092362,
+    "patch_precision": 0.7248983941572255,
+    "patch_recall": 0.8271769452900526,
+    "patch_vi": 0.35090675928415266,
+    "n_patches": 1300,
+    "cal_ece": 0.00821402980845419,
+    "cal_nll": 0.19145299976349267,
+    "cal_brier": 0.05514656907793308,
+    "cal_mean_confidence": 0.9346074646634128,
+    "cal_accuracy": 0.9264100060096154,
+    "cal_predicted_prevalence": 0.23219651442307693,
+    "cal_true_prevalence": 0.21078049879807692,
+    "cal_mean_probability": 0.244326443916351
+  },
+  "test": {
+    "patch_dice": 0.7240810547947243,
+    "patch_iou": 0.6434686049972183,
+    "patch_precision": 0.7535363328863108,
+    "patch_recall": 0.8265486522250965,
+    "patch_vi": 0.3295530513344462,
+    "n_patches": 1300,
+    "image_dice": 0.8312241136800578,
+    "image_iou": 0.726982790269668,
+    "image_precision": 0.8133640751108497,
+    "image_recall": 0.8773653978839356,
+    "image_vi": 0.38227195684253606,
+    "n_images": 52,
+    "cal_ece": 0.00327916983898916,
+    "cal_nll": 0.17840624343436598,
+    "cal_brier": 0.05143879160628608,
+    "cal_mean_confidence": 0.9305477422689723,
+    "cal_accuracy": 0.9314761117788461,
+    "cal_predicted_prevalence": 0.2275420673076923,
+    "cal_true_prevalence": 0.21078049879807692,
+    "cal_mean_probability": 0.2429929328227741
+  },
+  "eval_seconds": 68.5,
+  "total_seconds": 3939.2,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 1.48,
+        "peak_cpu_rss_mb": 2192.2,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 84.15,
+        "peak_cpu_rss_mb": 4183.1,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 1122.94,
+        "peak_cpu_rss_mb": 5940.4,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4618.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2655.43,
+        "peak_cpu_rss_mb": 5940.4,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4626.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 68.47,
+        "peak_cpu_rss_mb": 5940.4,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4626.0,
+        "patches": 1300,
+        "images": 52,
+        "patches_per_second": 19.0,
+        "seconds_per_image": 1.317
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5940.4,
+    "total_seconds": 3932.5
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/whole-mouse-brain/macs-v11-norestart-acq-entropy_b5_s2.pth"
+}

macs_v2_r26/runs/membrane/whole-mouse-brain/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
+1,49,113.6,573.4,"[0.41667508794746294, 0.3592800679911643, 0.15312328284483975, 0.3102883118187754, 0.24828297584066997, 0.40443558048329065, 0.45309956020060516, 0.3151859425557686]",,745730,43,True,kept,0.2919523581210136,0.45309956020060516,"[0.052345161843924994, 0.0, 0.0, 0.0, 0.0, 0.0, 0.947654838156075, 0.0]",top2,"{'mixture': 0.41707512159255195, 'uniform': 0.39628150854804123, 'top1': 0.45309956020060516, 'top2': 0.4560005583432433, 'top3': 0.42428559646215114}",0.6299640712246265,0.5106772900495546,0.03300236752641396,,,,,
+2,98,31.1,78.3,"[0.3249594717244935, 0.2618002400057593, 0.11626361986650778, 0.2682125036278144, 0.19005829309477, 0.3157203697046871, 0.357571670301185, 0.26891050108232195]",,1155835,80,False,,,,"[0.0542688419808356, 0.0, 0.0, 0.0, 0.0, 0.0, 0.9457311580191643, 0.0]",top2,"{'mixture': 0.30164075804164814, 'uniform': 0.30147095589245054, 'top1': 0.357571670301185, 'top2': 0.35967682146094637, 'top3': 0.32455408079063386}",0.6671426429611568,0.42133914138201256,0.015966444291345954,0.33799498785878834,0.12497048992244641,student,0.0,
+3,147,30.8,394.7,"[0.3362206587862502, 0.33035776451296633, 0.11570615054827438, 0.25699455697450335, 0.23065137965115518, 0.34275950768617824, 0.4148087923555059, 0.29939804919802215]","[0.38312200607089864, 0.37357687854150773, 0.1536548607926944, 0.2644659774961156, 0.33231233949642547, 0.44608789466873955, 0.5548695967709021, 0.29333183491515874]",1901791,106,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]",top1,"{'mixture': 0.3647942449410265, 'uniform': 0.3424460851717929, 'top1': 0.4148087923555059, 'top2': 0.3811854830612246, 'top3': 0.3826582850005112}",0.6500437319549233,0.46661771310585776,0.026616772063627447,0.4902302504374955,0.11578052428171628,student,0.0,
+4,196,30.6,438.1,"[0.3519483730790272, 0.3550619667303971, 0.11032225058098022, 0.2531933417507679, 0.2190875017687, 0.3241839148029759, 0.3952944285319471, 0.32148791585466907]","[0.4121643222363875, 0.37964937638880697, 0.15275562810191143, 0.3040249259871205, 0.3021430411408676, 0.4332669406483501, 0.5293982426240719, 0.3288904127260634]",2568324,139,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]",top1,"{'mixture': 0.36402664300994514, 'uniform': 0.3222085755550682, 'top1': 0.3952944285319471, 'top2': 0.3887335829311641, 'top3': 0.3864764612355511}",0.6867063671616198,0.4334551684265923,0.006732057576624399,0.53600076651133,0.22566554473495054,student,0.12,
+5,245,30.3,458.7,"[0.3584230095199944, 0.38096205524839705, 0.11117752910496465, 0.2439638336867353, 0.21809780395525596, 0.31570369170261414, 0.3941231445083987, 0.33228651701690143]","[0.3875614767122367, 0.400028641768668, 0.15084672589212084, 0.2667770048170303, 0.31216759610745726, 0.4302824637590289, 0.5282474877153657, 0.31952189062440434]",3582198,178,False,,,,"[0.01637661851566227, 0.6892814651914534, 0.0, 0.0, 0.0, 0.0, 0.29434191629288425, 0.0]",top3,"{'mixture': 0.3906889930951955, 'uniform': 0.31898344193918804, 'top1': 0.3941231445083987, 'top2': 0.40592475485919916, 'top3': 0.40537019629518667}",0.7010087760930706,0.4148022188492324,0.009600339954623855,0.6111341036273602,0.448249058234845,student,0.37,
+6,294,30.1,481.0,"[0.37813906679059617, 0.3929658426455666, 0.10759623232507429, 0.2470899709358254, 0.2097623786143264, 0.3154036114080012, 0.3847248817593051, 0.3467007071990453]","[0.4000509789602617, 0.39176571714136516, 0.1482235973977921, 0.27994162679950685, 0.2977878407705155, 0.4326847309976392, 0.5262451617634437, 0.3340916004034753]",4504610,224,False,,,,"[0.013069321687079801, 0.5092504873046682, 0.004433236159290681, 0.03485500253066859, 0.11990260091834273, 0.1232158588333893, 0.1755654455535038, 0.01970804701305701]",mixture,"{'mixture': 0.40943043039142457, 'uniform': 0.30811234119371467, 'top1': 0.3929658426455666, 'top2': 0.41077618458479637, 'top3': 0.41068862692388175}",0.7046566133521246,0.39338360372380154,0.007922761299779176,0.6500897437229997,0.2980986095351606,student,0.5,
+7,294,0.0,190.4,,,,,,,,,,,,0.715600054385459,0.3905030362479474,0.0081281918504199,,,,,True

macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-random/budget-5/seed-1/log.txt

@@ -0,0 +1,103 @@
+[02:45:19][      0s][START] target=whole-mouse-brain variant=macs-v11-norestart-acq-random budget=5.0% seed=1 | {"task": "membrane", "target": "whole-mouse-brain", "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": 8, "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}
+[02:45:19][      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.
+[02:45:21][      2s][DATA] whole-mouse-brain: train=5925 val=1250 test=1300 patches
+[02:45:23][      4s][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
+[02:45:23][      4s][FEATURES] 72 domain x encoder feature sets (72 cached) | n=512/domain | 0s
+[02:45:29][     11s][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
+[02:45:29][     11s][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
+[02:46:44][     85s][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
+[02:46:44][     85s][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
+[02:46:44][     85s][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
+[02:46:44][     85s][INIT] student initialised from teacher microns (weight 0.376)
+[02:46:59][    100s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[02:46:59][    100s][KD] distilling the mixture ensemble (prob space) into the student
+[02:51:00][    341s][KD] soft labels for 5925 patches from 8 teachers in 241s
+[02:53:56][    517s][KD] epoch 1/5 | patches=5925 | loss=0.6724 | 176s
+[02:56:52][    694s][KD] epoch 2/5 | patches=5925 | loss=0.4930 | 176s
+[02:59:49][    870s][KD] epoch 3/5 | patches=5925 | loss=0.4658 | 176s
+[03:02:45][   1046s][KD] epoch 4/5 | patches=5925 | loss=0.4598 | 176s
+[03:05:41][   1222s][KD] epoch 5/5 | patches=5925 | loss=0.4555 | 176s
+[03:05:47][   1228s][ROUND0] before any target label: val_dice=0.3878 val_vi=0.5059
+[03:05:47][   1228s][AL] patches=5925 (pool 5925 after dropping blank; free blank negatives: True) | budget=296 (5.0%) | rounds=6 | per_round=49 | acquisition=random | first_round=stratified
+[03:06:18][   1259s][AL] round-1 uncertainty quality on 5925 patches: {'spearman_uncertainty_error': -0.006734909767982264, 'auroc_top20_error': 0.5056221403265146}
+[03:07:40][   1341s][AL] round 1: stratified cold start: 25 patches from the membrane-rich half (ensemble prevalence >= 0.214) + 24 from the membrane-poor half of 2000 candidates
+[03:07:48][   1349s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.415, uniform=0.393, top1=0.460, top2=0.464, top3=0.490 -> using top3
+[03:07:48][   1349s][WEIGHTS] round 1: student kept (labeled Dice 0.314 vs best teacher oct-sfl 0.460); distilling the corrected ensemble into it
+[03:07:48][   1349s][WEIGHTS] round 1: teacher Dice on 49 labeled patches + 0 blank (774194 fg px): c-elegans=0.431, cerebellum=0.372, flywire=0.134, berghia=0.291, microns=0.284, oct-glia=0.371, oct-sfl=0.460, snemi=0.295 | weights: c-elegans=0.027, cerebellum=0.511, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.462, snemi=0.000
+[03:09:19][   1440s][KD] soft labels for 5876 patches from 8 teachers in 90s
+[03:12:13][   1615s][KD] epoch 1/2 | patches=5876 | loss=0.4892 | 175s
+[03:15:08][   1789s][KD] epoch 2/2 | patches=5876 | loss=0.4853 | 174s
+[03:15:39][   1820s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 31s
+[03:15:43][   1824s][FT] round 1/6 epoch 1/5 | patches=49 | loss=1.3866
+[03:15:46][   1827s][FT] round 1/6 epoch 2/5 | patches=49 | loss=1.2764
+[03:15:50][   1831s][FT] round 1/6 epoch 3/5 | patches=49 | loss=1.2397
+[03:15:54][   1835s][FT] round 1/6 epoch 4/5 | patches=49 | loss=1.2255
+[03:15:57][   1838s][FT] round 1/6 epoch 5/5 | patches=49 | loss=1.2378
+[03:16:03][   1844s][AL] round 1/6 | labeled=49 | selection_seconds=112.9000 round_seconds=610.4000 restart_student_dice=0.3142 restart_teacher_dice=0.4601 val_dice=0.6513 val_vi=0.4647 val_ece=0.0896
+[03:16:18][   1859s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.436, uniform=0.403, top1=0.475, top2=0.478, top3=0.452 -> using top2
+[03:16:18][   1859s][WEIGHTS] round 2: teacher Dice on 98 labeled patches + 0 blank (1631452 fg px): c-elegans=0.442, cerebellum=0.374, flywire=0.140, berghia=0.295, microns=0.276, oct-glia=0.378, oct-sfl=0.475, snemi=0.324 | weights: c-elegans=0.054, cerebellum=0.000, flywire=0.000, berghia=0.000, microns=0.000, oct-glia=0.000, oct-sfl=0.946, snemi=0.000
+[03:16:34][   1875s][CONS] round 2: on the 49 new patches student Dice=0.828 vs ensemble 0.491 -> pseudo-labels from the student | consistency x1.00
+[03:16:40][   1881s][FT] round 2/6 epoch 1/5 | patches=98 | loss=1.0456
+[03:16:47][   1888s][FT] round 2/6 epoch 2/5 | patches=98 | loss=0.9731
+[03:16:53][   1894s][FT] round 2/6 epoch 3/5 | patches=98 | loss=0.9812
+[03:17:00][   1901s][FT] round 2/6 epoch 4/5 | patches=98 | loss=0.8799
+[03:17:06][   1907s][FT] round 2/6 epoch 5/5 | patches=98 | loss=0.9492
+[03:17:12][   1913s][AL] round 2/6 | labeled=98 | selection_seconds=0.2000 round_seconds=63.2000 pseudo_student_dice=0.8275 pseudo_ensemble_dice=0.4908 consistency_scale=1.0000 val_dice=0.6900 val_vi=0.3935 val_ece=0.0466
+[03:19:59][   2080s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.396, cerebellum=0.379, flywire=0.160, berghia=0.279, microns=0.333, oct-glia=0.456, oct-sfl=0.545, snemi=0.289 | blended with labeled Dice at 0.50
+[03:22:36][   2237s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.461, uniform=0.447, top1=0.505, top2=0.477, top3=0.480 -> using top1
+[03:22:36][   2237s][WEIGHTS] round 3: teacher Dice on 147 labeled patches + 0 blank (2470847 fg px): c-elegans=0.421, cerebellum=0.375, flywire=0.151, berghia=0.280, microns=0.307, oct-glia=0.421, oct-sfl=0.505, snemi=0.310 | 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
+[03:22:51][   2253s][CONS] round 3: on the 49 new patches student Dice=0.848 vs ensemble 0.445 -> pseudo-labels from the student | consistency x1.00
+[03:23:01][   2262s][FT] round 3/6 epoch 1/5 | patches=147 | loss=0.8251
+[03:23:10][   2271s][FT] round 3/6 epoch 2/5 | patches=147 | loss=0.8697
+[03:23:19][   2281s][FT] round 3/6 epoch 3/5 | patches=147 | loss=0.8617
+[03:23:29][   2290s][FT] round 3/6 epoch 4/5 | patches=147 | loss=0.8097
+[03:23:38][   2299s][FT] round 3/6 epoch 5/5 | patches=147 | loss=0.8147
+[03:23:44][   2305s][AL] round 3/6 | labeled=147 | selection_seconds=0.2000 round_seconds=386.1000 pseudo_student_dice=0.8483 pseudo_ensemble_dice=0.4447 consistency_scale=1.0000 val_dice=0.6998 val_vi=0.3811 val_ece=0.0066
+[03:26:34][   2475s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.394, cerebellum=0.379, flywire=0.159, berghia=0.289, microns=0.326, oct-glia=0.460, oct-sfl=0.534, snemi=0.295 | blended with labeled Dice at 0.50
+[03:29:13][   2635s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.456, uniform=0.446, top1=0.497, top2=0.467, top3=0.469 -> using top1
+[03:29:13][   2635s][WEIGHTS] round 4: teacher Dice on 196 labeled patches + 0 blank (3255460 fg px): c-elegans=0.417, cerebellum=0.378, flywire=0.151, berghia=0.283, microns=0.306, oct-glia=0.423, oct-sfl=0.497, snemi=0.314 | 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
+[03:29:29][   2650s][CONS] round 4: on the 49 new patches student Dice=0.889 vs ensemble 0.444 -> pseudo-labels from the student | consistency x1.00
+[03:29:41][   2662s][FT] round 4/6 epoch 1/5 | patches=196 | loss=0.7037
+[03:29:53][   2674s][FT] round 4/6 epoch 2/5 | patches=196 | loss=0.6993
+[03:30:05][   2687s][FT] round 4/6 epoch 3/5 | patches=196 | loss=0.6914
+[03:30:18][   2699s][FT] round 4/6 epoch 4/5 | patches=196 | loss=0.6840
+[03:30:30][   2711s][FT] round 4/6 epoch 5/5 | patches=196 | loss=0.6446
+[03:30:36][   2717s][AL] round 4/6 | labeled=196 | selection_seconds=0.2000 round_seconds=405.8000 pseudo_student_dice=0.8894 pseudo_ensemble_dice=0.4436 consistency_scale=1.0000 val_dice=0.7284 val_vi=0.3506 val_ece=0.0106
+[03:33:29][   2890s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.397, cerebellum=0.379, flywire=0.154, berghia=0.284, microns=0.330, oct-glia=0.463, oct-sfl=0.543, snemi=0.305 | blended with labeled Dice at 0.50
+[03:36:12][   3054s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.462, uniform=0.453, top1=0.509, top2=0.477, top3=0.479 -> using top1
+[03:36:12][   3054s][WEIGHTS] round 5: teacher Dice on 245 labeled patches + 0 blank (3796031 fg px): c-elegans=0.407, cerebellum=0.378, flywire=0.149, berghia=0.280, microns=0.311, oct-glia=0.434, oct-sfl=0.509, snemi=0.312 | 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
+[03:36:28][   3069s][CONS] round 5: on the 49 new patches student Dice=0.815 vs ensemble 0.562 -> pseudo-labels from the student | consistency x1.00
+[03:36:43][   3084s][FT] round 5/6 epoch 1/5 | patches=245 | loss=0.6626
+[03:36:58][   3099s][FT] round 5/6 epoch 2/5 | patches=245 | loss=0.6472
+[03:37:13][   3114s][FT] round 5/6 epoch 3/5 | patches=245 | loss=0.6301
+[03:37:28][   3129s][FT] round 5/6 epoch 4/5 | patches=245 | loss=0.6262
+[03:37:43][   3144s][FT] round 5/6 epoch 5/5 | patches=245 | loss=0.6015
+[03:37:49][   3150s][AL] round 5/6 | labeled=245 | selection_seconds=0.2000 round_seconds=427.2000 pseudo_student_dice=0.8150 pseudo_ensemble_dice=0.5618 consistency_scale=1.0000 val_dice=0.7361 val_vi=0.3468 val_ece=0.0179
+[03:40:46][   3327s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: c-elegans=0.411, cerebellum=0.385, flywire=0.156, berghia=0.288, microns=0.321, oct-glia=0.459, oct-sfl=0.540, snemi=0.320 | blended with labeled Dice at 0.50
+[03:43:33][   3494s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.461, uniform=0.448, top1=0.509, top2=0.476, top3=0.478 -> using top1
+[03:43:33][   3494s][WEIGHTS] round 6: teacher Dice on 294 labeled patches + 0 blank (4607078 fg px): c-elegans=0.413, cerebellum=0.381, flywire=0.150, berghia=0.286, microns=0.307, oct-glia=0.431, oct-sfl=0.509, snemi=0.321 | 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
+[03:43:48][   3509s][CONS] round 6: on the 49 new patches student Dice=0.801 vs ensemble 0.486 -> pseudo-labels from the student | consistency x1.00
+[03:44:06][   3527s][FT] round 6/6 epoch 1/5 | patches=294 | loss=0.6618
+[03:44:24][   3545s][FT] round 6/6 epoch 2/5 | patches=294 | loss=0.6521
+[03:44:41][   3563s][FT] round 6/6 epoch 3/5 | patches=294 | loss=0.5995
+[03:44:59][   3581s][FT] round 6/6 epoch 4/5 | patches=294 | loss=0.5934
+[03:45:17][   3598s][FT] round 6/6 epoch 5/5 | patches=294 | loss=0.5715
+[03:45:23][   3604s][AL] round 6/6 | labeled=294 | selection_seconds=0.2000 round_seconds=448.7000 pseudo_student_dice=0.8012 pseudo_ensemble_dice=0.4864 consistency_scale=1.0000 val_dice=0.7406 val_vi=0.3158 val_ece=0.0196
+[03:45:33][   3614s][CONS] final consolidation: pseudo-labels refreshed from the student
+[03:45:51][   3632s][FT] final 1/10 epoch 1/1 | patches=294 | loss=0.5299
+[03:46:09][   3650s][FT] final 2/10 epoch 1/1 | patches=294 | loss=0.5280
+[03:46:27][   3668s][FT] final 3/10 epoch 1/1 | patches=294 | loss=0.5384
+[03:46:45][   3686s][FT] final 4/10 epoch 1/1 | patches=294 | loss=0.5264
+[03:47:03][   3704s][FT] final 5/10 epoch 1/1 | patches=294 | loss=0.5097
+[03:47:20][   3722s][FT] final 6/10 epoch 1/1 | patches=294 | loss=0.5057
+[03:47:38][   3740s][FT] final 7/10 epoch 1/1 | patches=294 | loss=0.4599
+[03:47:56][   3757s][FT] final 8/10 epoch 1/1 | patches=294 | loss=0.4586
+[03:48:14][   3775s][FT] final 9/10 epoch 1/1 | patches=294 | loss=0.4521
+[03:48:32][   3793s][FT] final 10/10 epoch 1/1 | patches=294 | loss=0.4527
+[03:48:38][   3799s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=189.0000 val_dice=0.7512 val_vi=0.3181 val_ece=0.0216
+[03:48:38][   3799s][WEIGHTS] final: 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
+[03:49:39][   3860s][TEST] raw student (no EMA, no TTA): patch_dice=0.7474 patch_vi=0.2821
+[03:49:47][   3868s][TEST] final model without TTA: patch_dice=0.7474 patch_vi=0.2821
+[03:49:47][   3868s][RESOURCES] {"phases": {"setup": {"seconds": 1.47, "peak_cpu_rss_mb": 2183.9, "peak_gpu_alloc_mb": 1066.3, "peak_gpu_reserved_mb": 1086.0}, "features_and_weights": {"seconds": 95.65, "peak_cpu_rss_mb": 4183.8, "peak_gpu_alloc_mb": 2820.1, "peak_gpu_reserved_mb": 3212.0}, "distillation": {"seconds": 1122.62, "peak_cpu_rss_mb": 5939.8, "peak_gpu_alloc_mb": 4300.6, "peak_gpu_reserved_mb": 4618.0, "kd_epochs": 5}, "active_learning": {"seconds": 2570.92, "peak_cpu_rss_mb": 5939.8, "peak_gpu_alloc_mb": 4420.9, "peak_gpu_reserved_mb": 4626.0, "rounds": 6}, "inference": {"seconds": 68.69, "peak_cpu_rss_mb": 5939.8, "peak_gpu_alloc_mb": 2623.4, "peak_gpu_reserved_mb": 4626.0, "patches": 1300, "images": 52, "patches_per_second": 18.9, "seconds_per_image": 1.321}}, "peak_gpu_alloc_mb": 4420.9, "peak_cpu_rss_mb": 5939.8, "total_seconds": 3859.3}
+[03:49:47][   3868s][TEST] patch_dice=0.7625 patch_iou=0.6916 patch_precision=0.8210 patch_recall=0.8202 patch_vi=0.2623 image_dice=0.8727 image_iou=0.7844 image_precision=0.8902 image_recall=0.8710 image_vi=0.3002 cal_ece=0.0089 cal_nll=0.1345 cal_brier=0.0377 cal_mean_confidence=0.9586 cal_accuracy=0.9497 cal_predicted_prevalence=0.2075 cal_true_prevalence=0.2108 cal_mean_probability=0.2073
+[03:49:47][   3869s][DONE] whole-mouse-brain macs-v11-norestart-acq-random b5 s1 | test patch Dice=0.7625 VI=0.2623 | image Dice=0.8727 VI=0.3002 | 3868s

macs_v2_r26/runs/membrane/whole-mouse-brain/macs-v11-norestart-acq-random/budget-5/seed-1/result.json

@@ -0,0 +1,785 @@
+{
+  "target": "whole-mouse-brain",
+  "variant": "macs-v11-norestart-acq-random",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "membrane",
+    "target": "whole-mouse-brain",
+    "variant": "macs-v11-norestart-acq-random",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "whole-mouse-brain",
+      "--variants",
+      "macs-v11-norestart-acq-random",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "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": [
+    0.00761798355588039,
+    0.2559331647259477,
+    0.03865987551029595,
+    0.0753340114213908,
+    0.3764144201177104,
+    0.13449626142769927,
+    0.09581294628140101,
+    0.015731336959674627
+  ],
+  "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": [
+      0.00761798355588039,
+      0.2559331647259477,
+      0.03865987551029595,
+      0.0753340114213908,
+      0.3764144201177104,
+      0.13449626142769927,
+      0.09581294628140101,
+      0.015731336959674627
+    ],
+    "per_source": [
+      0.14507434834977073,
+      0.041950586009985594,
+      0.0008814746660836073,
+      0.14183397467001313,
+      0.02035313133415202,
+      0.40237139544038975,
+      0.24338615668507596,
+      0.0041489328445293286
+    ],
+    "uniform": [
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125,
+      0.125
+    ],
+    "mixture_info": {
+      "objective": 0.11311692308315584,
+      "mmd2_mixture": 0.12945845805399875,
+      "mmd2_uniform": 0.15528887887137932,
+      "mmd2_single_source": [
+        0.38071558003618083,
+        0.17697908713268734,
+        0.3194072222380283,
+        0.24037631453841346,
+        0.16266629855891124,
+        0.27345204518121097,
+        0.2563983777819251,
+        0.2660430684822974
+      ],
+      "median_squared_distance": 0.1859244048719758,
+      "gammas": [
+        1.3446325143390696,
+        2.689265028678139,
+        5.378530057356278,
+        10.757060114712557,
+        21.514120229425114
+      ],
+      "max_weight": 0.3764144201177104,
+      "entropy": 1.6341534970843021,
+      "effective_sources": 5.125117733158912,
+      "min_weight": 0.00761798355588039
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.3322006978705003,
+        0.46442575425639476,
+        0.8760620010344873,
+        0.3346079883794868,
+        0.5415020997962443,
+        0.22348725249801393,
+        0.277061938603162,
+        0.710986599609967
+      ],
+      "standardized": [
+        -0.6467264485566062,
+        -0.026349470463598257,
+        1.9049762391239466,
+        -0.6354318599803043,
+        0.33527934358594014,
+        -1.1567910351159896,
+        -0.9054279215228022,
+        1.130471152929414
+      ],
+      "tau": 0.5,
+      "max_weight": 0.40237139544038975,
+      "entropy": 1.508582930948667,
+      "effective_sources": 4.5203206470852155,
+      "min_weight": 0.0008814746660836073
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "octopus-vulgaris-vertical-lobe-glia-deep-neuropil",
+    "seconds": 6.6
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.3857342957712184,
+      0.3164308639401021,
+      0.10997194512386699,
+      0.2588511091009309,
+      0.32881765823766074,
+      0.40721108437406883,
+      0.43978952859529485,
+      0.30535964718503905
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.3845120898706078,
+      "mixture@raw": 0.3998895285882327,
+      "mixture@logit": 0.38020955262748224,
+      "per_source": 0.4199047055822174,
+      "per_source@raw": 0.42186208777273215,
+      "per_source@logit": 0.41929804644233215,
+      "uniform": 0.41282903446624,
+      "uniform@raw": 0.3877116419741835,
+      "uniform@logit": 0.4054966159612228
+    },
+    "val_patches": 1024
+  },
+  "temperatures": [
+    2.575598955154419,
+    4.557764530181885,
+    1.9903231859207153,
+    8.0,
+    1.9153105020523071,
+    2.5512874126434326,
+    2.642308473587036,
+    3.547258138656616
+  ],
+  "top_source": "microns_basil",
+  "weight_stats": {
+    "max_weight": 0.3764144201177104,
+    "entropy": 1.6341534970843021,
+    "effective_sources": 5.125117733158912,
+    "min_weight": 0.00761798355588039
+  },
+  "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": 1122.6,
+  "round0_val": {
+    "val_dice": 0.38775386975069404,
+    "val_vi": 0.5058585267638911,
+    "val_ece": 0.13684088848086562
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 49,
+      "selection_seconds": 112.9,
+      "round_seconds": 610.4,
+      "teacher_dice_labeled": [
+        0.4308437371432156,
+        0.3720464899554072,
+        0.13400588777866665,
+        0.2906832527572902,
+        0.28441244487328066,
+        0.3714445937083196,
+        0.46009083351660035,
+        0.2953659731768761
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 774194,
+      "evidence_nonempty_patches": 43,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.3141552482233243,
+      "restart_teacher_dice": 0.46009083351660035,
+      "weights": [
+        0.027388975830910146,
+        0.511103822265685,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.46150720190340494,
+        0.0
+      ],
+      "ensemble_choice": "top3",
+      "candidate_dice": {
+        "mixture": 0.4149047036603486,
+        "uniform": 0.3931444911550303,
+        "top1": 0.46009083351660035,
+        "top2": 0.46380715367111625,
+        "top3": 0.49022779815240064
+      },
+      "val_dice": 0.6512893174377166,
+      "val_vi": 0.4646900588599244,
+      "val_ece": 0.0896417007453064
+    },
+    {
+      "round": 2,
+      "n_labeled": 98,
+      "selection_seconds": 0.2,
+      "round_seconds": 63.2,
+      "teacher_dice_labeled": [
+        0.4424330801851372,
+        0.37408893009446964,
+        0.13996203868108806,
+        0.29503398267841263,
+        0.2762732338980843,
+        0.3782590298845219,
+        0.47481626651490355,
+        0.3238189071307628
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1631452,
+      "evidence_nonempty_patches": 85,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0543864996701088,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.9456135003298912,
+        0.0
+      ],
+      "ensemble_choice": "top2",
+      "candidate_dice": {
+        "mixture": 0.4356210342800381,
+        "uniform": 0.40298612721687643,
+        "top1": 0.47481626651490355,
+        "top2": 0.47797929634414377,
+        "top3": 0.4516145396104275
+      },
+      "pseudo_student_dice": 0.8275029820365678,
+      "pseudo_ensemble_dice": 0.49077896830410456,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.6899674120835443,
+      "val_vi": 0.3934530539606663,
+      "val_ece": 0.046598598690968274
+    },
+    {
+      "round": 3,
+      "n_labeled": 147,
+      "selection_seconds": 0.2,
+      "round_seconds": 386.1,
+      "teacher_dice_labeled": [
+        0.4208794202637337,
+        0.37488266988173913,
+        0.150732512981738,
+        0.280229848328845,
+        0.30736600511815093,
+        0.42137673253425356,
+        0.504630320454826,
+        0.3103878885056744
+      ],
+      "teacher_dice_pool": [
+        0.3964222900856918,
+        0.37934827125456433,
+        0.15986144969132293,
+        0.2785786219247478,
+        0.33299853995204853,
+        0.45638475100222214,
+        0.5447891954337357,
+        0.2891779812953886
+      ],
+      "evidence_fg_pixels": 2470847,
+      "evidence_nonempty_patches": 129,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        1.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.46107124265136823,
+        "uniform": 0.4470064837241954,
+        "top1": 0.504630320454826,
+        "top2": 0.4774312960056267,
+        "top3": 0.47954494564006567
+      },
+      "pseudo_student_dice": 0.8483273711237067,
+      "pseudo_ensemble_dice": 0.4447035907635971,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.6997838980735469,
+      "val_vi": 0.38111564948839316,
+      "val_ece": 0.006551502400643054
+    },
+    {
+      "round": 4,
+      "n_labeled": 196,
+      "selection_seconds": 0.2,
+      "round_seconds": 405.8,
+      "teacher_dice_labeled": [
+        0.41735012268474536,
+        0.37774968729190495,
+        0.15129542306014881,
+        0.2833705136853922,
+        0.3061931648198789,
+        0.4232066747177724,
+        0.49688937580258463,
+        0.314315337148227
+      ],
+      "teacher_dice_pool": [
+        0.3943931472745527,
+        0.37855105448483334,
+        0.15943182820909615,
+        0.2892929128772673,
+        0.3257258256062749,
+        0.4601414870055501,
+        0.5342473085263869,
+        0.2950977113256563
+      ],
+      "evidence_fg_pixels": 3255460,
+      "evidence_nonempty_patches": 173,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        1.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.45595241469488296,
+        "uniform": 0.44581060629678865,
+        "top1": 0.49688937580258463,
+        "top2": 0.46677430571359046,
+        "top3": 0.4690154172263659
+      },
+      "pseudo_student_dice": 0.8894096619935941,
+      "pseudo_ensemble_dice": 0.4435759384979648,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.7284449934265951,
+      "val_vi": 0.35062932040639544,
+      "val_ece": 0.01061670729460243
+    },
+    {
+      "round": 5,
+      "n_labeled": 245,
+      "selection_seconds": 0.2,
+      "round_seconds": 427.2,
+      "teacher_dice_labeled": [
+        0.40692426871897414,
+        0.3776928989394102,
+        0.14920549941794428,
+        0.2795811790533572,
+        0.31140825070370776,
+        0.43368676727852196,
+        0.5091293274911578,
+        0.3121180397885524
+      ],
+      "teacher_dice_pool": [
+        0.3970244948724663,
+        0.37928996723211816,
+        0.15356131391556158,
+        0.28419405601396547,
+        0.3296259383047535,
+        0.46253273706373266,
+        0.542636728401153,
+        0.3050203065494246
+      ],
+      "evidence_fg_pixels": 3796031,
+      "evidence_nonempty_patches": 213,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        1.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.4624701194922756,
+        "uniform": 0.45307112409178935,
+        "top1": 0.5091293274911578,
+        "top2": 0.4770389469575239,
+        "top3": 0.4788482169595161
+      },
+      "pseudo_student_dice": 0.8150319479999724,
+      "pseudo_ensemble_dice": 0.5618041860169222,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.7360976720287663,
+      "val_vi": 0.3467971465172063,
+      "val_ece": 0.01788726699061594
+    },
+    {
+      "round": 6,
+      "n_labeled": 294,
+      "selection_seconds": 0.2,
+      "round_seconds": 448.7,
+      "teacher_dice_labeled": [
+        0.4131523804861451,
+        0.38148291144188595,
+        0.14959132590158677,
+        0.2859449485429898,
+        0.3070357015540196,
+        0.4312325733658978,
+        0.5087118932311401,
+        0.321087704901554
+      ],
+      "teacher_dice_pool": [
+        0.4114461558819054,
+        0.3849032866661021,
+        0.1555940064347239,
+        0.2883347744854101,
+        0.32091747898050904,
+        0.4585541603506631,
+        0.5398959859343434,
+        0.31990600159271626
+      ],
+      "evidence_fg_pixels": 4607078,
+      "evidence_nonempty_patches": 257,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        1.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.46080784643360345,
+        "uniform": 0.4478514590552423,
+        "top1": 0.5087118932311401,
+        "top2": 0.47603001745833207,
+        "top3": 0.47780834110988624
+      },
+      "pseudo_student_dice": 0.8012167018453978,
+      "pseudo_ensemble_dice": 0.4863777827338713,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.7405788014844665,
+      "val_vi": 0.31581319789840717,
+      "val_ece": 0.019612879670904602
+    },
+    {
+      "round": 7,
+      "n_labeled": 294,
+      "selection_seconds": 0.0,
+      "round_seconds": 189.0,
+      "final_consolidation": true,
+      "val_dice": 0.7512189727699001,
+      "val_vi": 0.3181009168739638,
+      "val_ece": 0.021623735484845367
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": -0.006734909767982264,
+    "auroc_top20_error": 0.5056221403265146
+  },
+  "al_seconds": 2570.9,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.0,
+    0.0,
+    0.0,
+    0.0,
+    0.0,
+    0.0,
+    1.0,
+    0.0
+  ],
+  "weight_stats_final": {
+    "max_weight": 1.0,
+    "entropy": -0.0,
+    "effective_sources": 1.0,
+    "min_weight": 0.0
+  },
+  "top_source_final": "octopus-vulgaris-vertical-lobe-sfltract",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.7474120376241677,
+    "patch_iou": 0.6728028364833846,
+    "patch_precision": 0.7935101241758076,
+    "patch_recall": 0.8178482491923542,
+    "patch_vi": 0.28209865954645047,
+    "n_patches": 1300,
+    "cal_ece": 0.0176318543517363,
+    "cal_nll": 0.1534192941139603,
+    "cal_brier": 0.04106024181296851,
+    "cal_mean_confidence": 0.9636918002651976,
+    "cal_accuracy": 0.9460599459134615,
+    "cal_predicted_prevalence": 0.2094666466346154,
+    "cal_true_prevalence": 0.21078049879807692,
+    "cal_mean_probability": 0.20719564276506328
+  },
+  "test_notta": {
+    "patch_dice": 0.7474120376241677,
+    "patch_iou": 0.6728028364833846,
+    "patch_precision": 0.7935101241758076,
+    "patch_recall": 0.8178482491923542,
+    "patch_vi": 0.28209865954645047,
+    "n_patches": 1300,
+    "cal_ece": 0.0176318543517363,
+    "cal_nll": 0.1534192941139603,
+    "cal_brier": 0.04106024181296851,
+    "cal_mean_confidence": 0.9636918002651976,
+    "cal_accuracy": 0.9460599459134615,
+    "cal_predicted_prevalence": 0.2094666466346154,
+    "cal_true_prevalence": 0.21078049879807692,
+    "cal_mean_probability": 0.20719564276506328
+  },
+  "test": {
+    "patch_dice": 0.7625086122496587,
+    "patch_iou": 0.6915794905331696,
+    "patch_precision": 0.8210280221083377,
+    "patch_recall": 0.8201574493133729,
+    "patch_vi": 0.26227260743302666,
+    "n_patches": 1300,
+    "image_dice": 0.8726720894721708,
+    "image_iou": 0.784447861167505,
+    "image_precision": 0.8901989738983472,
+    "image_recall": 0.87096714633107,
+    "image_vi": 0.3001978081447007,
+    "n_images": 52,
+    "cal_ece": 0.008869097294647845,
+    "cal_nll": 0.1345410909698496,
+    "cal_brier": 0.037687427717699104,
+    "cal_mean_confidence": 0.9585626016781935,
+    "cal_accuracy": 0.9497438401442307,
+    "cal_predicted_prevalence": 0.20746243990384616,
+    "cal_true_prevalence": 0.21078049879807692,
+    "cal_mean_probability": 0.2073494734025003
+  },
+  "eval_seconds": 68.7,
+  "total_seconds": 3867.7,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 1.47,
+        "peak_cpu_rss_mb": 2183.9,
+        "peak_gpu_alloc_mb": 1066.3,
+        "peak_gpu_reserved_mb": 1086.0
+      },
+      "features_and_weights": {
+        "seconds": 95.65,
+        "peak_cpu_rss_mb": 4183.8,
+        "peak_gpu_alloc_mb": 2820.1,
+        "peak_gpu_reserved_mb": 3212.0
+      },
+      "distillation": {
+        "seconds": 1122.62,
+        "peak_cpu_rss_mb": 5939.8,
+        "peak_gpu_alloc_mb": 4300.6,
+        "peak_gpu_reserved_mb": 4618.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2570.92,
+        "peak_cpu_rss_mb": 5939.8,
+        "peak_gpu_alloc_mb": 4420.9,
+        "peak_gpu_reserved_mb": 4626.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 68.69,
+        "peak_cpu_rss_mb": 5939.8,
+        "peak_gpu_alloc_mb": 2623.4,
+        "peak_gpu_reserved_mb": 4626.0,
+        "patches": 1300,
+        "images": 52,
+        "patches_per_second": 18.9,
+        "seconds_per_image": 1.321
+      }
+    },
+    "peak_gpu_alloc_mb": 4420.9,
+    "peak_cpu_rss_mb": 5939.8,
+    "total_seconds": 3859.3
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/whole-mouse-brain/macs-v11-norestart-acq-random_b5_s1.pth"
+}

macs_v2_r26/runs/membrane/whole-mouse-brain/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
+1,49,112.9,610.4,"[0.4308437371432156, 0.3720464899554072, 0.13400588777866665, 0.2906832527572902, 0.28441244487328066, 0.3714445937083196, 0.46009083351660035, 0.2953659731768761]",,774194,43,True,kept,0.3141552482233243,0.46009083351660035,"[0.027388975830910146, 0.511103822265685, 0.0, 0.0, 0.0, 0.0, 0.46150720190340494, 0.0]",top3,"{'mixture': 0.4149047036603486, 'uniform': 0.3931444911550303, 'top1': 0.46009083351660035, 'top2': 0.46380715367111625, 'top3': 0.49022779815240064}",0.6512893174377166,0.4646900588599244,0.0896417007453064,,,,,
+2,98,0.2,63.2,"[0.4424330801851372, 0.37408893009446964, 0.13996203868108806, 0.29503398267841263, 0.2762732338980843, 0.3782590298845219, 0.47481626651490355, 0.3238189071307628]",,1631452,85,False,,,,"[0.0543864996701088, 0.0, 0.0, 0.0, 0.0, 0.0, 0.9456135003298912, 0.0]",top2,"{'mixture': 0.4356210342800381, 'uniform': 0.40298612721687643, 'top1': 0.47481626651490355, 'top2': 0.47797929634414377, 'top3': 0.4516145396104275}",0.6899674120835443,0.3934530539606663,0.046598598690968274,0.8275029820365678,0.49077896830410456,student,1.0,
+3,147,0.2,386.1,"[0.4208794202637337, 0.37488266988173913, 0.150732512981738, 0.280229848328845, 0.30736600511815093, 0.42137673253425356, 0.504630320454826, 0.3103878885056744]","[0.3964222900856918, 0.37934827125456433, 0.15986144969132293, 0.2785786219247478, 0.33299853995204853, 0.45638475100222214, 0.5447891954337357, 0.2891779812953886]",2470847,129,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]",top1,"{'mixture': 0.46107124265136823, 'uniform': 0.4470064837241954, 'top1': 0.504630320454826, 'top2': 0.4774312960056267, 'top3': 0.47954494564006567}",0.6997838980735469,0.38111564948839316,0.006551502400643054,0.8483273711237067,0.4447035907635971,student,1.0,
+4,196,0.2,405.8,"[0.41735012268474536, 0.37774968729190495, 0.15129542306014881, 0.2833705136853922, 0.3061931648198789, 0.4232066747177724, 0.49688937580258463, 0.314315337148227]","[0.3943931472745527, 0.37855105448483334, 0.15943182820909615, 0.2892929128772673, 0.3257258256062749, 0.4601414870055501, 0.5342473085263869, 0.2950977113256563]",3255460,173,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]",top1,"{'mixture': 0.45595241469488296, 'uniform': 0.44581060629678865, 'top1': 0.49688937580258463, 'top2': 0.46677430571359046, 'top3': 0.4690154172263659}",0.7284449934265951,0.35062932040639544,0.01061670729460243,0.8894096619935941,0.4435759384979648,student,1.0,
+5,245,0.2,427.2,"[0.40692426871897414, 0.3776928989394102, 0.14920549941794428, 0.2795811790533572, 0.31140825070370776, 0.43368676727852196, 0.5091293274911578, 0.3121180397885524]","[0.3970244948724663, 0.37928996723211816, 0.15356131391556158, 0.28419405601396547, 0.3296259383047535, 0.46253273706373266, 0.542636728401153, 0.3050203065494246]",3796031,213,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]",top1,"{'mixture': 0.4624701194922756, 'uniform': 0.45307112409178935, 'top1': 0.5091293274911578, 'top2': 0.4770389469575239, 'top3': 0.4788482169595161}",0.7360976720287663,0.3467971465172063,0.01788726699061594,0.8150319479999724,0.5618041860169222,student,1.0,
+6,294,0.2,448.7,"[0.4131523804861451, 0.38148291144188595, 0.14959132590158677, 0.2859449485429898, 0.3070357015540196, 0.4312325733658978, 0.5087118932311401, 0.321087704901554]","[0.4114461558819054, 0.3849032866661021, 0.1555940064347239, 0.2883347744854101, 0.32091747898050904, 0.4585541603506631, 0.5398959859343434, 0.31990600159271626]",4607078,257,False,,,,"[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]",top1,"{'mixture': 0.46080784643360345, 'uniform': 0.4478514590552423, 'top1': 0.5087118932311401, 'top2': 0.47603001745833207, 'top3': 0.47780834110988624}",0.7405788014844665,0.31581319789840717,0.019612879670904602,0.8012167018453978,0.4863777827338713,student,1.0,
+7,294,0.0,189.0,,,,,,,,,,,,0.7512189727699001,0.3181009168739638,0.021623735484845367,,,,,True

macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-1/log.txt

@@ -0,0 +1,105 @@
+[23:08:10][      0s][START] target=drosophila-vnc variant=macs-v11-norestart-acq-laplace budget=5.0% seed=1 | {"task": "mitochondria", "target": "drosophila-vnc", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 8, "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}
+[23:08:10][      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.
+[23:08:11][      0s][DATA] drosophila-vnc: train=686 val=147 test=147 patches
+[23:08:20][     10s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, multiclass=2.60
+[23:08:27][     17s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 7s
+[23:08:28][     18s][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
+[23:08:28][     18s][WEIGHTS] per-source: fly-mito=0.039, kasthuri=0.943, lucchi=0.014, multiclass=0.004 | min-MMD source (NeuroADDA rule) = kasthuri
+[23:08:35][     24s][TEACHERS] target-val Dice: fly-mito=0.278, kasthuri=0.183, lucchi=0.058, multiclass=0.017
+[23:08:35][     24s][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
+[23:08:35][     24s][WEIGHTS] using mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020
+[23:08:35][     24s][INIT] student initialised from teacher kasthuri (weight 0.838)
+[23:08:38][     27s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[23:08:38][     27s][KD] distilling the mixture ensemble (prob space) into the student
+[23:08:51][     41s][KD] soft labels for 686 patches from 4 teachers in 14s
+[23:09:12][     61s][KD] epoch 1/5 | patches=686 | loss=0.2005 | 20s
+[23:09:32][     81s][KD] epoch 2/5 | patches=686 | loss=0.1322 | 20s
+[23:09:52][    101s][KD] epoch 3/5 | patches=686 | loss=0.1259 | 20s
+[23:10:12][    121s][KD] epoch 4/5 | patches=686 | loss=0.1216 | 20s
+[23:10:32][    142s][KD] epoch 5/5 | patches=686 | loss=0.1204 | 20s
+[23:10:33][    143s][ROUND0] before any target label: val_dice=0.1919 val_vi=0.2213
+[23:10:33][    143s][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
+[23:10:40][    150s][AL] round-1 uncertainty quality on 686 patches: {'spearman_uncertainty_error': 0.1281956487029202, 'auroc_top20_error': 0.6150618810253443}
+[23:10:54][    164s][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
+[23:10:55][    165s][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
+[23:10:55][    165s][WEIGHTS] round 1: student kept (labeled Dice 0.123 vs best teacher fly-mito 0.312); distilling the corrected ensemble into it
+[23:10:55][    165s][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
+[23:10:59][    169s][KD] soft labels for 681 patches from 4 teachers in 4s
+[23:11:19][    189s][KD] epoch 1/2 | patches=681 | loss=0.2934 | 20s
+[23:11:39][    209s][KD] epoch 2/2 | patches=681 | loss=0.2402 | 20s
+[23:11:43][    212s][CONS] pseudo-labels for 681 unlabeled patches from the current ensemble in 4s
+[23:11:43][    213s][FT] round 1/6 epoch 1/5 | patches=5 | loss=1.4014
+[23:11:44][    214s][FT] round 1/6 epoch 2/5 | patches=5 | loss=1.3138
+[23:11:45][    215s][FT] round 1/6 epoch 3/5 | patches=5 | loss=1.2332
+[23:11:46][    216s][FT] round 1/6 epoch 4/5 | patches=5 | loss=1.1817
+[23:11:47][    217s][FT] round 1/6 epoch 5/5 | patches=5 | loss=1.2248
+[23:11:48][    218s][AL] round 1/6 | labeled=5 | selection_seconds=21.2000 round_seconds=74.2000 restart_student_dice=0.1231 restart_teacher_dice=0.3119 val_dice=0.2369 val_vi=0.2879 val_ece=0.0156
+[23:11:55][    225s][AL] round 2: 14/681 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:11:57][    226s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.141, uniform=0.025, top1=0.172, top2=0.142, top3=0.142 -> using top1
+[23:11:57][    226s][WEIGHTS] round 2: teacher Dice on 10 labeled patches + 0 blank (17002 fg px): fly-mito=0.172, kasthuri=0.135, lucchi=0.001, multiclass=0.003 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[23:11:58][    227s][CONS] round 2: on the 5 new patches student Dice=0.042 vs ensemble 0.040 -> pseudo-labels from the ensemble | consistency x1.00 | judgement set below the foreground gate: source kept
+[23:11:59][    229s][FT] round 2/6 epoch 1/5 | patches=10 | loss=1.1973
+[23:12:00][    230s][FT] round 2/6 epoch 2/5 | patches=10 | loss=1.1801
+[23:12:01][    231s][FT] round 2/6 epoch 3/5 | patches=10 | loss=1.1914
+[23:12:03][    232s][FT] round 2/6 epoch 4/5 | patches=10 | loss=1.1851
+[23:12:04][    234s][FT] round 2/6 epoch 5/5 | patches=10 | loss=1.1409
+[23:12:05][    235s][AL] round 2/6 | labeled=10 | selection_seconds=7.4000 round_seconds=16.0000 pseudo_student_dice=0.0421 pseudo_ensemble_dice=0.0401 consistency_scale=1.0000 val_dice=0.2866 val_vi=0.2713 val_ece=0.0474
+[23:12:12][    242s][AL] round 3: 31/676 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:12:14][    244s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.146, uniform=0.018, top1=0.182, top2=0.147, top3=0.146 -> using top1
+[23:12:14][    244s][WEIGHTS] round 3: teacher Dice on 15 labeled patches + 0 blank (24575 fg px): fly-mito=0.182, kasthuri=0.137, lucchi=0.001, multiclass=0.002 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[23:12:15][    245s][CONS] round 3: on the 5 new patches student Dice=0.203 vs ensemble 0.197 -> pseudo-labels from the ensemble | consistency x1.00
+[23:12:16][    246s][FT] round 3/6 epoch 1/5 | patches=15 | loss=1.1297
+[23:12:18][    248s][FT] round 3/6 epoch 2/5 | patches=15 | loss=1.1117
+[23:12:19][    249s][FT] round 3/6 epoch 3/5 | patches=15 | loss=1.0964
+[23:12:21][    250s][FT] round 3/6 epoch 4/5 | patches=15 | loss=1.0400
+[23:12:22][    252s][FT] round 3/6 epoch 5/5 | patches=15 | loss=1.0601
+[23:12:23][    253s][AL] round 3/6 | labeled=15 | selection_seconds=7.3000 round_seconds=17.0000 pseudo_student_dice=0.2030 pseudo_ensemble_dice=0.1971 consistency_scale=1.0000 val_dice=0.4354 val_vi=0.3016 val_ece=0.0098
+[23:12:30][    260s][AL] round 4: 15/671 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:12:33][    262s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.128, uniform=0.017, top1=0.147, top2=0.129, top3=0.129 -> using top1
+[23:12:33][    262s][WEIGHTS] round 4: teacher Dice on 20 labeled patches + 0 blank (25508 fg px): fly-mito=0.147, kasthuri=0.122, lucchi=0.001, multiclass=0.002 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[23:12:33][    263s][CONS] round 4: on the 5 new patches student Dice=0.026 vs ensemble 0.011 -> pseudo-labels from the ensemble | consistency x1.00 | judgement set below the foreground gate: source kept
+[23:12:35][    265s][FT] round 4/6 epoch 1/5 | patches=20 | loss=1.0636
+[23:12:37][    267s][FT] round 4/6 epoch 2/5 | patches=20 | loss=1.0478
+[23:12:39][    268s][FT] round 4/6 epoch 3/5 | patches=20 | loss=1.0256
+[23:12:41][    270s][FT] round 4/6 epoch 4/5 | patches=20 | loss=1.0184
+[23:12:42][    272s][FT] round 4/6 epoch 5/5 | patches=20 | loss=1.0131
+[23:12:43][    273s][AL] round 4/6 | labeled=20 | selection_seconds=7.3000 round_seconds=19.3000 pseudo_student_dice=0.0263 pseudo_ensemble_dice=0.0112 consistency_scale=1.0000 val_dice=0.5388 val_vi=0.1580 val_ece=0.0072
+[23:12:51][    280s][AL] round 5: 99/666 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:12:53][    283s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.117, uniform=0.011, top1=0.150, top2=0.118, top3=0.117 -> using top1
+[23:12:53][    283s][WEIGHTS] round 5: teacher Dice on 25 labeled patches + 0 blank (41206 fg px): fly-mito=0.150, kasthuri=0.112, lucchi=0.001, multiclass=0.001 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[23:12:58][    287s][CONS] round 5: on the 5 new patches student Dice=0.588 vs ensemble 0.157 -> pseudo-labels from the student | consistency x0.29
+[23:13:00][    290s][FT] round 5/6 epoch 1/5 | patches=25 | loss=0.7865
+[23:13:02][    292s][FT] round 5/6 epoch 2/5 | patches=25 | loss=0.7621
+[23:13:04][    294s][FT] round 5/6 epoch 3/5 | patches=25 | loss=0.7242
+[23:13:06][    296s][FT] round 5/6 epoch 4/5 | patches=25 | loss=0.6684
+[23:13:09][    298s][FT] round 5/6 epoch 5/5 | patches=25 | loss=0.6240
+[23:13:10][    299s][AL] round 5/6 | labeled=25 | selection_seconds=7.2000 round_seconds=25.2000 pseudo_student_dice=0.5880 pseudo_ensemble_dice=0.1565 consistency_scale=0.2930 val_dice=0.6531 val_vi=0.1316 val_ece=0.0065
+[23:13:17][    307s][AL] round 6: 116/661 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:13:45][    335s][WEIGHTS] round 6: teacher agreement with the student on 656 unlabeled patches: fly-mito=0.387, kasthuri=0.467, lucchi=0.009, multiclass=0.001 | blended with labeled Dice at 0.50
+[23:14:11][    360s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.284, uniform=0.034, top1=0.288, top2=0.284, top3=0.284 -> using mixture
+[23:14:11][    360s][WEIGHTS] round 6: teacher Dice on 30 labeled patches + 0 blank (67637 fg px): fly-mito=0.288, kasthuri=0.281, lucchi=0.005, multiclass=0.001 | weights: fly-mito=0.130, kasthuri=0.867, lucchi=0.002, multiclass=0.001
+[23:14:15][    365s][CONS] round 6: on the 5 new patches student Dice=0.411 vs ensemble 0.063 -> pseudo-labels from the student | consistency x0.00
+[23:14:16][    366s][FT] round 6/6 epoch 1/5 | patches=30 | loss=0.6123
+[23:14:17][    367s][FT] round 6/6 epoch 2/5 | patches=30 | loss=0.5794
+[23:14:19][    368s][FT] round 6/6 epoch 3/5 | patches=30 | loss=0.5905
+[23:14:20][    370s][FT] round 6/6 epoch 4/5 | patches=30 | loss=0.5808
+[23:14:21][    371s][FT] round 6/6 epoch 5/5 | patches=30 | loss=0.5486
+[23:14:22][    372s][AL] round 6/6 | labeled=30 | selection_seconds=7.2000 round_seconds=71.4000 pseudo_student_dice=0.4114 pseudo_ensemble_dice=0.0634 consistency_scale=0.0000 val_dice=0.6087 val_vi=0.1785 val_ece=0.0197
+[23:14:26][    375s][CONS] final consolidation: pseudo-labels refreshed from the student
+[23:14:27][    377s][FT] final 1/10 epoch 1/1 | patches=30 | loss=0.5766
+[23:14:28][    378s][FT] final 2/10 epoch 1/1 | patches=30 | loss=0.4982
+[23:14:29][    379s][FT] final 3/10 epoch 1/1 | patches=30 | loss=0.5141
+[23:14:31][    380s][FT] final 4/10 epoch 1/1 | patches=30 | loss=0.5012
+[23:14:32][    381s][FT] final 5/10 epoch 1/1 | patches=30 | loss=0.4460
+[23:14:33][    383s][FT] final 6/10 epoch 1/1 | patches=30 | loss=0.4272
+[23:14:34][    384s][FT] final 7/10 epoch 1/1 | patches=30 | loss=0.4768
+[23:14:35][    385s][FT] final 8/10 epoch 1/1 | patches=30 | loss=0.4484
+[23:14:37][    386s][FT] final 9/10 epoch 1/1 | patches=30 | loss=0.4726
+[23:14:38][    387s][FT] final 10/10 epoch 1/1 | patches=30 | loss=0.4304
+[23:14:39][    389s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=15.6000 val_dice=0.6910 val_vi=0.1376 val_ece=0.0151
+[23:14:39][    389s][WEIGHTS] final: fly-mito=0.130, kasthuri=0.867, lucchi=0.002, multiclass=0.001
+[23:14:46][    396s][TEST] raw student (no EMA, no TTA): patch_dice=0.7909 patch_vi=0.1394
+[23:14:47][    397s][TEST] final model without TTA: patch_dice=0.7909 patch_vi=0.1394
+[23:14:47][    397s][RESOURCES] {"phases": {"setup": {"seconds": 8.5, "peak_cpu_rss_mb": 2190.7, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 17.69, "peak_cpu_rss_mb": 4161.9, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 114.25, "peak_cpu_rss_mb": 4369.3, "peak_gpu_alloc_mb": 3826.7, "peak_gpu_reserved_mb": 4168.0, "kd_epochs": 5}, "active_learning": {"seconds": 246.0, "peak_cpu_rss_mb": 4472.1, "peak_gpu_alloc_mb": 3947.0, "peak_gpu_reserved_mb": 4178.0, "rounds": 6}, "inference": {"seconds": 8.46, "peak_cpu_rss_mb": 4472.1, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4178.0, "patches": 147, "images": 3, "patches_per_second": 17.4, "seconds_per_image": 2.82}}, "peak_gpu_alloc_mb": 3947.0, "peak_cpu_rss_mb": 4472.1, "total_seconds": 394.9}
+[23:14:47][    397s][TEST] patch_dice=0.7877 patch_iou=0.6951 patch_precision=0.7281 patch_recall=0.9348 patch_vi=0.1348 image_dice=0.8425 image_iou=0.7285 image_precision=0.7509 image_recall=0.9602 image_vi=0.1642 cal_ece=0.0075 cal_nll=0.0525 cal_brier=0.0154 cal_mean_confidence=0.9862 cal_accuracy=0.9787 cal_predicted_prevalence=0.0745 cal_true_prevalence=0.0594 cal_mean_probability=0.0762
+[23:14:48][    397s][DONE] drosophila-vnc macs-v11-norestart-acq-laplace b5 s1 | test patch Dice=0.7877 VI=0.1348 | image Dice=0.8425 VI=0.1642 | 397s

macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-1/result.json

@@ -0,0 +1,652 @@
+{
+  "target": "drosophila-vnc",
+  "variant": "macs-v11-norestart-acq-laplace",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "mitochondria",
+    "target": "drosophila-vnc",
+    "variant": "macs-v11-norestart-acq-laplace",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "fce9b15",
+  "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "drosophila-vnc",
+      "--variants",
+      "macs-v11-norestart-acq-laplace",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--task",
+      "mitochondria",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "commit": "fce9b15",
+    "dirty": true
+  },
+  "sources": [
+    "fly-brain-mito-benchmark",
+    "kasthuri-pp",
+    "lucchi-pp",
+    "multiclass"
+  ],
+  "alphas": [
+    0.11690492516893997,
+    0.8375897330633602,
+    0.0256027087898516,
+    0.019902632977848233
+  ],
+  "weights_all": {
+    "sources": [
+      "fly-brain-mito-benchmark",
+      "kasthuri-pp",
+      "lucchi-pp",
+      "multiclass"
+    ],
+    "mixture": [
+      0.11690492516893997,
+      0.8375897330633602,
+      0.0256027087898516,
+      0.019902632977848233
+    ],
+    "per_source": [
+      0.038599207620062845,
+      0.9432861284385984,
+      0.013712296406123614,
+      0.0044023675352152245
+    ],
+    "uniform": [
+      0.25,
+      0.25,
+      0.25,
+      0.25
+    ],
+    "mixture_info": {
+      "objective": 0.1917820351790263,
+      "mmd2_mixture": 0.19749363071106324,
+      "mmd2_uniform": 0.3743489659922599,
+      "mmd2_single_source": [
+        0.7042977370508033,
+        0.20706860465273969,
+        0.6307956892595754,
+        0.6534362949819378
+      ],
+      "median_squared_distance": 0.19645890982127545,
+      "gammas": [
+        1.272530730356961,
+        2.545061460713922,
+        5.090122921427844,
+        10.180245842855689,
+        20.360491685711377
+      ],
+      "max_weight": 0.8375897330633602,
+      "entropy": 0.5711595532036933,
+      "effective_sources": 1.7703186403916105,
+      "min_weight": 0.019902632977848233
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.605076346545893,
+        0.31101778520145595,
+        0.7002952057460718,
+        0.8048259839560096
+      ],
+      "standardized": [
+        -0.0012362667606482206,
+        -1.5993052227157551,
+        0.5162331182950395,
+        1.0843083711813646
+      ],
+      "tau": 0.5,
+      "max_weight": 0.9432861284385984,
+      "entropy": 0.26340029394025066,
+      "effective_sources": 1.301347536237684,
+      "min_weight": 0.0044023675352152245
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "kasthuri-pp",
+    "seconds": 8.5
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.27840715557699913,
+      0.18341499813983803,
+      0.0577032083417681,
+      0.01681466600081078
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.18652277313599772,
+      "mixture@raw": 0.18548230728775292,
+      "mixture@logit": 0.19296074616265912,
+      "per_source": 0.18415443581907806,
+      "per_source@raw": 0.1835051571090219,
+      "per_source@logit": 0.18905485087217583,
+      "uniform": 0.0696943056822927,
+      "uniform@raw": 0.08847159266818436,
+      "uniform@logit": 0.06549734203813108
+    },
+    "val_patches": 147
+  },
+  "temperatures": [
+    1.3822230100631714,
+    2.5090832710266113,
+    3.80090069770813,
+    2.597146987915039
+  ],
+  "top_source": "kasthuri-pp",
+  "weight_stats": {
+    "max_weight": 0.8375897330633602,
+    "entropy": 0.5711595532036933,
+    "effective_sources": 1.7703186403916105,
+    "min_weight": 0.019902632977848233
+  },
+  "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": 114.3,
+  "round0_val": {
+    "val_dice": 0.19189352403793783,
+    "val_vi": 0.22130833251687648,
+    "val_ece": 0.02232954733842891
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 5,
+      "selection_seconds": 21.2,
+      "round_seconds": 74.2,
+      "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": 0.12307992202729044,
+      "restart_teacher_dice": 0.3118739095489208,
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.30330376940133036,
+        "uniform": 0.018090264826557255,
+        "top1": 0.3118739095489208,
+        "top2": 0.30483917067915484,
+        "top3": 0.3042653191818001
+      },
+      "val_dice": 0.23685093051502523,
+      "val_vi": 0.28794167838117374,
+      "val_ece": 0.015598701639074236
+    },
+    {
+      "round": 2,
+      "n_labeled": 10,
+      "selection_seconds": 7.4,
+      "round_seconds": 16.0,
+      "teacher_dice_labeled": [
+        0.17236378900771807,
+        0.134718532257256,
+        0.0011034322550670772,
+        0.0025857610756766076
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 17002,
+      "evidence_nonempty_patches": 7,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.14129939786469667,
+        "uniform": 0.025314333612740988,
+        "top1": 0.17236378900771807,
+        "top2": 0.14178256808945655,
+        "top3": 0.14157927984838914
+      },
+      "pseudo_student_dice": 0.042085187539733,
+      "pseudo_ensemble_dice": 0.04012414510853405,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.2866201808181058,
+      "val_vi": 0.27125518727925535,
+      "val_ece": 0.04738608310345371
+    },
+    {
+      "round": 3,
+      "n_labeled": 15,
+      "selection_seconds": 7.3,
+      "round_seconds": 17.0,
+      "teacher_dice_labeled": [
+        0.1816465346770543,
+        0.13722353870458137,
+        0.0007644952319639481,
+        0.0017971963736570949
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 24575,
+      "evidence_nonempty_patches": 12,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.14601875873079226,
+        "uniform": 0.017784931883318283,
+        "top1": 0.1816465346770543,
+        "top2": 0.14662073360149747,
+        "top3": 0.14634583948008453
+      },
+      "pseudo_student_dice": 0.2029553264604811,
+      "pseudo_ensemble_dice": 0.1971301728238465,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.43543202231132283,
+      "val_vi": 0.30158674255549045,
+      "val_ece": 0.009805639594878747
+    },
+    {
+      "round": 4,
+      "n_labeled": 20,
+      "selection_seconds": 7.3,
+      "round_seconds": 19.3,
+      "teacher_dice_labeled": [
+        0.14676982261976956,
+        0.12248808874980222,
+        0.0007346686257830021,
+        0.002188183807439825
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 25508,
+      "evidence_nonempty_patches": 15,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.12848106904231626,
+        "uniform": 0.017156491440690803,
+        "top1": 0.14676982261976956,
+        "top2": 0.12918176006818666,
+        "top3": 0.1288194073915982
+      },
+      "pseudo_student_dice": 0.02631578947368421,
+      "pseudo_ensemble_dice": 0.01118521606224816,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.538829033552655,
+      "val_vi": 0.15799834303392746,
+      "val_ece": 0.007239701747245221
+    },
+    {
+      "round": 5,
+      "n_labeled": 25,
+      "selection_seconds": 7.2,
+      "round_seconds": 25.2,
+      "teacher_dice_labeled": [
+        0.14991183354172677,
+        0.11196244467753076,
+        0.0005527650267970872,
+        0.0013616492678149112
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 41206,
+      "evidence_nonempty_patches": 20,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.11684489229345535,
+        "uniform": 0.011136968085106383,
+        "top1": 0.14991183354172677,
+        "top2": 0.11754230882560181,
+        "top3": 0.11732180427977626
+      },
+      "pseudo_student_dice": 0.5880473899154907,
+      "pseudo_ensemble_dice": 0.15650009421145017,
+      "pseudo_source": "student",
+      "consistency_scale": 0.293,
+      "val_dice": 0.6531419798130339,
+      "val_vi": 0.13160857087554498,
+      "val_ece": 0.006511332969372438
+    },
+    {
+      "round": 6,
+      "n_labeled": 30,
+      "selection_seconds": 7.2,
+      "round_seconds": 71.4,
+      "teacher_dice_labeled": [
+        0.28795011474485044,
+        0.280563902813102,
+        0.0054258056716156265,
+        0.0008719495006685415
+      ],
+      "teacher_dice_pool": [
+        0.3870062594431254,
+        0.4669729462302163,
+        0.009022441201141316,
+        0.0009105291166930196
+      ],
+      "evidence_fg_pixels": 67637,
+      "evidence_nonempty_patches": 24,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.13025564297709144,
+        0.8667965482101361,
+        0.0016914617097016046,
+        0.0012563471030710067
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.2837908050450364,
+        "uniform": 0.034392525491845516,
+        "top1": 0.28795011474485044,
+        "top2": 0.2838905935041172,
+        "top3": 0.2838305958183957
+      },
+      "pseudo_student_dice": 0.41143576121906894,
+      "pseudo_ensemble_dice": 0.06339725733876152,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.608747389325827,
+      "val_vi": 0.17846436977004534,
+      "val_ece": 0.01971239731023785
+    },
+    {
+      "round": 7,
+      "n_labeled": 30,
+      "selection_seconds": 0.0,
+      "round_seconds": 15.6,
+      "final_consolidation": true,
+      "val_dice": 0.6909546100135161,
+      "val_vi": 0.13761017851695848,
+      "val_ece": 0.015077464841051817
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.1281956487029202,
+    "auroc_top20_error": 0.6150618810253443
+  },
+  "al_seconds": 246.0,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.13025564297709144,
+    0.8667965482101361,
+    0.0016914617097016046,
+    0.0012563471030710067
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.8667965482101361,
+    "entropy": 0.4085908200891099,
+    "effective_sources": 1.5046959031685392,
+    "min_weight": 0.0012563471030710067
+  },
+  "top_source_final": "kasthuri-pp",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.79094270337799,
+    "patch_iou": 0.6974377079183249,
+    "patch_precision": 0.7248810371901514,
+    "patch_recall": 0.9449957914805975,
+    "patch_vi": 0.13943115167621045,
+    "n_patches": 147,
+    "cal_ece": 0.014094231018419017,
+    "cal_nll": 0.061826090976241234,
+    "cal_brier": 0.016988086156188117,
+    "cal_mean_confidence": 0.9927206806975483,
+    "cal_accuracy": 0.9786484906462585,
+    "cal_predicted_prevalence": 0.07579985119047619,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.0756547042173509
+  },
+  "test_notta": {
+    "patch_dice": 0.79094270337799,
+    "patch_iou": 0.6974377079183249,
+    "patch_precision": 0.7248810371901514,
+    "patch_recall": 0.9449957914805975,
+    "patch_vi": 0.13943115167621045,
+    "n_patches": 147,
+    "cal_ece": 0.014094231018419017,
+    "cal_nll": 0.061826090976241234,
+    "cal_brier": 0.016988086156188117,
+    "cal_mean_confidence": 0.9927206806975483,
+    "cal_accuracy": 0.9786484906462585,
+    "cal_predicted_prevalence": 0.07579985119047619,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.0756547042173509
+  },
+  "test": {
+    "patch_dice": 0.7877385379966639,
+    "patch_iou": 0.6951161067329877,
+    "patch_precision": 0.7280763728942042,
+    "patch_recall": 0.9348244922204363,
+    "patch_vi": 0.1348154343767868,
+    "n_patches": 147,
+    "image_dice": 0.8425339902518142,
+    "image_iou": 0.728532153833435,
+    "image_precision": 0.7508811622248409,
+    "image_recall": 0.9602402210801045,
+    "image_vi": 0.1642342012294927,
+    "n_images": 3,
+    "cal_ece": 0.00746777378342397,
+    "cal_nll": 0.05250644355455002,
+    "cal_brier": 0.015442382133295839,
+    "cal_mean_confidence": 0.9861893405351245,
+    "cal_accuracy": 0.9787215667517006,
+    "cal_predicted_prevalence": 0.07450441113945579,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.07621637154438468
+  },
+  "eval_seconds": 8.5,
+  "total_seconds": 397.0,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 8.5,
+        "peak_cpu_rss_mb": 2190.7,
+        "peak_gpu_alloc_mb": 592.4,
+        "peak_gpu_reserved_mb": 594.0
+      },
+      "features_and_weights": {
+        "seconds": 17.69,
+        "peak_cpu_rss_mb": 4161.9,
+        "peak_gpu_alloc_mb": 2298.2,
+        "peak_gpu_reserved_mb": 2640.0
+      },
+      "distillation": {
+        "seconds": 114.25,
+        "peak_cpu_rss_mb": 4369.3,
+        "peak_gpu_alloc_mb": 3826.7,
+        "peak_gpu_reserved_mb": 4168.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 246.0,
+        "peak_cpu_rss_mb": 4472.1,
+        "peak_gpu_alloc_mb": 3947.0,
+        "peak_gpu_reserved_mb": 4178.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 8.46,
+        "peak_cpu_rss_mb": 4472.1,
+        "peak_gpu_alloc_mb": 2149.5,
+        "peak_gpu_reserved_mb": 4178.0,
+        "patches": 147,
+        "images": 3,
+        "patches_per_second": 17.4,
+        "seconds_per_image": 2.82
+      }
+    },
+    "peak_gpu_alloc_mb": 3947.0,
+    "peak_cpu_rss_mb": 4472.1,
+    "total_seconds": 394.9
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/drosophila-vnc/macs-v11-norestart-acq-laplace_b5_s1.pth"
+}

macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/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
+1,5,21.2,74.2,"[0.3118739095489208, 0.0968435913812229, 0.0004512053628980276, 0.00288]",,15368,5,True,kept,0.12307992202729044,0.3118739095489208,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.30330376940133036, 'uniform': 0.018090264826557255, 'top1': 0.3118739095489208, 'top2': 0.30483917067915484, 'top3': 0.3042653191818001}",0.23685093051502523,0.28794167838117374,0.015598701639074236,,,,,
+2,10,7.4,16.0,"[0.17236378900771807, 0.134718532257256, 0.0011034322550670772, 0.0025857610756766076]",,17002,7,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.14129939786469667, 'uniform': 0.025314333612740988, 'top1': 0.17236378900771807, 'top2': 0.14178256808945655, 'top3': 0.14157927984838914}",0.2866201808181058,0.27125518727925535,0.04738608310345371,0.042085187539733,0.04012414510853405,ensemble,1.0,
+3,15,7.3,17.0,"[0.1816465346770543, 0.13722353870458137, 0.0007644952319639481, 0.0017971963736570949]",,24575,12,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.14601875873079226, 'uniform': 0.017784931883318283, 'top1': 0.1816465346770543, 'top2': 0.14662073360149747, 'top3': 0.14634583948008453}",0.43543202231132283,0.30158674255549045,0.009805639594878747,0.2029553264604811,0.1971301728238465,ensemble,1.0,
+4,20,7.3,19.3,"[0.14676982261976956, 0.12248808874980222, 0.0007346686257830021, 0.002188183807439825]",,25508,15,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.12848106904231626, 'uniform': 0.017156491440690803, 'top1': 0.14676982261976956, 'top2': 0.12918176006818666, 'top3': 0.1288194073915982}",0.538829033552655,0.15799834303392746,0.007239701747245221,0.02631578947368421,0.01118521606224816,ensemble,1.0,
+5,25,7.2,25.2,"[0.14991183354172677, 0.11196244467753076, 0.0005527650267970872, 0.0013616492678149112]",,41206,20,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.11684489229345535, 'uniform': 0.011136968085106383, 'top1': 0.14991183354172677, 'top2': 0.11754230882560181, 'top3': 0.11732180427977626}",0.6531419798130339,0.13160857087554498,0.006511332969372438,0.5880473899154907,0.15650009421145017,student,0.293,
+6,30,7.2,71.4,"[0.28795011474485044, 0.280563902813102, 0.0054258056716156265, 0.0008719495006685415]","[0.3870062594431254, 0.4669729462302163, 0.009022441201141316, 0.0009105291166930196]",67637,24,False,,,,"[0.13025564297709144, 0.8667965482101361, 0.0016914617097016046, 0.0012563471030710067]",mixture,"{'mixture': 0.2837908050450364, 'uniform': 0.034392525491845516, 'top1': 0.28795011474485044, 'top2': 0.2838905935041172, 'top3': 0.2838305958183957}",0.608747389325827,0.17846436977004534,0.01971239731023785,0.41143576121906894,0.06339725733876152,student,0.0,
+7,30,0.0,15.6,,,,,,,,,,,,0.6909546100135161,0.13761017851695848,0.015077464841051817,,,,,True

macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt

@@ -0,0 +1,105 @@
+[12:09:17][      0s][START] target=drosophila-vnc variant=macs-v11-norestart-acq-laplace budget=5.0% seed=2 | {"task": "mitochondria", "target": "drosophila-vnc", "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": 8, "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}
+[12:09:17][      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.
+[12:09:17][      0s][DATA] drosophila-vnc: train=686 val=147 test=147 patches
+[12:09:18][      2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, multiclass=2.60
+[12:09:19][      2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s
+[12:09:20][      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
+[12:09:20][      3s][WEIGHTS] per-source: fly-mito=0.039, kasthuri=0.943, lucchi=0.014, multiclass=0.004 | min-MMD source (NeuroADDA rule) = kasthuri
+[12:09:26][     10s][TEACHERS] target-val Dice: fly-mito=0.278, kasthuri=0.183, lucchi=0.058, multiclass=0.017
+[12:09:26][     10s][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
+[12:09:26][     10s][WEIGHTS] using mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020
+[12:09:26][     10s][INIT] student initialised from teacher kasthuri (weight 0.838)
+[12:09:30][     13s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[12:09:30][     13s][KD] distilling the mixture ensemble (prob space) into the student
+[12:09:44][     27s][KD] soft labels for 686 patches from 4 teachers in 14s
+[12:10:05][     48s][KD] epoch 1/5 | patches=686 | loss=0.1999 | 21s
+[12:10:25][     69s][KD] epoch 2/5 | patches=686 | loss=0.1335 | 21s
+[12:10:46][     89s][KD] epoch 3/5 | patches=686 | loss=0.1266 | 21s
+[12:11:07][    110s][KD] epoch 4/5 | patches=686 | loss=0.1231 | 21s
+[12:11:28][    131s][KD] epoch 5/5 | patches=686 | loss=0.1203 | 21s
+[12:11:29][    132s][ROUND0] before any target label: val_dice=0.2059 val_vi=0.2207
+[12:11:29][    132s][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
+[12:11:36][    140s][AL] round-1 uncertainty quality on 686 patches: {'spearman_uncertainty_error': 0.126648544840516, 'auroc_top20_error': 0.6094229345181423}
+[12:11:51][    154s][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
+[12:11:52][    155s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.531, uniform=0.067, top1=0.531, top2=0.531, top3=0.531 -> using mixture
+[12:11:52][    156s][WEIGHTS] round 1: student kept (labeled Dice 0.583 vs best teacher kasthuri 0.531); distilling the corrected ensemble into it
+[12:11:52][    156s][WEIGHTS] round 1: teacher Dice on 5 labeled patches + 0 blank (19579 fg px): fly-mito=0.363, kasthuri=0.531, lucchi=0.014, multiclass=0.002 | weights: fly-mito=0.025, kasthuri=0.974, lucchi=0.000, multiclass=0.000
+[12:12:06][    170s][KD] soft labels for 681 patches from 4 teachers in 14s
+[12:12:27][    190s][KD] epoch 1/2 | patches=681 | loss=0.0886 | 21s
+[12:12:48][    211s][KD] epoch 2/2 | patches=681 | loss=0.0859 | 21s
+[12:13:02][    225s][CONS] pseudo-labels for 681 unlabeled patches from the current ensemble in 14s
+[12:13:03][    226s][FT] round 1/6 epoch 1/5 | patches=5 | loss=0.9031
+[12:13:04][    227s][FT] round 1/6 epoch 2/5 | patches=5 | loss=0.8632
+[12:13:05][    228s][FT] round 1/6 epoch 3/5 | patches=5 | loss=0.8447
+[12:13:06][    229s][FT] round 1/6 epoch 4/5 | patches=5 | loss=0.8470
+[12:13:07][    230s][FT] round 1/6 epoch 5/5 | patches=5 | loss=0.8177
+[12:13:08][    232s][AL] round 1/6 | labeled=5 | selection_seconds=22.1000 round_seconds=98.5000 restart_student_dice=0.5832 restart_teacher_dice=0.5310 val_dice=0.1982 val_vi=0.2106 val_ece=0.0303
+[12:13:16][    239s][AL] round 2: 242/681 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:13:18][    241s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.363, uniform=0.033, top1=0.364, top2=0.363, top3=0.363 -> using mixture
+[12:13:18][    241s][WEIGHTS] round 2: teacher Dice on 10 labeled patches + 0 blank (46033 fg px): fly-mito=0.334, kasthuri=0.364, lucchi=0.006, multiclass=0.001 | weights: fly-mito=0.094, kasthuri=0.905, lucchi=0.001, multiclass=0.001
+[12:13:19][    242s][CONS] round 2: on the 5 new patches student Dice=0.153 vs ensemble 0.218 -> pseudo-labels from the ensemble | consistency x1.00
+[12:13:20][    243s][FT] round 2/6 epoch 1/5 | patches=10 | loss=0.9990
+[12:13:21][    245s][FT] round 2/6 epoch 2/5 | patches=10 | loss=0.9318
+[12:13:23][    246s][FT] round 2/6 epoch 3/5 | patches=10 | loss=0.8588
+[12:13:24][    248s][FT] round 2/6 epoch 4/5 | patches=10 | loss=0.7714
+[12:13:26][    249s][FT] round 2/6 epoch 5/5 | patches=10 | loss=0.6310
+[12:13:27][    250s][AL] round 2/6 | labeled=10 | selection_seconds=7.6000 round_seconds=17.6000 pseudo_student_dice=0.1526 pseudo_ensemble_dice=0.2182 consistency_scale=1.0000 val_dice=0.4146 val_vi=0.1946 val_ece=0.0208
+[12:13:35][    258s][AL] round 3: 141/676 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:13:37][    260s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.242, uniform=0.021, top1=0.334, top2=0.243, top3=0.243 -> using top1
+[12:13:37][    260s][WEIGHTS] round 3: teacher Dice on 15 labeled patches + 0 blank (71466 fg px): fly-mito=0.334, kasthuri=0.243, lucchi=0.005, multiclass=0.001 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[12:13:38][    261s][CONS] round 3: on the 5 new patches student Dice=0.270 vs ensemble 0.334 -> pseudo-labels from the ensemble | consistency x1.00
+[12:13:39][    262s][FT] round 3/6 epoch 1/5 | patches=15 | loss=0.9433
+[12:13:41][    264s][FT] round 3/6 epoch 2/5 | patches=15 | loss=0.8414
+[12:13:42][    266s][FT] round 3/6 epoch 3/5 | patches=15 | loss=0.6321
+[12:13:44][    267s][FT] round 3/6 epoch 4/5 | patches=15 | loss=0.5810
+[12:13:45][    269s][FT] round 3/6 epoch 5/5 | patches=15 | loss=0.5354
+[12:13:47][    270s][AL] round 3/6 | labeled=15 | selection_seconds=7.6000 round_seconds=18.4000 pseudo_student_dice=0.2699 pseudo_ensemble_dice=0.3335 consistency_scale=1.0000 val_dice=0.5844 val_vi=0.1675 val_ece=0.0049
+[12:13:54][    277s][AL] round 4: 64/671 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:13:56][    280s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.194, uniform=0.017, top1=0.322, top2=0.194, top3=0.194 -> using top1
+[12:13:56][    280s][WEIGHTS] round 4: teacher Dice on 20 labeled patches + 0 blank (88303 fg px): fly-mito=0.322, kasthuri=0.195, lucchi=0.006, multiclass=0.000 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[12:13:57][    281s][CONS] round 4: on the 5 new patches student Dice=0.113 vs ensemble 0.276 -> pseudo-labels from the ensemble | consistency x1.00
+[12:13:59][    283s][FT] round 4/6 epoch 1/5 | patches=20 | loss=0.6223
+[12:14:01][    285s][FT] round 4/6 epoch 2/5 | patches=20 | loss=0.6012
+[12:14:03][    286s][FT] round 4/6 epoch 3/5 | patches=20 | loss=0.5775
+[12:14:05][    288s][FT] round 4/6 epoch 4/5 | patches=20 | loss=0.5477
+[12:14:07][    290s][FT] round 4/6 epoch 5/5 | patches=20 | loss=0.5058
+[12:14:08][    292s][AL] round 4/6 | labeled=20 | selection_seconds=7.5000 round_seconds=20.7000 pseudo_student_dice=0.1127 pseudo_ensemble_dice=0.2760 consistency_scale=1.0000 val_dice=0.6994 val_vi=0.1209 val_ece=0.0073
+[12:14:16][    299s][AL] round 5: 133/666 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:14:19][    302s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.182, uniform=0.023, top1=0.284, top2=0.182, top3=0.182 -> using top1
+[12:14:19][    302s][WEIGHTS] round 5: teacher Dice on 25 labeled patches + 0 blank (100786 fg px): fly-mito=0.284, kasthuri=0.184, lucchi=0.018, multiclass=0.000 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[12:14:23][    306s][CONS] round 5: on the 5 new patches student Dice=0.494 vs ensemble 0.137 -> pseudo-labels from the student | consistency x0.00
+[12:14:24][    308s][FT] round 5/6 epoch 1/5 | patches=25 | loss=0.4278
+[12:14:25][    309s][FT] round 5/6 epoch 2/5 | patches=25 | loss=0.3891
+[12:14:27][    310s][FT] round 5/6 epoch 3/5 | patches=25 | loss=0.3514
+[12:14:28][    311s][FT] round 5/6 epoch 4/5 | patches=25 | loss=0.3588
+[12:14:29][    312s][FT] round 5/6 epoch 5/5 | patches=25 | loss=0.3080
+[12:14:30][    313s][AL] round 5/6 | labeled=25 | selection_seconds=7.4000 round_seconds=20.5000 pseudo_student_dice=0.4939 pseudo_ensemble_dice=0.1373 consistency_scale=0.0000 val_dice=0.6982 val_vi=0.1222 val_ece=0.0099
+[12:14:37][    321s][AL] round 6: 123/661 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:15:07][    350s][WEIGHTS] round 6: teacher agreement with the student on 656 unlabeled patches: fly-mito=0.359, kasthuri=0.416, lucchi=0.011, multiclass=0.001 | blended with labeled Dice at 0.50
+[12:15:32][    376s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.306, uniform=0.035, top1=0.317, top2=0.306, top3=0.306 -> using top1
+[12:15:32][    376s][WEIGHTS] round 6: teacher Dice on 30 labeled patches + 0 blank (109183 fg px): fly-mito=0.317, kasthuri=0.306, lucchi=0.014, multiclass=0.001 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[12:15:37][    380s][CONS] round 6: on the 5 new patches student Dice=0.583 vs ensemble 0.190 -> pseudo-labels from the student | consistency x0.28
+[12:15:40][    383s][FT] round 6/6 epoch 1/5 | patches=30 | loss=0.3750
+[12:15:42][    385s][FT] round 6/6 epoch 2/5 | patches=30 | loss=0.3679
+[12:15:45][    388s][FT] round 6/6 epoch 3/5 | patches=30 | loss=0.3280
+[12:15:47][    390s][FT] round 6/6 epoch 4/5 | patches=30 | loss=0.3097
+[12:15:50][    393s][FT] round 6/6 epoch 5/5 | patches=30 | loss=0.3035
+[12:15:51][    394s][AL] round 6/6 | labeled=30 | selection_seconds=7.4000 round_seconds=79.7000 pseudo_student_dice=0.5830 pseudo_ensemble_dice=0.1899 consistency_scale=0.2770 val_dice=0.7796 val_vi=0.1032 val_ece=0.0075
+[12:15:54][    398s][CONS] final consolidation: pseudo-labels refreshed from the student
+[12:15:57][    400s][FT] final 1/10 epoch 1/1 | patches=30 | loss=0.3236
+[12:15:59][    403s][FT] final 2/10 epoch 1/1 | patches=30 | loss=0.2798
+[12:16:02][    405s][FT] final 3/10 epoch 1/1 | patches=30 | loss=0.2968
+[12:16:04][    408s][FT] final 4/10 epoch 1/1 | patches=30 | loss=0.2693
+[12:16:07][    410s][FT] final 5/10 epoch 1/1 | patches=30 | loss=0.2312
+[12:16:10][    413s][FT] final 6/10 epoch 1/1 | patches=30 | loss=0.2003
+[12:16:12][    415s][FT] final 7/10 epoch 1/1 | patches=30 | loss=0.2203
+[12:16:15][    418s][FT] final 8/10 epoch 1/1 | patches=30 | loss=0.1931
+[12:16:17][    420s][FT] final 9/10 epoch 1/1 | patches=30 | loss=0.2420
+[12:16:20][    423s][FT] final 10/10 epoch 1/1 | patches=30 | loss=0.1780
+[12:16:21][    424s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=28.9000 val_dice=0.7771 val_vi=0.0989 val_ece=0.0088
+[12:16:21][    424s][WEIGHTS] final: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[12:16:28][    431s][TEST] raw student (no EMA, no TTA): patch_dice=0.8138 patch_vi=0.1142
+[12:16:29][    433s][TEST] final model without TTA: patch_dice=0.8138 patch_vi=0.1142
+[12:16:29][    433s][RESOURCES] {"phases": {"setup": {"seconds": 0.75, "peak_cpu_rss_mb": 2191.5, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 11.38, "peak_cpu_rss_mb": 4114.7, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 117.78, "peak_cpu_rss_mb": 4292.5, "peak_gpu_alloc_mb": 3826.7, "peak_gpu_reserved_mb": 4168.0, "kd_epochs": 5}, "active_learning": {"seconds": 292.01, "peak_cpu_rss_mb": 4424.3, "peak_gpu_alloc_mb": 3947.0, "peak_gpu_reserved_mb": 4178.0, "rounds": 6}, "inference": {"seconds": 8.63, "peak_cpu_rss_mb": 4424.3, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4178.0, "patches": 147, "images": 3, "patches_per_second": 17.0, "seconds_per_image": 2.877}}, "peak_gpu_alloc_mb": 3947.0, "peak_cpu_rss_mb": 4424.3, "total_seconds": 430.6}
+[12:16:29][    433s][TEST] patch_dice=0.8323 patch_iou=0.7579 patch_precision=0.8281 patch_recall=0.9090 patch_vi=0.1053 image_dice=0.8863 image_iou=0.7972 image_precision=0.8331 image_recall=0.9485 image_vi=0.1257 cal_ece=0.0064 cal_nll=0.0393 cal_brier=0.0111 cal_mean_confidence=0.9915 cal_accuracy=0.9852 cal_predicted_prevalence=0.0663 cal_true_prevalence=0.0594 cal_mean_probability=0.0664
+[12:16:30][    433s][DONE] drosophila-vnc macs-v11-norestart-acq-laplace b5 s2 | test patch Dice=0.8323 VI=0.1053 | image Dice=0.8863 VI=0.1257 | 433s

macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json

@@ -0,0 +1,652 @@
+{
+  "target": "drosophila-vnc",
+  "variant": "macs-v11-norestart-acq-laplace",
+  "budget": 5.0,
+  "seed": 2,
+  "config": {
+    "task": "mitochondria",
+    "target": "drosophila-vnc",
+    "variant": "macs-v11-norestart-acq-laplace",
+    "budget": 5.0,
+    "seed": 2,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "drosophila-vnc",
+      "--variants",
+      "macs-v11-norestart-acq-laplace",
+      "--budgets",
+      "5",
+      "--seeds",
+      "2",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--task",
+      "mitochondria",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "commit": "aecf380",
+    "dirty": true
+  },
+  "sources": [
+    "fly-brain-mito-benchmark",
+    "kasthuri-pp",
+    "lucchi-pp",
+    "multiclass"
+  ],
+  "alphas": [
+    0.11690492516893997,
+    0.8375897330633602,
+    0.0256027087898516,
+    0.019902632977848233
+  ],
+  "weights_all": {
+    "sources": [
+      "fly-brain-mito-benchmark",
+      "kasthuri-pp",
+      "lucchi-pp",
+      "multiclass"
+    ],
+    "mixture": [
+      0.11690492516893997,
+      0.8375897330633602,
+      0.0256027087898516,
+      0.019902632977848233
+    ],
+    "per_source": [
+      0.038599207620062845,
+      0.9432861284385984,
+      0.013712296406123614,
+      0.0044023675352152245
+    ],
+    "uniform": [
+      0.25,
+      0.25,
+      0.25,
+      0.25
+    ],
+    "mixture_info": {
+      "objective": 0.1917820351790263,
+      "mmd2_mixture": 0.19749363071106324,
+      "mmd2_uniform": 0.3743489659922599,
+      "mmd2_single_source": [
+        0.7042977370508033,
+        0.20706860465273969,
+        0.6307956892595754,
+        0.6534362949819378
+      ],
+      "median_squared_distance": 0.19645890982127545,
+      "gammas": [
+        1.272530730356961,
+        2.545061460713922,
+        5.090122921427844,
+        10.180245842855689,
+        20.360491685711377
+      ],
+      "max_weight": 0.8375897330633602,
+      "entropy": 0.5711595532036933,
+      "effective_sources": 1.7703186403916105,
+      "min_weight": 0.019902632977848233
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.605076346545893,
+        0.31101778520145595,
+        0.7002952057460718,
+        0.8048259839560096
+      ],
+      "standardized": [
+        -0.0012362667606482206,
+        -1.5993052227157551,
+        0.5162331182950395,
+        1.0843083711813646
+      ],
+      "tau": 0.5,
+      "max_weight": 0.9432861284385984,
+      "entropy": 0.26340029394025066,
+      "effective_sources": 1.301347536237684,
+      "min_weight": 0.0044023675352152245
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "kasthuri-pp",
+    "seconds": 1.4
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.27840715557699913,
+      0.18341499813983803,
+      0.0577032083417681,
+      0.01681466600081078
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.18652277313599772,
+      "mixture@raw": 0.18548230728775292,
+      "mixture@logit": 0.19296074616265912,
+      "per_source": 0.18415443581907806,
+      "per_source@raw": 0.1835051571090219,
+      "per_source@logit": 0.18905485087217583,
+      "uniform": 0.0696943056822927,
+      "uniform@raw": 0.08847159266818436,
+      "uniform@logit": 0.06549734203813108
+    },
+    "val_patches": 147
+  },
+  "temperatures": [
+    1.3822230100631714,
+    2.5090832710266113,
+    3.80090069770813,
+    2.597146987915039
+  ],
+  "top_source": "kasthuri-pp",
+  "weight_stats": {
+    "max_weight": 0.8375897330633602,
+    "entropy": 0.5711595532036933,
+    "effective_sources": 1.7703186403916105,
+    "min_weight": 0.019902632977848233
+  },
+  "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": 117.8,
+  "round0_val": {
+    "val_dice": 0.2058871697531777,
+    "val_vi": 0.22072946276036065,
+    "val_ece": 0.019761043084348236
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 5,
+      "selection_seconds": 22.1,
+      "round_seconds": 98.5,
+      "teacher_dice_labeled": [
+        0.36295908671523197,
+        0.5310285360135809,
+        0.013736820762368207,
+        0.0019869574077847974
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 19579,
+      "evidence_nonempty_patches": 5,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.5831796951819076,
+      "restart_teacher_dice": 0.5310285360135809,
+      "weights": [
+        0.025328879887354556,
+        0.9743856190601974,
+        0.00016881695486452877,
+        0.0001166840975834896
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.531149750640248,
+        "uniform": 0.0667619235317304,
+        "top1": 0.5310285360135809,
+        "top2": 0.5311354323916325,
+        "top3": 0.5311354323916325
+      },
+      "val_dice": 0.19819172565480536,
+      "val_vi": 0.21059685476621876,
+      "val_ece": 0.030250068283869655
+    },
+    {
+      "round": 2,
+      "n_labeled": 10,
+      "selection_seconds": 7.6,
+      "round_seconds": 17.6,
+      "teacher_dice_labeled": [
+        0.3342600608199596,
+        0.36433835993119296,
+        0.0058655469460196525,
+        0.0008455101244417466
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 46033,
+      "evidence_nonempty_patches": 10,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.09351739812091488,
+        0.9051474776121796,
+        0.0007676199810288104,
+        0.0005675042858766415
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.36326152072288853,
+        "uniform": 0.032810331945778634,
+        "top1": 0.36433835993119296,
+        "top2": 0.3632638740962572,
+        "top3": 0.3632524156254308
+      },
+      "pseudo_student_dice": 0.15260812899705997,
+      "pseudo_ensemble_dice": 0.21822732695501648,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.41463101296538946,
+      "val_vi": 0.19460083461243838,
+      "val_ece": 0.020840921132885007
+    },
+    {
+      "round": 3,
+      "n_labeled": 15,
+      "selection_seconds": 7.6,
+      "round_seconds": 18.4,
+      "teacher_dice_labeled": [
+        0.3339991879232587,
+        0.24314318156375442,
+        0.005117478909572614,
+        0.0005446547028838769
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 71466,
+      "evidence_nonempty_patches": 14,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.2424702747842685,
+        "uniform": 0.021214923601429972,
+        "top1": 0.3339991879232587,
+        "top2": 0.24252174949350494,
+        "top3": 0.2425010931351115
+      },
+      "pseudo_student_dice": 0.2699233539569674,
+      "pseudo_ensemble_dice": 0.3335354950696935,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.5843989124516004,
+      "val_vi": 0.16750684661110102,
+      "val_ece": 0.004899511730954534
+    },
+    {
+      "round": 4,
+      "n_labeled": 20,
+      "selection_seconds": 7.5,
+      "round_seconds": 20.7,
+      "teacher_dice_labeled": [
+        0.32164944334569073,
+        0.19456246320800594,
+        0.006338947937308593,
+        0.000440294884676609
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 88303,
+      "evidence_nonempty_patches": 18,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.19353505028082205,
+        "uniform": 0.017062989853238304,
+        "top1": 0.32164944334569073,
+        "top2": 0.19381819105561976,
+        "top3": 0.19362159275834012
+      },
+      "pseudo_student_dice": 0.11274676329596497,
+      "pseudo_ensemble_dice": 0.2759716745795445,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.6993540102840254,
+      "val_vi": 0.12093193704679728,
+      "val_ece": 0.007255122349904359
+    },
+    {
+      "round": 5,
+      "n_labeled": 25,
+      "selection_seconds": 7.4,
+      "round_seconds": 20.5,
+      "teacher_dice_labeled": [
+        0.28431461799312685,
+        0.1835229273015314,
+        0.0176183379740382,
+        0.00038568800806977984
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 100786,
+      "evidence_nonempty_patches": 23,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.18152320914405376,
+        "uniform": 0.023056570492629457,
+        "top1": 0.28431461799312685,
+        "top2": 0.1816722357064336,
+        "top3": 0.18160171744910858
+      },
+      "pseudo_student_dice": 0.49388154674498286,
+      "pseudo_ensemble_dice": 0.13730756264748847,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.6982076177663978,
+      "val_vi": 0.12219064496731252,
+      "val_ece": 0.009867805570635917
+    },
+    {
+      "round": 6,
+      "n_labeled": 30,
+      "selection_seconds": 7.4,
+      "round_seconds": 79.7,
+      "teacher_dice_labeled": [
+        0.31740826893484075,
+        0.30628203034957413,
+        0.013855732783164331,
+        0.0007705463176043627
+      ],
+      "teacher_dice_pool": [
+        0.358785858335439,
+        0.4158121410750507,
+        0.011455642146002096,
+        0.0011850973087883954
+      ],
+      "evidence_fg_pixels": 109183,
+      "evidence_nonempty_patches": 27,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.3063970994102441,
+        "uniform": 0.03496920739101875,
+        "top1": 0.31740826893484075,
+        "top2": 0.30644533343595115,
+        "top3": 0.3064190484860375
+      },
+      "pseudo_student_dice": 0.5830105146651909,
+      "pseudo_ensemble_dice": 0.18987521606326807,
+      "pseudo_source": "student",
+      "consistency_scale": 0.277,
+      "val_dice": 0.7795886737738887,
+      "val_vi": 0.10321806923354346,
+      "val_ece": 0.007524647510925679
+    },
+    {
+      "round": 7,
+      "n_labeled": 30,
+      "selection_seconds": 0.0,
+      "round_seconds": 28.9,
+      "final_consolidation": true,
+      "val_dice": 0.7771382821975249,
+      "val_vi": 0.09893786775402841,
+      "val_ece": 0.008758624116729442
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.126648544840516,
+    "auroc_top20_error": 0.6094229345181423
+  },
+  "al_seconds": 292.0,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    1.0,
+    0.0,
+    0.0,
+    0.0
+  ],
+  "weight_stats_final": {
+    "max_weight": 1.0,
+    "entropy": -0.0,
+    "effective_sources": 1.0,
+    "min_weight": 0.0
+  },
+  "top_source_final": "fly-brain-mito-benchmark",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.8138396684227535,
+    "patch_iou": 0.7341542006754566,
+    "patch_precision": 0.8001245475759902,
+    "patch_recall": 0.9119421120838823,
+    "patch_vi": 0.11419680921494532,
+    "n_patches": 147,
+    "cal_ece": 0.011223098567597907,
+    "cal_nll": 0.05504557527311981,
+    "cal_brier": 0.013421014793791255,
+    "cal_mean_confidence": 0.9949232920450249,
+    "cal_accuracy": 0.9837239583333334,
+    "cal_predicted_prevalence": 0.06801392431972789,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.06709936536024964
+  },
+  "test_notta": {
+    "patch_dice": 0.8138396684227535,
+    "patch_iou": 0.7341542006754566,
+    "patch_precision": 0.8001245475759902,
+    "patch_recall": 0.9119421120838823,
+    "patch_vi": 0.11419680921494532,
+    "n_patches": 147,
+    "cal_ece": 0.011223098567597907,
+    "cal_nll": 0.05504557527311981,
+    "cal_brier": 0.013421014793791255,
+    "cal_mean_confidence": 0.9949232920450249,
+    "cal_accuracy": 0.9837239583333334,
+    "cal_predicted_prevalence": 0.06801392431972789,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.06709936536024964
+  },
+  "test": {
+    "patch_dice": 0.8322875309966076,
+    "patch_iou": 0.7578700879195538,
+    "patch_precision": 0.8280970395677735,
+    "patch_recall": 0.9090226856163043,
+    "patch_vi": 0.10528060887477136,
+    "n_patches": 147,
+    "image_dice": 0.8862696494450427,
+    "image_iou": 0.7971637240024779,
+    "image_precision": 0.8330888536136852,
+    "image_recall": 0.9484565471613607,
+    "image_vi": 0.125726327245971,
+    "n_images": 3,
+    "cal_ece": 0.0064357235725075925,
+    "cal_nll": 0.03925682527344598,
+    "cal_brier": 0.011056755049628232,
+    "cal_mean_confidence": 0.9915352959508038,
+    "cal_accuracy": 0.9852452699829932,
+    "cal_predicted_prevalence": 0.06629331420068027,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.06635219970296335
+  },
+  "eval_seconds": 8.6,
+  "total_seconds": 432.6,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 0.75,
+        "peak_cpu_rss_mb": 2191.5,
+        "peak_gpu_alloc_mb": 592.4,
+        "peak_gpu_reserved_mb": 594.0
+      },
+      "features_and_weights": {
+        "seconds": 11.38,
+        "peak_cpu_rss_mb": 4114.7,
+        "peak_gpu_alloc_mb": 2298.2,
+        "peak_gpu_reserved_mb": 2640.0
+      },
+      "distillation": {
+        "seconds": 117.78,
+        "peak_cpu_rss_mb": 4292.5,
+        "peak_gpu_alloc_mb": 3826.7,
+        "peak_gpu_reserved_mb": 4168.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 292.01,
+        "peak_cpu_rss_mb": 4424.3,
+        "peak_gpu_alloc_mb": 3947.0,
+        "peak_gpu_reserved_mb": 4178.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 8.63,
+        "peak_cpu_rss_mb": 4424.3,
+        "peak_gpu_alloc_mb": 2149.5,
+        "peak_gpu_reserved_mb": 4178.0,
+        "patches": 147,
+        "images": 3,
+        "patches_per_second": 17.0,
+        "seconds_per_image": 2.877
+      }
+    },
+    "peak_gpu_alloc_mb": 3947.0,
+    "peak_cpu_rss_mb": 4424.3,
+    "total_seconds": 430.6
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/drosophila-vnc/macs-v11-norestart-acq-laplace_b5_s2.pth"
+}

macs_v2_r26/runs/mitochondria/drosophila-vnc/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
+1,5,22.1,98.5,"[0.36295908671523197, 0.5310285360135809, 0.013736820762368207, 0.0019869574077847974]",,19579,5,True,kept,0.5831796951819076,0.5310285360135809,"[0.025328879887354556, 0.9743856190601974, 0.00016881695486452877, 0.0001166840975834896]",mixture,"{'mixture': 0.531149750640248, 'uniform': 0.0667619235317304, 'top1': 0.5310285360135809, 'top2': 0.5311354323916325, 'top3': 0.5311354323916325}",0.19819172565480536,0.21059685476621876,0.030250068283869655,,,,,
+2,10,7.6,17.6,"[0.3342600608199596, 0.36433835993119296, 0.0058655469460196525, 0.0008455101244417466]",,46033,10,False,,,,"[0.09351739812091488, 0.9051474776121796, 0.0007676199810288104, 0.0005675042858766415]",mixture,"{'mixture': 0.36326152072288853, 'uniform': 0.032810331945778634, 'top1': 0.36433835993119296, 'top2': 0.3632638740962572, 'top3': 0.3632524156254308}",0.41463101296538946,0.19460083461243838,0.020840921132885007,0.15260812899705997,0.21822732695501648,ensemble,1.0,
+3,15,7.6,18.4,"[0.3339991879232587, 0.24314318156375442, 0.005117478909572614, 0.0005446547028838769]",,71466,14,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.2424702747842685, 'uniform': 0.021214923601429972, 'top1': 0.3339991879232587, 'top2': 0.24252174949350494, 'top3': 0.2425010931351115}",0.5843989124516004,0.16750684661110102,0.004899511730954534,0.2699233539569674,0.3335354950696935,ensemble,1.0,
+4,20,7.5,20.7,"[0.32164944334569073, 0.19456246320800594, 0.006338947937308593, 0.000440294884676609]",,88303,18,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.19353505028082205, 'uniform': 0.017062989853238304, 'top1': 0.32164944334569073, 'top2': 0.19381819105561976, 'top3': 0.19362159275834012}",0.6993540102840254,0.12093193704679728,0.007255122349904359,0.11274676329596497,0.2759716745795445,ensemble,1.0,
+5,25,7.4,20.5,"[0.28431461799312685, 0.1835229273015314, 0.0176183379740382, 0.00038568800806977984]",,100786,23,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.18152320914405376, 'uniform': 0.023056570492629457, 'top1': 0.28431461799312685, 'top2': 0.1816722357064336, 'top3': 0.18160171744910858}",0.6982076177663978,0.12219064496731252,0.009867805570635917,0.49388154674498286,0.13730756264748847,student,0.0,
+6,30,7.4,79.7,"[0.31740826893484075, 0.30628203034957413, 0.013855732783164331, 0.0007705463176043627]","[0.358785858335439, 0.4158121410750507, 0.011455642146002096, 0.0011850973087883954]",109183,27,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.3063970994102441, 'uniform': 0.03496920739101875, 'top1': 0.31740826893484075, 'top2': 0.30644533343595115, 'top3': 0.3064190484860375}",0.7795886737738887,0.10321806923354346,0.007524647510925679,0.5830105146651909,0.18987521606326807,student,0.277,
+7,30,0.0,28.9,,,,,,,,,,,,0.7771382821975249,0.09893786775402841,0.008758624116729442,,,,,True

macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-2/log.txt

@@ -0,0 +1,107 @@
+[13:30:41][      0s][START] target=drosophila-vnc variant=macs-v11-norestart-acq-margin budget=5.0% seed=2 | {"task": "mitochondria", "target": "drosophila-vnc", "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": 8, "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}
+[13:30:41][      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.
+[13:30:41][      0s][DATA] drosophila-vnc: train=686 val=147 test=147 patches
+[13:30:42][      2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, multiclass=2.60
+[13:30:42][      2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s
+[13:30:44][      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
+[13:30:44][      3s][WEIGHTS] per-source: fly-mito=0.039, kasthuri=0.943, lucchi=0.014, multiclass=0.004 | min-MMD source (NeuroADDA rule) = kasthuri
+[13:30:50][     10s][TEACHERS] target-val Dice: fly-mito=0.278, kasthuri=0.183, lucchi=0.058, multiclass=0.017
+[13:30:50][     10s][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
+[13:30:50][     10s][WEIGHTS] using mixture: fly-mito=0.117, kasthuri=0.838, lucchi=0.026, multiclass=0.020
+[13:30:50][     10s][INIT] student initialised from teacher kasthuri (weight 0.838)
+[13:30:53][     13s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[13:30:53][     13s][KD] distilling the mixture ensemble (prob space) into the student
+[13:31:08][     27s][KD] soft labels for 686 patches from 4 teachers in 14s
+[13:31:28][     48s][KD] epoch 1/5 | patches=686 | loss=0.1999 | 21s
+[13:31:49][     68s][KD] epoch 2/5 | patches=686 | loss=0.1335 | 21s
+[13:32:10][     89s][KD] epoch 3/5 | patches=686 | loss=0.1266 | 21s
+[13:32:30][    110s][KD] epoch 4/5 | patches=686 | loss=0.1231 | 21s
+[13:32:51][    131s][KD] epoch 5/5 | patches=686 | loss=0.1203 | 21s
+[13:32:52][    132s][ROUND0] before any target label: val_dice=0.2058 val_vi=0.2207
+[13:32:52][    132s][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
+[13:32:56][    136s][AL] round-1 uncertainty quality on 686 patches: {'spearman_uncertainty_error': 0.07993511898383816, 'auroc_top20_error': 0.5728207976303819}
+[13:33:11][    150s][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
+[13:33:12][    151s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.531, uniform=0.067, top1=0.531, top2=0.531, top3=0.531 -> using mixture
+[13:33:12][    151s][WEIGHTS] round 1: student kept (labeled Dice 0.583 vs best teacher kasthuri 0.531); distilling the corrected ensemble into it
+[13:33:12][    151s][WEIGHTS] round 1: teacher Dice on 5 labeled patches + 0 blank (19579 fg px): fly-mito=0.363, kasthuri=0.531, lucchi=0.014, multiclass=0.002 | weights: fly-mito=0.025, kasthuri=0.974, lucchi=0.000, multiclass=0.000
+[13:33:26][    166s][KD] soft labels for 681 patches from 4 teachers in 14s
+[13:33:47][    186s][KD] epoch 1/2 | patches=681 | loss=0.0885 | 21s
+[13:34:08][    207s][KD] epoch 2/2 | patches=681 | loss=0.0859 | 21s
+[13:34:22][    221s][CONS] pseudo-labels for 681 unlabeled patches from the current ensemble in 14s
+[13:34:23][    222s][FT] round 1/6 epoch 1/5 | patches=5 | loss=0.9208
+[13:34:24][    223s][FT] round 1/6 epoch 2/5 | patches=5 | loss=0.8768
+[13:34:25][    224s][FT] round 1/6 epoch 3/5 | patches=5 | loss=0.8140
+[13:34:26][    225s][FT] round 1/6 epoch 4/5 | patches=5 | loss=0.8660
+[13:34:27][    227s][FT] round 1/6 epoch 5/5 | patches=5 | loss=0.7604
+[13:34:28][    228s][AL] round 1/6 | labeled=5 | selection_seconds=18.3000 round_seconds=94.9000 restart_student_dice=0.5832 restart_teacher_dice=0.5310 val_dice=0.2276 val_vi=0.2140 val_ece=0.0322
+[13:34:32][    231s][AL] round 2: 210/681 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[13:34:34][    233s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.293, uniform=0.025, top1=0.384, top2=0.293, top3=0.293 -> using top1
+[13:34:34][    233s][WEIGHTS] round 2: teacher Dice on 10 labeled patches + 0 blank (56186 fg px): fly-mito=0.384, kasthuri=0.287, lucchi=0.007, multiclass=0.002 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[13:34:35][    234s][CONS] round 2: on the 5 new patches student Dice=0.097 vs ensemble 0.398 -> pseudo-labels from the ensemble | consistency x1.00
+[13:34:36][    236s][FT] round 2/6 epoch 1/5 | patches=10 | loss=1.1322
+[13:34:38][    237s][FT] round 2/6 epoch 2/5 | patches=10 | loss=1.0126
+[13:34:39][    239s][FT] round 2/6 epoch 3/5 | patches=10 | loss=0.9161
+[13:34:41][    240s][FT] round 2/6 epoch 4/5 | patches=10 | loss=0.8452
+[13:34:42][    241s][FT] round 2/6 epoch 5/5 | patches=10 | loss=0.8654
+[13:34:43][    243s][AL] round 2/6 | labeled=10 | selection_seconds=3.9000 round_seconds=13.8000 pseudo_student_dice=0.0965 pseudo_ensemble_dice=0.3979 consistency_scale=1.0000 val_dice=0.4151 val_vi=0.1911 val_ece=0.0070
+[13:34:47][    246s][AL] round 3: 147/676 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[13:34:49][    248s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.263, uniform=0.022, top1=0.307, top2=0.263, top3=0.263 -> using top1
+[13:34:49][    248s][WEIGHTS] round 3: teacher Dice on 15 labeled patches + 0 blank (65212 fg px): fly-mito=0.307, kasthuri=0.261, lucchi=0.014, multiclass=0.001 | weights: fly-mito=1.000, kasthuri=0.000, lucchi=0.000, multiclass=0.000
+[13:34:54][    253s][CONS] round 3: on the 5 new patches student Dice=0.400 vs ensemble 0.128 -> pseudo-labels from the student | consistency x0.00
+[13:34:54][    254s][FT] round 3/6 epoch 1/5 | patches=15 | loss=0.8282
+[13:34:55][    255s][FT] round 3/6 epoch 2/5 | patches=15 | loss=0.6492
+[13:34:56][    255s][FT] round 3/6 epoch 3/5 | patches=15 | loss=0.5065
+[13:34:57][    256s][FT] round 3/6 epoch 4/5 | patches=15 | loss=0.5487
+[13:34:58][    257s][FT] round 3/6 epoch 5/5 | patches=15 | loss=0.5168
+[13:34:59][    258s][AL] round 3/6 | labeled=15 | selection_seconds=3.9000 round_seconds=14.6000 pseudo_student_dice=0.3998 pseudo_ensemble_dice=0.1283 consistency_scale=0.0000 val_dice=0.5541 val_vi=0.1877 val_ece=0.0034
+[13:35:03][    262s][AL] round 4: 43/671 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[13:35:32][    291s][WEIGHTS] round 4: teacher agreement with the student on 666 unlabeled patches: fly-mito=0.377, kasthuri=0.417, lucchi=0.009, multiclass=0.002 | blended with labeled Dice at 0.50
+[13:35:58][    317s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.325, uniform=0.031, top1=0.328, top2=0.326, top3=0.325 -> using mixture
+[13:35:58][    317s][WEIGHTS] round 4: teacher Dice on 20 labeled patches + 0 blank (71602 fg px): fly-mito=0.328, kasthuri=0.325, lucchi=0.011, multiclass=0.002 | weights: fly-mito=0.126, kasthuri=0.872, lucchi=0.001, multiclass=0.001
+[13:36:03][    322s][CONS] round 4: on the 5 new patches student Dice=0.338 vs ensemble 0.000 -> pseudo-labels from the student | consistency x0.00
+[13:36:04][    323s][FT] round 4/6 epoch 1/5 | patches=20 | loss=0.5445
+[13:36:05][    324s][FT] round 4/6 epoch 2/5 | patches=20 | loss=0.5171
+[13:36:06][    325s][FT] round 4/6 epoch 3/5 | patches=20 | loss=0.4741
+[13:36:06][    326s][FT] round 4/6 epoch 4/5 | patches=20 | loss=0.4724
+[13:36:07][    327s][FT] round 4/6 epoch 5/5 | patches=20 | loss=0.4460
+[13:36:09][    328s][AL] round 4/6 | labeled=20 | selection_seconds=3.9000 round_seconds=68.6000 pseudo_student_dice=0.3375 pseudo_ensemble_dice=0.0001 consistency_scale=0.0000 val_dice=0.7049 val_vi=0.1246 val_ece=0.0086
+[13:36:12][    332s][AL] round 5: 136/666 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[13:36:42][    361s][WEIGHTS] round 5: teacher agreement with the student on 661 unlabeled patches: fly-mito=0.343, kasthuri=0.434, lucchi=0.010, multiclass=0.001 | blended with labeled Dice at 0.50
+[13:37:08][    387s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.312, uniform=0.035, top1=0.312, top2=0.312, top3=0.312 -> using mixture
+[13:37:08][    387s][WEIGHTS] round 5: teacher Dice on 25 labeled patches + 0 blank (93337 fg px): fly-mito=0.312, kasthuri=0.311, lucchi=0.010, multiclass=0.001 | weights: fly-mito=0.123, kasthuri=0.874, lucchi=0.001, multiclass=0.001
+[13:37:12][    392s][CONS] round 5: on the 5 new patches student Dice=0.425 vs ensemble 0.000 -> pseudo-labels from the student | consistency x0.00
+[13:37:14][    393s][FT] round 5/6 epoch 1/5 | patches=25 | loss=0.4196
+[13:37:15][    394s][FT] round 5/6 epoch 2/5 | patches=25 | loss=0.3617
+[13:37:16][    395s][FT] round 5/6 epoch 3/5 | patches=25 | loss=0.4179
+[13:37:17][    396s][FT] round 5/6 epoch 4/5 | patches=25 | loss=0.3490
+[13:37:18][    397s][FT] round 5/6 epoch 5/5 | patches=25 | loss=0.4248
+[13:37:19][    399s][AL] round 5/6 | labeled=25 | selection_seconds=3.8000 round_seconds=69.5000 pseudo_student_dice=0.4253 pseudo_ensemble_dice=0.0000 consistency_scale=0.0000 val_dice=0.6628 val_vi=0.1320 val_ece=0.0144
+[13:37:23][    402s][AL] round 6: 141/661 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[13:37:52][    432s][WEIGHTS] round 6: teacher agreement with the student on 656 unlabeled patches: fly-mito=0.359, kasthuri=0.426, lucchi=0.007, multiclass=0.001 | blended with labeled Dice at 0.50
+[13:38:18][    457s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.317, uniform=0.037, top1=0.316, top2=0.317, top3=0.317 -> using mixture
+[13:38:18][    457s][WEIGHTS] round 6: teacher Dice on 30 labeled patches + 0 blank (101937 fg px): fly-mito=0.314, kasthuri=0.316, lucchi=0.008, multiclass=0.001 | weights: fly-mito=0.121, kasthuri=0.877, lucchi=0.001, multiclass=0.001
+[13:38:23][    462s][CONS] round 6: on the 5 new patches student Dice=0.619 vs ensemble 0.407 -> pseudo-labels from the student | consistency x0.40
+[13:38:25][    465s][FT] round 6/6 epoch 1/5 | patches=30 | loss=0.4783
+[13:38:28][    467s][FT] round 6/6 epoch 2/5 | patches=30 | loss=0.4042
+[13:38:30][    470s][FT] round 6/6 epoch 3/5 | patches=30 | loss=0.3390
+[13:38:33][    472s][FT] round 6/6 epoch 4/5 | patches=30 | loss=0.3294
+[13:38:35][    475s][FT] round 6/6 epoch 5/5 | patches=30 | loss=0.3309
+[13:38:37][    476s][AL] round 6/6 | labeled=30 | selection_seconds=3.8000 round_seconds=76.3000 pseudo_student_dice=0.6186 pseudo_ensemble_dice=0.4073 consistency_scale=0.3950 val_dice=0.6835 val_vi=0.1245 val_ece=0.0127
+[13:38:40][    480s][CONS] final consolidation: pseudo-labels refreshed from the student
+[13:38:43][    482s][FT] final 1/10 epoch 1/1 | patches=30 | loss=0.3316
+[13:38:45][    485s][FT] final 2/10 epoch 1/1 | patches=30 | loss=0.2807
+[13:38:48][    487s][FT] final 3/10 epoch 1/1 | patches=30 | loss=0.3052
+[13:38:50][    490s][FT] final 4/10 epoch 1/1 | patches=30 | loss=0.2432
+[13:38:53][    492s][FT] final 5/10 epoch 1/1 | patches=30 | loss=0.2443
+[13:38:55][    495s][FT] final 6/10 epoch 1/1 | patches=30 | loss=0.2739
+[13:38:58][    497s][FT] final 7/10 epoch 1/1 | patches=30 | loss=0.2722
+[13:39:00][    500s][FT] final 8/10 epoch 1/1 | patches=30 | loss=0.2558
+[13:39:03][    502s][FT] final 9/10 epoch 1/1 | patches=30 | loss=0.2214
+[13:39:05][    505s][FT] final 10/10 epoch 1/1 | patches=30 | loss=0.2567
+[13:39:06][    506s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=28.7000 val_dice=0.7389 val_vi=0.1129 val_ece=0.0112
+[13:39:06][    506s][WEIGHTS] final: fly-mito=0.121, kasthuri=0.877, lucchi=0.001, multiclass=0.001
+[13:39:14][    513s][TEST] raw student (no EMA, no TTA): patch_dice=0.8171 patch_vi=0.1140
+[13:39:15][    514s][TEST] final model without TTA: patch_dice=0.8171 patch_vi=0.1140
+[13:39:15][    514s][RESOURCES] {"phases": {"setup": {"seconds": 0.77, "peak_cpu_rss_mb": 2191.8, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 11.09, "peak_cpu_rss_mb": 4114.3, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 117.87, "peak_cpu_rss_mb": 4292.8, "peak_gpu_alloc_mb": 3826.7, "peak_gpu_reserved_mb": 4168.0, "kd_epochs": 5}, "active_learning": {"seconds": 374.13, "peak_cpu_rss_mb": 4406.5, "peak_gpu_alloc_mb": 3947.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 8.56, "peak_cpu_rss_mb": 4406.5, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 147, "images": 3, "patches_per_second": 17.2, "seconds_per_image": 2.853}}, "peak_gpu_alloc_mb": 3947.0, "peak_cpu_rss_mb": 4406.5, "total_seconds": 512.4}
+[13:39:15][    514s][TEST] patch_dice=0.8349 patch_iou=0.7587 patch_precision=0.8561 patch_recall=0.8840 patch_vi=0.1017 image_dice=0.9032 image_iou=0.8241 image_precision=0.8944 image_recall=0.9141 image_vi=0.1090 cal_ece=0.0030 cal_nll=0.0326 cal_brier=0.0094 cal_mean_confidence=0.9897 cal_accuracy=0.9869 cal_predicted_prevalence=0.0606 cal_true_prevalence=0.0594 cal_mean_probability=0.0604
+[13:39:15][    515s][DONE] drosophila-vnc macs-v11-norestart-acq-margin b5 s2 | test patch Dice=0.8349 VI=0.1017 | image Dice=0.9032 VI=0.1090 | 514s

macs_v2_r26/runs/mitochondria/drosophila-vnc/macs-v11-norestart-acq-margin/budget-5/seed-2/result.json

@@ -0,0 +1,662 @@
+{
+  "target": "drosophila-vnc",
+  "variant": "macs-v11-norestart-acq-margin",
+  "budget": 5.0,
+  "seed": 2,
+  "config": {
+    "task": "mitochondria",
+    "target": "drosophila-vnc",
+    "variant": "macs-v11-norestart-acq-margin",
+    "budget": 5.0,
+    "seed": 2,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "drosophila-vnc",
+      "--variants",
+      "macs-v11-norestart-acq-margin",
+      "--budgets",
+      "5",
+      "--seeds",
+      "2",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--task",
+      "mitochondria",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "commit": "aecf380",
+    "dirty": true
+  },
+  "sources": [
+    "fly-brain-mito-benchmark",
+    "kasthuri-pp",
+    "lucchi-pp",
+    "multiclass"
+  ],
+  "alphas": [
+    0.11690492516893997,
+    0.8375897330633602,
+    0.0256027087898516,
+    0.019902632977848233
+  ],
+  "weights_all": {
+    "sources": [
+      "fly-brain-mito-benchmark",
+      "kasthuri-pp",
+      "lucchi-pp",
+      "multiclass"
+    ],
+    "mixture": [
+      0.11690492516893997,
+      0.8375897330633602,
+      0.0256027087898516,
+      0.019902632977848233
+    ],
+    "per_source": [
+      0.038599207620062845,
+      0.9432861284385984,
+      0.013712296406123614,
+      0.0044023675352152245
+    ],
+    "uniform": [
+      0.25,
+      0.25,
+      0.25,
+      0.25
+    ],
+    "mixture_info": {
+      "objective": 0.1917820351790263,
+      "mmd2_mixture": 0.19749363071106324,
+      "mmd2_uniform": 0.3743489659922599,
+      "mmd2_single_source": [
+        0.7042977370508033,
+        0.20706860465273969,
+        0.6307956892595754,
+        0.6534362949819378
+      ],
+      "median_squared_distance": 0.19645890982127545,
+      "gammas": [
+        1.272530730356961,
+        2.545061460713922,
+        5.090122921427844,
+        10.180245842855689,
+        20.360491685711377
+      ],
+      "max_weight": 0.8375897330633602,
+      "entropy": 0.5711595532036933,
+      "effective_sources": 1.7703186403916105,
+      "min_weight": 0.019902632977848233
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.605076346545893,
+        0.31101778520145595,
+        0.7002952057460718,
+        0.8048259839560096
+      ],
+      "standardized": [
+        -0.0012362667606482206,
+        -1.5993052227157551,
+        0.5162331182950395,
+        1.0843083711813646
+      ],
+      "tau": 0.5,
+      "max_weight": 0.9432861284385984,
+      "entropy": 0.26340029394025066,
+      "effective_sources": 1.301347536237684,
+      "min_weight": 0.0044023675352152245
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "kasthuri-pp",
+    "seconds": 1.7
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.27840715557699913,
+      0.18341499813983803,
+      0.0577032083417681,
+      0.01681466600081078
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.18652277313599772,
+      "mixture@raw": 0.18548230728775292,
+      "mixture@logit": 0.19296074616265912,
+      "per_source": 0.18415443581907806,
+      "per_source@raw": 0.1835051571090219,
+      "per_source@logit": 0.18905485087217583,
+      "uniform": 0.0696943056822927,
+      "uniform@raw": 0.08847159266818436,
+      "uniform@logit": 0.06549734203813108
+    },
+    "val_patches": 147
+  },
+  "temperatures": [
+    1.3822230100631714,
+    2.5090832710266113,
+    3.80090069770813,
+    2.597146987915039
+  ],
+  "top_source": "kasthuri-pp",
+  "weight_stats": {
+    "max_weight": 0.8375897330633602,
+    "entropy": 0.5711595532036933,
+    "effective_sources": 1.7703186403916105,
+    "min_weight": 0.019902632977848233
+  },
+  "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": 117.9,
+  "round0_val": {
+    "val_dice": 0.20583852352297866,
+    "val_vi": 0.2206894295217048,
+    "val_ece": 0.019786654639967937
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 5,
+      "selection_seconds": 18.3,
+      "round_seconds": 94.9,
+      "teacher_dice_labeled": [
+        0.36295908671523197,
+        0.5310285360135809,
+        0.013736820762368207,
+        0.0019869574077847974
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 19579,
+      "evidence_nonempty_patches": 5,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.583215536844693,
+      "restart_teacher_dice": 0.5310285360135809,
+      "weights": [
+        0.025328879887354556,
+        0.9743856190601974,
+        0.00016881695486452877,
+        0.0001166840975834896
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.531149750640248,
+        "uniform": 0.0667619235317304,
+        "top1": 0.5310285360135809,
+        "top2": 0.5311354323916325,
+        "top3": 0.5311354323916325
+      },
+      "val_dice": 0.22762768970263805,
+      "val_vi": 0.2139711243244451,
+      "val_ece": 0.03221462987886444
+    },
+    {
+      "round": 2,
+      "n_labeled": 10,
+      "selection_seconds": 3.9,
+      "round_seconds": 13.8,
+      "teacher_dice_labeled": [
+        0.38401783708865406,
+        0.28656576040595816,
+        0.006609724978735469,
+        0.001720193654790828
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 56186,
+      "evidence_nonempty_patches": 10,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.29303699199791533,
+        "uniform": 0.024812635707781746,
+        "top1": 0.38401783708865406,
+        "top2": 0.29324058902043343,
+        "top3": 0.2931189822302791
+      },
+      "pseudo_student_dice": 0.09651973250447396,
+      "pseudo_ensemble_dice": 0.3979095217735406,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.41508196440410405,
+      "val_vi": 0.1910820541092168,
+      "val_ece": 0.007020516502335978
+    },
+    {
+      "round": 3,
+      "n_labeled": 15,
+      "selection_seconds": 3.9,
+      "round_seconds": 14.6,
+      "teacher_dice_labeled": [
+        0.3067076899556643,
+        0.2610554773924798,
+        0.014201309254393293,
+        0.0014810969278690527
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 65212,
+      "evidence_nonempty_patches": 14,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "ensemble_choice": "top1",
+      "candidate_dice": {
+        "mixture": 0.26331644529645354,
+        "uniform": 0.022352339093262688,
+        "top1": 0.3067076899556643,
+        "top2": 0.2632606335914767,
+        "top3": 0.26327461896504517
+      },
+      "pseudo_student_dice": 0.39983894945490583,
+      "pseudo_ensemble_dice": 0.12834515276777023,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.5541464148558664,
+      "val_vi": 0.18765088084462966,
+      "val_ece": 0.0034033482403390374
+    },
+    {
+      "round": 4,
+      "n_labeled": 20,
+      "selection_seconds": 3.9,
+      "round_seconds": 68.6,
+      "teacher_dice_labeled": [
+        0.32843252673042683,
+        0.32500226774835783,
+        0.011205289601600254,
+        0.001607866862380934
+      ],
+      "teacher_dice_pool": [
+        0.37703511707654025,
+        0.41697163557551786,
+        0.009469044489436806,
+        0.001867650226971613
+      ],
+      "evidence_fg_pixels": 71602,
+      "evidence_nonempty_patches": 15,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.12596412966410978,
+        0.8720632836330818,
+        0.0011561112109388508,
+        0.0008164754918695407
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.3254468576327058,
+        "uniform": 0.03129050120454881,
+        "top1": 0.32843252673042683,
+        "top2": 0.3255205919573236,
+        "top3": 0.32547291784264754
+      },
+      "pseudo_student_dice": 0.33754523002699444,
+      "pseudo_ensemble_dice": 7.686395080707149e-05,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.7049201214465788,
+      "val_vi": 0.12461696322525986,
+      "val_ece": 0.008641253938814855
+    },
+    {
+      "round": 5,
+      "n_labeled": 25,
+      "selection_seconds": 3.8,
+      "round_seconds": 69.5,
+      "teacher_dice_labeled": [
+        0.31201821320261003,
+        0.3108151802191612,
+        0.009948750155924245,
+        0.0009971843286295514
+      ],
+      "teacher_dice_pool": [
+        0.3428589401009912,
+        0.43397047373891273,
+        0.00959118250712609,
+        0.0009592051120907552
+      ],
+      "evidence_fg_pixels": 93337,
+      "evidence_nonempty_patches": 20,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.12349790890107877,
+        0.8742455597184858,
+        0.0013189906070092267,
+        0.0009375407734261823
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.3118025581344997,
+        "uniform": 0.03485867907823771,
+        "top1": 0.31201821320261003,
+        "top2": 0.3118660509609209,
+        "top3": 0.31182082107141296
+      },
+      "pseudo_student_dice": 0.42525555049156843,
+      "pseudo_ensemble_dice": 3.389256058295204e-05,
+      "pseudo_source": "student",
+      "consistency_scale": 0.0,
+      "val_dice": 0.6627517799862803,
+      "val_vi": 0.13200599901825022,
+      "val_ece": 0.014428097276413115
+    },
+    {
+      "round": 6,
+      "n_labeled": 30,
+      "selection_seconds": 3.8,
+      "round_seconds": 76.3,
+      "teacher_dice_labeled": [
+        0.31439930418264694,
+        0.3158112661728686,
+        0.008469861850983751,
+        0.000731929251973725
+      ],
+      "teacher_dice_pool": [
+        0.35933004834555277,
+        0.42554124651196956,
+        0.006534919388207661,
+        0.0005166021294979816
+      ],
+      "evidence_fg_pixels": 101937,
+      "evidence_nonempty_patches": 25,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.12071086575697663,
+        0.8771562753946593,
+        0.0012404086509053586,
+        0.0008924501974586932
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.3167068292410805,
+        "uniform": 0.03696126287935871,
+        "top1": 0.3158112661728686,
+        "top2": 0.3167952146594093,
+        "top3": 0.31674336464528546
+      },
+      "pseudo_student_dice": 0.6185659343554081,
+      "pseudo_ensemble_dice": 0.40730560578661845,
+      "pseudo_source": "student",
+      "consistency_scale": 0.395,
+      "val_dice": 0.6835278718987173,
+      "val_vi": 0.12453932816493471,
+      "val_ece": 0.012681996756738407
+    },
+    {
+      "round": 7,
+      "n_labeled": 30,
+      "selection_seconds": 0.0,
+      "round_seconds": 28.7,
+      "final_consolidation": true,
+      "val_dice": 0.7389110295338063,
+      "val_vi": 0.11285964003514103,
+      "val_ece": 0.011176637690902614
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.07993511898383816,
+    "auroc_top20_error": 0.5728207976303819
+  },
+  "al_seconds": 374.1,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.12071086575697663,
+    0.8771562753946593,
+    0.0012404086509053586,
+    0.0008924501974586932
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.8771562753946593,
+    "entropy": 0.38476242814608885,
+    "effective_sources": 1.4692652239117807,
+    "min_weight": 0.0008924501974586932
+  },
+  "top_source_final": "kasthuri-pp",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.8170945417514479,
+    "patch_iou": 0.7325863544605868,
+    "patch_precision": 0.8206006516040203,
+    "patch_recall": 0.8926114311565548,
+    "patch_vi": 0.11397135888121473,
+    "n_patches": 147,
+    "cal_ece": 0.009251723975275686,
+    "cal_nll": 0.059870382370229576,
+    "cal_brier": 0.0121247786045106,
+    "cal_mean_confidence": 0.9941307853684778,
+    "cal_accuracy": 0.9849728954081632,
+    "cal_predicted_prevalence": 0.06171609268707483,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.06030104652931691
+  },
+  "test_notta": {
+    "patch_dice": 0.8170945417514479,
+    "patch_iou": 0.7325863544605868,
+    "patch_precision": 0.8206006516040203,
+    "patch_recall": 0.8926114311565548,
+    "patch_vi": 0.11397135888121473,
+    "n_patches": 147,
+    "cal_ece": 0.009251723975275686,
+    "cal_nll": 0.059870382370229576,
+    "cal_brier": 0.0121247786045106,
+    "cal_mean_confidence": 0.9941307853684778,
+    "cal_accuracy": 0.9849728954081632,
+    "cal_predicted_prevalence": 0.06171609268707483,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.06030104652931691
+  },
+  "test": {
+    "patch_dice": 0.8348826233141908,
+    "patch_iou": 0.7586723040300558,
+    "patch_precision": 0.8560631492580926,
+    "patch_recall": 0.8840063786551214,
+    "patch_vi": 0.10169670823361057,
+    "n_patches": 147,
+    "image_dice": 0.9031688922650402,
+    "image_iou": 0.8240584503474547,
+    "image_precision": 0.894351360788363,
+    "image_recall": 0.914061361626028,
+    "image_vi": 0.10897988313866609,
+    "n_images": 3,
+    "cal_ece": 0.003019011167729832,
+    "cal_nll": 0.0325651706703299,
+    "cal_brier": 0.009432980359557166,
+    "cal_mean_confidence": 0.9897270955516048,
+    "cal_accuracy": 0.9869393069727891,
+    "cal_predicted_prevalence": 0.06056016156462585,
+    "cal_true_prevalence": 0.05939094387755102,
+    "cal_mean_probability": 0.06044280983044895
+  },
+  "eval_seconds": 8.6,
+  "total_seconds": 514.4,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 0.77,
+        "peak_cpu_rss_mb": 2191.8,
+        "peak_gpu_alloc_mb": 592.4,
+        "peak_gpu_reserved_mb": 594.0
+      },
+      "features_and_weights": {
+        "seconds": 11.09,
+        "peak_cpu_rss_mb": 4114.3,
+        "peak_gpu_alloc_mb": 2298.2,
+        "peak_gpu_reserved_mb": 2640.0
+      },
+      "distillation": {
+        "seconds": 117.87,
+        "peak_cpu_rss_mb": 4292.8,
+        "peak_gpu_alloc_mb": 3826.7,
+        "peak_gpu_reserved_mb": 4168.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 374.13,
+        "peak_cpu_rss_mb": 4406.5,
+        "peak_gpu_alloc_mb": 3947.0,
+        "peak_gpu_reserved_mb": 4176.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 8.56,
+        "peak_cpu_rss_mb": 4406.5,
+        "peak_gpu_alloc_mb": 2149.5,
+        "peak_gpu_reserved_mb": 4176.0,
+        "patches": 147,
+        "images": 3,
+        "patches_per_second": 17.2,
+        "seconds_per_image": 2.853
+      }
+    },
+    "peak_gpu_alloc_mb": 3947.0,
+    "peak_cpu_rss_mb": 4406.5,
+    "total_seconds": 512.4
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/drosophila-vnc/macs-v11-norestart-acq-margin_b5_s2.pth"
+}

macs_v2_r26/runs/mitochondria/drosophila-vnc/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
+1,5,18.3,94.9,"[0.36295908671523197, 0.5310285360135809, 0.013736820762368207, 0.0019869574077847974]",,19579,5,True,kept,0.583215536844693,0.5310285360135809,"[0.025328879887354556, 0.9743856190601974, 0.00016881695486452877, 0.0001166840975834896]",mixture,"{'mixture': 0.531149750640248, 'uniform': 0.0667619235317304, 'top1': 0.5310285360135809, 'top2': 0.5311354323916325, 'top3': 0.5311354323916325}",0.22762768970263805,0.2139711243244451,0.03221462987886444,,,,,
+2,10,3.9,13.8,"[0.38401783708865406, 0.28656576040595816, 0.006609724978735469, 0.001720193654790828]",,56186,10,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.29303699199791533, 'uniform': 0.024812635707781746, 'top1': 0.38401783708865406, 'top2': 0.29324058902043343, 'top3': 0.2931189822302791}",0.41508196440410405,0.1910820541092168,0.007020516502335978,0.09651973250447396,0.3979095217735406,ensemble,1.0,
+3,15,3.9,14.6,"[0.3067076899556643, 0.2610554773924798, 0.014201309254393293, 0.0014810969278690527]",,65212,14,False,,,,"[1.0, 0.0, 0.0, 0.0]",top1,"{'mixture': 0.26331644529645354, 'uniform': 0.022352339093262688, 'top1': 0.3067076899556643, 'top2': 0.2632606335914767, 'top3': 0.26327461896504517}",0.5541464148558664,0.18765088084462966,0.0034033482403390374,0.39983894945490583,0.12834515276777023,student,0.0,
+4,20,3.9,68.6,"[0.32843252673042683, 0.32500226774835783, 0.011205289601600254, 0.001607866862380934]","[0.37703511707654025, 0.41697163557551786, 0.009469044489436806, 0.001867650226971613]",71602,15,False,,,,"[0.12596412966410978, 0.8720632836330818, 0.0011561112109388508, 0.0008164754918695407]",mixture,"{'mixture': 0.3254468576327058, 'uniform': 0.03129050120454881, 'top1': 0.32843252673042683, 'top2': 0.3255205919573236, 'top3': 0.32547291784264754}",0.7049201214465788,0.12461696322525986,0.008641253938814855,0.33754523002699444,7.686395080707149e-05,student,0.0,
+5,25,3.8,69.5,"[0.31201821320261003, 0.3108151802191612, 0.009948750155924245, 0.0009971843286295514]","[0.3428589401009912, 0.43397047373891273, 0.00959118250712609, 0.0009592051120907552]",93337,20,False,,,,"[0.12349790890107877, 0.8742455597184858, 0.0013189906070092267, 0.0009375407734261823]",mixture,"{'mixture': 0.3118025581344997, 'uniform': 0.03485867907823771, 'top1': 0.31201821320261003, 'top2': 0.3118660509609209, 'top3': 0.31182082107141296}",0.6627517799862803,0.13200599901825022,0.014428097276413115,0.42525555049156843,3.389256058295204e-05,student,0.0,
+6,30,3.8,76.3,"[0.31439930418264694, 0.3158112661728686, 0.008469861850983751, 0.000731929251973725]","[0.35933004834555277, 0.42554124651196956, 0.006534919388207661, 0.0005166021294979816]",101937,25,False,,,,"[0.12071086575697663, 0.8771562753946593, 0.0012404086509053586, 0.0008924501974586932]",mixture,"{'mixture': 0.3167068292410805, 'uniform': 0.03696126287935871, 'top1': 0.3158112661728686, 'top2': 0.3167952146594093, 'top3': 0.31674336464528546}",0.6835278718987173,0.12453932816493471,0.012681996756738407,0.6185659343554081,0.40730560578661845,student,0.395,
+7,30,0.0,28.7,,,,,,,,,,,,0.7389110295338063,0.11285964003514103,0.011176637690902614,,,,,True

macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-entropy/budget-5/seed-1/log.txt

@@ -0,0 +1,108 @@
+[07:50:03][      0s][START] target=kasthuri-pp variant=macs-v11-norestart-acq-entropy budget=5.0% seed=1 | {"task": "mitochondria", "target": "kasthuri-pp", "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": 8, "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}
+[07:50:03][      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.
+[07:50:04][      1s][DATA] kasthuri-pp: train=12023 val=2376 test=2376 patches
+[07:50:05][      2s][CALIB] source-val temperatures: fly-mito=1.38, lucchi=3.80, dros-vnc=2.77, multiclass=2.60
+[07:50:05][      2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s
+[07:50:06][      4s][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
+[07:50:06][      4s][WEIGHTS] per-source: fly-mito=0.004, lucchi=0.027, dros-vnc=0.949, multiclass=0.020 | min-MMD source (NeuroADDA rule) = dros-vnc
+[07:50:45][     42s][TEACHERS] target-val Dice: fly-mito=0.194, lucchi=0.137, dros-vnc=0.349, multiclass=0.152
+[07:50:45][     42s][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
+[07:50:45][     42s][WEIGHTS] using mixture: fly-mito=0.055, lucchi=0.143, dros-vnc=0.678, multiclass=0.124
+[07:50:45][     42s][INIT] student initialised from teacher dros-vnc (weight 0.678)
+[07:50:54][     51s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[07:50:54][     51s][KD] distilling the mixture ensemble (prob space) into the student
+[07:52:56][    173s][KD] soft labels for 6000 patches from 4 teachers in 122s
+[07:55:54][    351s][KD] epoch 1/5 | patches=6000 | loss=0.1770 | 178s
+[07:58:51][    529s][KD] epoch 2/5 | patches=6000 | loss=0.1513 | 178s
+[08:01:49][    706s][KD] epoch 3/5 | patches=6000 | loss=0.1493 | 178s
+[08:04:47][    884s][KD] epoch 4/5 | patches=6000 | loss=0.1449 | 178s
+[08:07:45][   1062s][KD] epoch 5/5 | patches=6000 | loss=0.1438 | 178s
+[08:07:51][   1068s][ROUND0] before any target label: val_dice=0.2223 val_vi=0.1502
+[08:07:51][   1068s][AL] patches=12023 (pool 10327 after dropping blank; free blank negatives: True) | budget=601 (5.0%) | rounds=6 | per_round=100 | acquisition=entropy | first_round=stratified
+[08:08:33][   1110s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': -0.3458863863149128, 'auroc_top20_error': 0.36265166015625}
+[08:09:14][   1151s][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
+[08:09:30][   1167s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.203, uniform=0.004, top1=0.204, top2=0.205, top3=0.204 -> using mixture
+[08:09:32][   1169s][WEIGHTS] round 1: student kept (labeled Dice 0.038 vs best teacher dros-vnc 0.204); distilling the corrected ensemble into it
+[08:09:32][   1169s][WEIGHTS] round 1: teacher Dice on 100 labeled patches + 100 blank (277958 fg px): fly-mito=0.038, lucchi=0.009, dros-vnc=0.204, multiclass=0.002 | weights: fly-mito=0.014, lucchi=0.028, dros-vnc=0.935, multiclass=0.023
+[08:11:34][   1292s][KD] soft labels for 6000 patches from 4 teachers in 123s
+[08:14:33][   1470s][KD] epoch 1/2 | patches=6000 | loss=0.0603 | 178s
+[08:17:31][   1648s][KD] epoch 2/2 | patches=6000 | loss=0.0593 | 178s
+[08:18:12][   1689s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 41s
+[08:18:24][   1702s][FT] round 1/6 epoch 1/5 | patches=200 | loss=0.9295
+[08:18:37][   1714s][FT] round 1/6 epoch 2/5 | patches=200 | loss=0.8602
+[08:18:49][   1727s][FT] round 1/6 epoch 3/5 | patches=200 | loss=0.5114
+[08:19:02][   1739s][FT] round 1/6 epoch 4/5 | patches=200 | loss=0.4687
+[08:19:15][   1752s][FT] round 1/6 epoch 5/5 | patches=200 | loss=0.3942
+[08:19:20][   1758s][AL] round 1/6 | labeled=100 | selection_seconds=83.3000 round_seconds=683.6000 restart_student_dice=0.0379 restart_teacher_dice=0.2043 val_dice=0.7305 val_vi=0.0986 val_ece=0.0061
+[08:20:02][   1799s][AL] round 2: 2781/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[08:20:33][   1830s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.135, uniform=0.014, top1=0.137, top2=0.137, top3=0.136 -> using mixture
+[08:20:33][   1830s][WEIGHTS] round 2: teacher Dice on 200 labeled patches + 200 blank (1027506 fg px): fly-mito=0.076, lucchi=0.033, dros-vnc=0.137, multiclass=0.008 | weights: fly-mito=0.038, lucchi=0.064, dros-vnc=0.855, multiclass=0.043
+[08:20:54][   1851s][CONS] round 2: on the 100 new patches student Dice=0.590 vs ensemble 0.109 -> pseudo-labels from the student | consistency x0.30
+[08:21:18][   1875s][FT] round 2/6 epoch 1/5 | patches=400 | loss=0.2601
+[08:21:42][   1900s][FT] round 2/6 epoch 2/5 | patches=400 | loss=0.2320
+[08:22:07][   1924s][FT] round 2/6 epoch 3/5 | patches=400 | loss=0.2258
+[08:22:31][   1948s][FT] round 2/6 epoch 4/5 | patches=400 | loss=0.1876
+[08:22:55][   1973s][FT] round 2/6 epoch 5/5 | patches=400 | loss=0.1990
+[08:23:01][   1978s][AL] round 2/6 | labeled=200 | selection_seconds=42.0000 round_seconds=214.9000 pseudo_student_dice=0.5901 pseudo_ensemble_dice=0.1087 consistency_scale=0.3000 val_dice=0.8368 val_vi=0.0604 val_ece=0.0019
+[08:23:43][   2020s][AL] round 3: 2240/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[08:25:28][   2126s][WEIGHTS] round 3: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.116, lucchi=0.014, dros-vnc=0.264, multiclass=0.001 | blended with labeled Dice at 0.50
+[08:27:04][   2221s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.214, uniform=0.013, top1=0.215, top2=0.215, top3=0.214 -> using mixture
+[08:27:04][   2221s][WEIGHTS] round 3: teacher Dice on 300 labeled patches + 300 blank (1431921 fg px): fly-mito=0.094, lucchi=0.020, dros-vnc=0.215, multiclass=0.004 | weights: fly-mito=0.023, lucchi=0.028, dros-vnc=0.929, multiclass=0.021
+[08:27:25][   2242s][CONS] round 3: on the 100 new patches student Dice=0.816 vs ensemble 0.232 -> pseudo-labels from the student | consistency x1.00
+[08:28:01][   2278s][FT] round 3/6 epoch 1/5 | patches=600 | loss=0.2296
+[08:28:37][   2314s][FT] round 3/6 epoch 2/5 | patches=600 | loss=0.2064
+[08:29:13][   2350s][FT] round 3/6 epoch 3/5 | patches=600 | loss=0.1887
+[08:29:49][   2386s][FT] round 3/6 epoch 4/5 | patches=600 | loss=0.1811
+[08:30:25][   2422s][FT] round 3/6 epoch 5/5 | patches=600 | loss=0.1760
+[08:30:31][   2428s][AL] round 3/6 | labeled=300 | selection_seconds=41.9000 round_seconds=443.8000 pseudo_student_dice=0.8157 pseudo_ensemble_dice=0.2319 consistency_scale=1.0000 val_dice=0.8475 val_vi=0.0524 val_ece=0.0034
+[08:31:13][   2470s][AL] round 4: 2399/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[08:33:05][   2583s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.110, lucchi=0.014, dros-vnc=0.272, multiclass=0.002 | blended with labeled Dice at 0.50
+[08:34:48][   2685s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.213, uniform=0.012, top1=0.214, top2=0.214, top3=0.213 -> using mixture
+[08:34:48][   2685s][WEIGHTS] round 4: teacher Dice on 400 labeled patches + 400 blank (1887583 fg px): fly-mito=0.091, lucchi=0.018, dros-vnc=0.214, multiclass=0.004 | weights: fly-mito=0.022, lucchi=0.028, dros-vnc=0.930, multiclass=0.021
+[08:35:09][   2706s][CONS] round 4: on the 100 new patches student Dice=0.820 vs ensemble 0.124 -> pseudo-labels from the student | consistency x1.00
+[08:35:57][   2754s][FT] round 4/6 epoch 1/5 | patches=800 | loss=0.1679
+[08:36:44][   2801s][FT] round 4/6 epoch 2/5 | patches=800 | loss=0.1625
+[08:37:32][   2849s][FT] round 4/6 epoch 3/5 | patches=800 | loss=0.1553
+[08:38:19][   2897s][FT] round 4/6 epoch 4/5 | patches=800 | loss=0.1472
+[08:39:07][   2944s][FT] round 4/6 epoch 5/5 | patches=800 | loss=0.1651
+[08:39:13][   2950s][AL] round 4/6 | labeled=400 | selection_seconds=42.0000 round_seconds=516.3000 pseudo_student_dice=0.8200 pseudo_ensemble_dice=0.1235 consistency_scale=1.0000 val_dice=0.8402 val_vi=0.0520 val_ece=0.0027
+[08:39:55][   2992s][AL] round 5: 2595/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[08:41:55][   3112s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.111, lucchi=0.013, dros-vnc=0.290, multiclass=0.001 | blended with labeled Dice at 0.50
+[08:43:45][   3222s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.220, uniform=0.013, top1=0.221, top2=0.221, top3=0.221 -> using mixture
+[08:43:45][   3222s][WEIGHTS] round 5: teacher Dice on 500 labeled patches + 500 blank (2417674 fg px): fly-mito=0.091, lucchi=0.017, dros-vnc=0.221, multiclass=0.005 | weights: fly-mito=0.021, lucchi=0.026, dros-vnc=0.934, multiclass=0.020
+[08:44:06][   3243s][CONS] round 5: on the 100 new patches student Dice=0.806 vs ensemble 0.138 -> pseudo-labels from the student | consistency x1.00
+[08:45:05][   3302s][FT] round 5/6 epoch 1/5 | patches=1000 | loss=0.1638
+[08:46:04][   3362s][FT] round 5/6 epoch 2/5 | patches=1000 | loss=0.1541
+[08:47:04][   3421s][FT] round 5/6 epoch 3/5 | patches=1000 | loss=0.1552
+[08:48:03][   3480s][FT] round 5/6 epoch 4/5 | patches=1000 | loss=0.1423
+[08:49:02][   3540s][FT] round 5/6 epoch 5/5 | patches=1000 | loss=0.1438
+[08:49:08][   3546s][AL] round 5/6 | labeled=500 | selection_seconds=42.0000 round_seconds=589.6000 pseudo_student_dice=0.8063 pseudo_ensemble_dice=0.1379 consistency_scale=1.0000 val_dice=0.8604 val_vi=0.0508 val_ece=0.0027
+[08:49:50][   3587s][AL] round 6: 2336/8000 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[08:51:51][   3708s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.116, lucchi=0.015, dros-vnc=0.272, multiclass=0.003 | blended with labeled Dice at 0.50
+[08:53:42][   3819s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.214, uniform=0.013, top1=0.216, top2=0.216, top3=0.215 -> using mixture
+[08:53:42][   3819s][WEIGHTS] round 6: teacher Dice on 512 labeled patches + 512 blank (2411238 fg px): fly-mito=0.093, lucchi=0.018, dros-vnc=0.216, multiclass=0.005 | weights: fly-mito=0.022, lucchi=0.027, dros-vnc=0.930, multiclass=0.021
+[08:54:03][   3840s][CONS] round 6: on the 100 new patches student Dice=0.830 vs ensemble 0.171 -> pseudo-labels from the student | consistency x1.00
+[08:55:14][   3911s][FT] round 6/6 epoch 1/5 | patches=1200 | loss=0.1388
+[08:56:25][   3982s][FT] round 6/6 epoch 2/5 | patches=1200 | loss=0.1365
+[08:57:36][   4053s][FT] round 6/6 epoch 3/5 | patches=1200 | loss=0.1608
+[08:58:47][   4124s][FT] round 6/6 epoch 4/5 | patches=1200 | loss=0.1408
+[08:59:58][   4196s][FT] round 6/6 epoch 5/5 | patches=1200 | loss=0.1164
+[09:00:04][   4201s][AL] round 6/6 | labeled=600 | selection_seconds=42.0000 round_seconds=650.0000 pseudo_student_dice=0.8301 pseudo_ensemble_dice=0.1706 consistency_scale=1.0000 val_dice=0.8608 val_vi=0.0520 val_ece=0.0029
+[09:00:14][   4212s][CONS] final consolidation: pseudo-labels refreshed from the student
+[09:01:25][   4282s][FT] final 1/10 epoch 1/1 | patches=1200 | loss=0.1211
+[09:02:36][   4353s][FT] final 2/10 epoch 1/1 | patches=1200 | loss=0.1199
+[09:03:47][   4425s][FT] final 3/10 epoch 1/1 | patches=1200 | loss=0.1234
+[09:04:58][   4496s][FT] final 4/10 epoch 1/1 | patches=1200 | loss=0.1219
+[09:06:09][   4567s][FT] final 5/10 epoch 1/1 | patches=1200 | loss=0.1027
+[09:07:21][   4638s][FT] final 6/10 epoch 1/1 | patches=1200 | loss=0.1009
+[09:08:32][   4709s][FT] final 7/10 epoch 1/1 | patches=1200 | loss=0.0915
+[09:09:43][   4780s][FT] final 8/10 epoch 1/1 | patches=1200 | loss=0.0918
+[09:10:54][   4851s][FT] final 9/10 epoch 1/1 | patches=1200 | loss=0.0996
+[09:12:05][   4923s][FT] final 10/10 epoch 1/1 | patches=1200 | loss=0.0947
+[09:12:11][   4928s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=721.2000 val_dice=0.8854 val_vi=0.0413 val_ece=0.0029
+[09:12:11][   4928s][WEIGHTS] final: fly-mito=0.022, lucchi=0.027, dros-vnc=0.930, multiclass=0.021
+[09:14:02][   5040s][TEST] raw student (no EMA, no TTA): patch_dice=0.8665 patch_vi=0.0276
+[09:14:15][   5053s][TEST] final model without TTA: patch_dice=0.8665 patch_vi=0.0276
+[09:14:15][   5053s][RESOURCES] {"phases": {"setup": {"seconds": 0.75, "peak_cpu_rss_mb": 2191.9, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 48.62, "peak_cpu_rss_mb": 4172.5, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 1011.42, "peak_cpu_rss_mb": 5948.1, "peak_gpu_alloc_mb": 3826.7, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 3860.3, "peak_cpu_rss_mb": 5948.1, "peak_gpu_alloc_mb": 3947.0, "peak_gpu_reserved_mb": 4170.0, "rounds": 6}, "inference": {"seconds": 124.22, "peak_cpu_rss_mb": 5948.1, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4172.0, "patches": 2376, "images": 24, "patches_per_second": 19.1, "seconds_per_image": 5.176}}, "peak_gpu_alloc_mb": 3947.0, "peak_cpu_rss_mb": 5948.1, "total_seconds": 5045.3}
+[09:14:15][   5053s][TEST] patch_dice=0.8759 patch_iou=0.8475 patch_precision=0.9330 patch_recall=0.9041 patch_vi=0.0263 image_dice=0.9468 image_iou=0.8995 image_precision=0.9461 image_recall=0.9481 image_vi=0.0320 cal_ece=0.0016 cal_nll=0.0114 cal_brier=0.0026 cal_mean_confidence=0.9985 cal_accuracy=0.9969 cal_predicted_prevalence=0.0271 cal_true_prevalence=0.0268 cal_mean_probability=0.0270
+[09:14:16][   5054s][DONE] kasthuri-pp macs-v11-norestart-acq-entropy b5 s1 | test patch Dice=0.8759 VI=0.0263 | image Dice=0.9468 VI=0.0320 | 5053s

macs_v2_r26/runs/mitochondria/kasthuri-pp/macs-v11-norestart-acq-entropy/budget-5/seed-1/result.json

@@ -0,0 +1,667 @@
+{
+  "target": "kasthuri-pp",
+  "variant": "macs-v11-norestart-acq-entropy",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "mitochondria",
+    "target": "kasthuri-pp",
+    "variant": "macs-v11-norestart-acq-entropy",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "kasthuri-pp",
+      "--variants",
+      "macs-v11-norestart-acq-entropy",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--task",
+      "mitochondria",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "commit": "aecf380",
+    "dirty": true
+  },
+  "sources": [
+    "fly-brain-mito-benchmark",
+    "lucchi-pp",
+    "drosophila-vnc",
+    "multiclass"
+  ],
+  "alphas": [
+    0.055262541562243715,
+    0.1430436080615447,
+    0.6777947281235579,
+    0.12389912225265376
+  ],
+  "weights_all": {
+    "sources": [
+      "fly-brain-mito-benchmark",
+      "lucchi-pp",
+      "drosophila-vnc",
+      "multiclass"
+    ],
+    "mixture": [
+      0.055262541562243715,
+      0.1430436080615447,
+      0.6777947281235579,
+      0.12389912225265376
+    ],
+    "per_source": [
+      0.00394341763649415,
+      0.027391591802199966,
+      0.9486104893290658,
+      0.020054501232240087
+    ],
+    "uniform": [
+      0.25,
+      0.25,
+      0.25,
+      0.25
+    ],
+    "mixture_info": {
+      "objective": 0.11610906469392875,
+      "mmd2_mixture": 0.12571430106500386,
+      "mmd2_uniform": 0.23862555861405582,
+      "mmd2_single_source": [
+        0.7126149676152926,
+        0.43664360111426287,
+        0.17981946303704704,
+        0.44826934222434894
+      ],
+      "median_squared_distance": 0.17346729616407242,
+      "gammas": [
+        1.4411938476491837,
+        2.8823876952983674,
+        5.764775390596735,
+        11.52955078119347,
+        23.05910156238694
+      ],
+      "max_weight": 0.6777947281235579,
+      "entropy": 0.9605236371075117,
+      "effective_sources": 2.6130644127307936,
+      "min_weight": 0.055262541562243715
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.6491305090540279,
+        0.5572005449783343,
+        0.3890693813540105,
+        0.571988659588714
+      ],
+      "standardized": [
+        1.1309430868146413,
+        0.16184892180501523,
+        -1.610532164849744,
+        0.31774015623008967
+      ],
+      "tau": 0.5,
+      "max_weight": 0.9486104893290658,
+      "entropy": 0.24881632938562975,
+      "effective_sources": 1.2825064526877779,
+      "min_weight": 0.00394341763649415
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "drosophila-vnc",
+    "seconds": 1.4
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.194080938472025,
+      0.1369313835617696,
+      0.3485636136721979,
+      0.15175646072001214
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.34088805889447826,
+      "mixture@raw": 0.34615631820143733,
+      "mixture@logit": 0.32096912740211536,
+      "per_source": 0.34760137972472194,
+      "per_source@raw": 0.3482221938843253,
+      "per_source@logit": 0.34643796682361483,
+      "uniform": 0.1882329719806286,
+      "uniform@raw": 0.18328511906937095,
+      "uniform@logit": 0.20666689954871842
+    },
+    "val_patches": 1024
+  },
+  "temperatures": [
+    1.3822230100631714,
+    3.80090069770813,
+    2.767641305923462,
+    2.597146987915039
+  ],
+  "top_source": "drosophila-vnc",
+  "weight_stats": {
+    "max_weight": 0.6777947281235579,
+    "entropy": 0.9605236371075117,
+    "effective_sources": 2.6130644127307936,
+    "min_weight": 0.055262541562243715
+  },
+  "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": 1011.4,
+  "round0_val": {
+    "val_dice": 0.22230521027783137,
+    "val_vi": 0.15021869142723604,
+    "val_ece": 0.024154192170000647
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 100,
+      "selection_seconds": 83.3,
+      "round_seconds": 683.6,
+      "teacher_dice_labeled": [
+        0.03781776614152498,
+        0.008631654211325996,
+        0.20429526102680404,
+        0.002462217997419853
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 277958,
+      "evidence_nonempty_patches": 81,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.037854588964460804,
+      "restart_teacher_dice": 0.20429526102680404,
+      "weights": [
+        0.014425496757348876,
+        0.027887828907199822,
+        0.9349764743833905,
+        0.022710199952060868
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.20296856291238885,
+        "uniform": 0.003791982665222102,
+        "top1": 0.20429526102680404,
+        "top2": 0.20459987978743366,
+        "top3": 0.20364223129632752
+      },
+      "val_dice": 0.7304906483112401,
+      "val_vi": 0.09860420800166039,
+      "val_ece": 0.006122067360769768
+    },
+    {
+      "round": 2,
+      "n_labeled": 200,
+      "selection_seconds": 42.0,
+      "round_seconds": 214.9,
+      "teacher_dice_labeled": [
+        0.07584090652827795,
+        0.03299669818917039,
+        0.13684430388898855,
+        0.0081834642784292
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1027506,
+      "evidence_nonempty_patches": 178,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0378790588809147,
+        0.06388027088587322,
+        0.8550684864561654,
+        0.04317218377704667
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.1345944737548064,
+        "uniform": 0.01422790922259778,
+        "top1": 0.13684430388898855,
+        "top2": 0.13741123518504308,
+        "top3": 0.13573152285302534
+      },
+      "pseudo_student_dice": 0.5900695150463006,
+      "pseudo_ensemble_dice": 0.10871404767259664,
+      "pseudo_source": "student",
+      "consistency_scale": 0.3,
+      "val_dice": 0.8367504873013976,
+      "val_vi": 0.0604071687246516,
+      "val_ece": 0.0018813678527083225
+    },
+    {
+      "round": 3,
+      "n_labeled": 300,
+      "selection_seconds": 41.9,
+      "round_seconds": 443.8,
+      "teacher_dice_labeled": [
+        0.09410491162631666,
+        0.019753152250683706,
+        0.21479288222850765,
+        0.004277525336689255
+      ],
+      "teacher_dice_pool": [
+        0.11583249102758529,
+        0.014004986028802935,
+        0.26411840031088085,
+        0.0010927551573748757
+      ],
+      "evidence_fg_pixels": 1431921,
+      "evidence_nonempty_patches": 269,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.02265206800706618,
+        0.027876604279621704,
+        0.928787550559032,
+        0.020683777154280256
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.21362074475237053,
+        "uniform": 0.013180744674238559,
+        "top1": 0.21479288222850765,
+        "top2": 0.21514335117603156,
+        "top3": 0.21427675306672375
+      },
+      "pseudo_student_dice": 0.8157384611809655,
+      "pseudo_ensemble_dice": 0.23191085602224795,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8474988516944891,
+      "val_vi": 0.052352724445911526,
+      "val_ece": 0.0033775932266113576
+    },
+    {
+      "round": 4,
+      "n_labeled": 400,
+      "selection_seconds": 42.0,
+      "round_seconds": 516.3,
+      "teacher_dice_labeled": [
+        0.0905930387266031,
+        0.017877815086345897,
+        0.21396067424212706,
+        0.004024708778821875
+      ],
+      "teacher_dice_pool": [
+        0.11028078012272939,
+        0.013610624233042259,
+        0.2721740555641988,
+        0.001608773147522873
+      ],
+      "evidence_fg_pixels": 1887583,
+      "evidence_nonempty_patches": 362,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.022070079558340224,
+        0.027608530821177382,
+        0.9295014342696468,
+        0.02081995535083564
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.21284120228465636,
+        "uniform": 0.012371191155121971,
+        "top1": 0.21396067424212706,
+        "top2": 0.21427996967549467,
+        "top3": 0.21346076574775963
+      },
+      "pseudo_student_dice": 0.820048584988824,
+      "pseudo_ensemble_dice": 0.12350500020243735,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8401894013194965,
+      "val_vi": 0.05195238749177214,
+      "val_ece": 0.002747599314205605
+    },
+    {
+      "round": 5,
+      "n_labeled": 500,
+      "selection_seconds": 42.0,
+      "round_seconds": 589.6,
+      "teacher_dice_labeled": [
+        0.09141055023752703,
+        0.017051576100546896,
+        0.22105584207629211,
+        0.004550119226199254
+      ],
+      "teacher_dice_pool": [
+        0.11144896610606583,
+        0.013122411638998625,
+        0.2900320636792412,
+        0.0014239416017628541
+      ],
+      "evidence_fg_pixels": 2417674,
+      "evidence_nonempty_patches": 461,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.020826516930618596,
+        0.025628153924054202,
+        0.9339558016859595,
+        0.019589527459367744
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.21996896223701284,
+        "uniform": 0.013380655892565459,
+        "top1": 0.22105584207629211,
+        "top2": 0.22140687033401255,
+        "top3": 0.22059841909763112
+      },
+      "pseudo_student_dice": 0.8062622159868228,
+      "pseudo_ensemble_dice": 0.13791089170693827,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8603952180050969,
+      "val_vi": 0.05084424495510591,
+      "val_ece": 0.002686674098639422
+    },
+    {
+      "round": 6,
+      "n_labeled": 600,
+      "selection_seconds": 42.0,
+      "round_seconds": 650.0,
+      "teacher_dice_labeled": [
+        0.0929883618669614,
+        0.0183669711352127,
+        0.2156353693180218,
+        0.004857743495817465
+      ],
+      "teacher_dice_pool": [
+        0.115820433241756,
+        0.015135867659614768,
+        0.27208450552603364,
+        0.0028672229574325868
+      ],
+      "evidence_fg_pixels": 2411238,
+      "evidence_nonempty_patches": 471,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.022237263579107578,
+        0.02729243992452504,
+        0.9298178119198955,
+        0.02065248457647184
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.21446330415652365,
+        "uniform": 0.013375940114072303,
+        "top1": 0.2156353693180218,
+        "top2": 0.2159996588536014,
+        "top3": 0.21513142769519916
+      },
+      "pseudo_student_dice": 0.8300632758147568,
+      "pseudo_ensemble_dice": 0.17062110708034692,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8607512127522224,
+      "val_vi": 0.05201825374881291,
+      "val_ece": 0.0029437962306920057
+    },
+    {
+      "round": 7,
+      "n_labeled": 600,
+      "selection_seconds": 0.0,
+      "round_seconds": 721.2,
+      "final_consolidation": true,
+      "val_dice": 0.8854431193714826,
+      "val_vi": 0.0413108489859979,
+      "val_ece": 0.0028508722852614143
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": -0.3458863863149128,
+    "auroc_top20_error": 0.36265166015625
+  },
+  "al_seconds": 3860.3,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.022237263579107578,
+    0.02729243992452504,
+    0.9298178119198955,
+    0.02065248457647184
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9298178119198955,
+    "entropy": 0.3307084327822969,
+    "effective_sources": 1.3919538850150148,
+    "min_weight": 0.02065248457647184
+  },
+  "top_source_final": "drosophila-vnc",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.8665314979226896,
+    "patch_iou": 0.8374277072784165,
+    "patch_precision": 0.9208533164412269,
+    "patch_recall": 0.9051563822798545,
+    "patch_vi": 0.027594797590052678,
+    "n_patches": 2376,
+    "cal_ece": 0.0021656661630985764,
+    "cal_nll": 0.01375716441216002,
+    "cal_brier": 0.002850755817302684,
+    "cal_mean_confidence": 0.9987719470711373,
+    "cal_accuracy": 0.9966062809080387,
+    "cal_predicted_prevalence": 0.027316623263888888,
+    "cal_true_prevalence": 0.026826286958122895,
+    "cal_mean_probability": 0.027120532269459516
+  },
+  "test_notta": {
+    "patch_dice": 0.8665314979226896,
+    "patch_iou": 0.8374277072784165,
+    "patch_precision": 0.9208533164412269,
+    "patch_recall": 0.9051563822798545,
+    "patch_vi": 0.027594797590052678,
+    "n_patches": 2376,
+    "cal_ece": 0.0021656661630985764,
+    "cal_nll": 0.01375716441216002,
+    "cal_brier": 0.002850755817302684,
+    "cal_mean_confidence": 0.9987719470711373,
+    "cal_accuracy": 0.9966062809080387,
+    "cal_predicted_prevalence": 0.027316623263888888,
+    "cal_true_prevalence": 0.026826286958122895,
+    "cal_mean_probability": 0.027120532269459516
+  },
+  "test": {
+    "patch_dice": 0.875913909393184,
+    "patch_iou": 0.8474997327678472,
+    "patch_precision": 0.9330206901599026,
+    "patch_recall": 0.904096587521667,
+    "patch_vi": 0.026343271704068902,
+    "n_patches": 2376,
+    "image_dice": 0.9468178241085186,
+    "image_iou": 0.8995131509390805,
+    "image_precision": 0.9460566273553056,
+    "image_recall": 0.9480638794469338,
+    "image_vi": 0.03204408346277166,
+    "n_images": 24,
+    "cal_ece": 0.0016236001779007755,
+    "cal_nll": 0.011411511158590052,
+    "cal_brier": 0.002567256205654599,
+    "cal_mean_confidence": 0.9984630915407051,
+    "cal_accuracy": 0.9968504215330387,
+    "cal_predicted_prevalence": 0.027095499263468013,
+    "cal_true_prevalence": 0.026826286958122895,
+    "cal_mean_probability": 0.02698464092476411
+  },
+  "eval_seconds": 124.2,
+  "total_seconds": 5052.7,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 0.75,
+        "peak_cpu_rss_mb": 2191.9,
+        "peak_gpu_alloc_mb": 592.4,
+        "peak_gpu_reserved_mb": 594.0
+      },
+      "features_and_weights": {
+        "seconds": 48.62,
+        "peak_cpu_rss_mb": 4172.5,
+        "peak_gpu_alloc_mb": 2298.2,
+        "peak_gpu_reserved_mb": 2640.0
+      },
+      "distillation": {
+        "seconds": 1011.42,
+        "peak_cpu_rss_mb": 5948.1,
+        "peak_gpu_alloc_mb": 3826.7,
+        "peak_gpu_reserved_mb": 4128.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 3860.3,
+        "peak_cpu_rss_mb": 5948.1,
+        "peak_gpu_alloc_mb": 3947.0,
+        "peak_gpu_reserved_mb": 4170.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 124.22,
+        "peak_cpu_rss_mb": 5948.1,
+        "peak_gpu_alloc_mb": 2149.5,
+        "peak_gpu_reserved_mb": 4172.0,
+        "patches": 2376,
+        "images": 24,
+        "patches_per_second": 19.1,
+        "seconds_per_image": 5.176
+      }
+    },
+    "peak_gpu_alloc_mb": 3947.0,
+    "peak_cpu_rss_mb": 5948.1,
+    "total_seconds": 5045.3
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/kasthuri-pp/macs-v11-norestart-acq-entropy_b5_s1.pth"
+}

macs_v2_r26/runs/mitochondria/kasthuri-pp/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
+1,100,83.3,683.6,"[0.03781776614152498, 0.008631654211325996, 0.20429526102680404, 0.002462217997419853]",,277958,81,True,kept,0.037854588964460804,0.20429526102680404,"[0.014425496757348876, 0.027887828907199822, 0.9349764743833905, 0.022710199952060868]",mixture,"{'mixture': 0.20296856291238885, 'uniform': 0.003791982665222102, 'top1': 0.20429526102680404, 'top2': 0.20459987978743366, 'top3': 0.20364223129632752}",0.7304906483112401,0.09860420800166039,0.006122067360769768,,,,,
+2,200,42.0,214.9,"[0.07584090652827795, 0.03299669818917039, 0.13684430388898855, 0.0081834642784292]",,1027506,178,False,,,,"[0.0378790588809147, 0.06388027088587322, 0.8550684864561654, 0.04317218377704667]",mixture,"{'mixture': 0.1345944737548064, 'uniform': 0.01422790922259778, 'top1': 0.13684430388898855, 'top2': 0.13741123518504308, 'top3': 0.13573152285302534}",0.8367504873013976,0.0604071687246516,0.0018813678527083225,0.5900695150463006,0.10871404767259664,student,0.3,
+3,300,41.9,443.8,"[0.09410491162631666, 0.019753152250683706, 0.21479288222850765, 0.004277525336689255]","[0.11583249102758529, 0.014004986028802935, 0.26411840031088085, 0.0010927551573748757]",1431921,269,False,,,,"[0.02265206800706618, 0.027876604279621704, 0.928787550559032, 0.020683777154280256]",mixture,"{'mixture': 0.21362074475237053, 'uniform': 0.013180744674238559, 'top1': 0.21479288222850765, 'top2': 0.21514335117603156, 'top3': 0.21427675306672375}",0.8474988516944891,0.052352724445911526,0.0033775932266113576,0.8157384611809655,0.23191085602224795,student,1.0,
+4,400,42.0,516.3,"[0.0905930387266031, 0.017877815086345897, 0.21396067424212706, 0.004024708778821875]","[0.11028078012272939, 0.013610624233042259, 0.2721740555641988, 0.001608773147522873]",1887583,362,False,,,,"[0.022070079558340224, 0.027608530821177382, 0.9295014342696468, 0.02081995535083564]",mixture,"{'mixture': 0.21284120228465636, 'uniform': 0.012371191155121971, 'top1': 0.21396067424212706, 'top2': 0.21427996967549467, 'top3': 0.21346076574775963}",0.8401894013194965,0.05195238749177214,0.002747599314205605,0.820048584988824,0.12350500020243735,student,1.0,
+5,500,42.0,589.6,"[0.09141055023752703, 0.017051576100546896, 0.22105584207629211, 0.004550119226199254]","[0.11144896610606583, 0.013122411638998625, 0.2900320636792412, 0.0014239416017628541]",2417674,461,False,,,,"[0.020826516930618596, 0.025628153924054202, 0.9339558016859595, 0.019589527459367744]",mixture,"{'mixture': 0.21996896223701284, 'uniform': 0.013380655892565459, 'top1': 0.22105584207629211, 'top2': 0.22140687033401255, 'top3': 0.22059841909763112}",0.8603952180050969,0.05084424495510591,0.002686674098639422,0.8062622159868228,0.13791089170693827,student,1.0,
+6,600,42.0,650.0,"[0.0929883618669614, 0.0183669711352127, 0.2156353693180218, 0.004857743495817465]","[0.115820433241756, 0.015135867659614768, 0.27208450552603364, 0.0028672229574325868]",2411238,471,False,,,,"[0.022237263579107578, 0.02729243992452504, 0.9298178119198955, 0.02065248457647184]",mixture,"{'mixture': 0.21446330415652365, 'uniform': 0.013375940114072303, 'top1': 0.2156353693180218, 'top2': 0.2159996588536014, 'top3': 0.21513142769519916}",0.8607512127522224,0.05201825374881291,0.0029437962306920057,0.8300632758147568,0.17062110708034692,student,1.0,
+7,600,0.0,721.2,,,,,,,,,,,,0.8854431193714826,0.0413108489859979,0.0028508722852614143,,,,,True

macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-laplace/budget-5/seed-1/log.txt

@@ -0,0 +1,148 @@
+[23:30:06][      0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-laplace budget=5.0% seed=1 | {"task": "mitochondria", "target": "lucchi-pp", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 8, "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}
+[23:30:06][      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.
+[23:30:06][      0s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches
+[23:30:08][      2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60
+[23:30:08][      2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s
+[23:30:09][      3s][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
+[23:30:09][      3s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass
+[23:30:47][     41s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833
+[23:30:47][     41s][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
+[23:30:47][     41s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920
+[23:30:47][     41s][INIT] student initialised from teacher multiclass (weight 0.920)
+[23:30:50][     44s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[23:30:50][     44s][KD] distilling the mixture ensemble (prob space) into the student
+[23:32:54][    168s][KD] soft labels for 6000 patches from 4 teachers in 123s
+[23:35:54][    348s][KD] epoch 1/5 | patches=6000 | loss=0.1185 | 180s
+[23:38:54][    528s][KD] epoch 2/5 | patches=6000 | loss=0.1064 | 181s
+[23:41:55][    709s][KD] epoch 3/5 | patches=6000 | loss=0.1046 | 181s
+[23:44:55][    889s][KD] epoch 4/5 | patches=6000 | loss=0.1035 | 180s
+[23:47:55][   1069s][KD] epoch 5/5 | patches=6000 | loss=0.1028 | 180s
+[23:48:02][   1075s][ROUND0] before any target label: val_dice=0.8843 val_vi=0.0624
+[23:48:02][   1075s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=laplace | first_round=stratified
+[23:49:29][   1163s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.0892883911765001, 'auroc_top20_error': 0.473226806640625}
+[23:50:11][   1205s][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
+[23:50:17][   1211s][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
+[23:50:18][   1212s][WEIGHTS] round 1: student kept (labeled Dice 0.940 vs best teacher multiclass 0.932); distilling the corrected ensemble into it
+[23:50:18][   1212s][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
+[23:52:22][   1336s][KD] soft labels for 6000 patches from 4 teachers in 124s
+[23:55:22][   1516s][KD] epoch 1/2 | patches=6000 | loss=0.0467 | 180s
+[23:58:23][   1697s][KD] epoch 2/2 | patches=6000 | loss=0.0462 | 181s
+[23:59:04][   1738s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 42s
+[23:59:09][   1743s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3159
+[23:59:14][   1748s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3304
+[23:59:19][   1753s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3587
+[23:59:24][   1758s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.3334
+[23:59:29][   1763s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.2772
+[23:59:35][   1769s][AL] round 1/6 | labeled=67 | selection_seconds=129.3000 round_seconds=687.5000 restart_student_dice=0.9401 restart_teacher_dice=0.9318 val_dice=0.8065 val_vi=0.0844 val_ece=0.0065
+[00:01:02][   1856s][AL] round 2: 1197/7983 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[00:01:13][   1867s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.840, uniform=0.218, top1=0.840, top2=0.840, top3=0.840 -> using mixture
+[00:01:13][   1867s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (778080 fg px): fly-mito=0.289, kasthuri=0.000, dros-vnc=0.012, multiclass=0.840 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[00:01:18][   1871s][CONS] round 2: on the 67 new patches student Dice=0.770 vs ensemble 0.763 -> pseudo-labels from the ensemble | consistency x1.00
+[00:01:26][   1880s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.4674
+[00:01:35][   1889s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.3997
+[00:01:44][   1898s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.3579
+[00:01:53][   1907s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.3644
+[00:02:02][   1915s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.3159
+[00:04:10][      0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-laplace budget=5.0% seed=1 | {"task": "mitochondria", "target": "lucchi-pp", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 8, "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}
+[00:04:10][      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.
+[00:04:15][      5s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches
+[00:04:26][     16s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60
+[00:04:36][     26s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 9s
+[00:04:37][     27s][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
+[00:04:37][     27s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass
+[00:05:16][     66s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833
+[00:05:16][     66s][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
+[00:05:16][     66s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920
+[00:05:16][     66s][INIT] student initialised from teacher multiclass (weight 0.920)
+[00:05:27][     76s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[00:05:27][     76s][KD] distilling the mixture ensemble (prob space) into the student
+[00:07:29][    199s][KD] soft labels for 6000 patches from 4 teachers in 123s
+[00:10:29][    379s][KD] epoch 1/5 | patches=6000 | loss=0.1185 | 179s
+[00:13:31][    561s][KD] epoch 2/5 | patches=6000 | loss=0.1064 | 182s
+[00:16:30][    740s][KD] epoch 3/5 | patches=6000 | loss=0.1043 | 179s
+[00:19:29][    918s][KD] epoch 4/5 | patches=6000 | loss=0.1035 | 179s
+[00:22:28][   1097s][KD] epoch 5/5 | patches=6000 | loss=0.1031 | 179s
+[00:22:33][   1103s][ROUND0] before any target label: val_dice=0.8827 val_vi=0.0617
+[00:22:33][   1103s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=laplace | first_round=stratified
+[00:24:00][   1190s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.08900566294125338, 'auroc_top20_error': 0.471344873046875}
+[00:24:42][   1232s][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
+[00:24:47][   1237s][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
+[00:24:48][   1238s][WEIGHTS] round 1: student kept (labeled Dice 0.940 vs best teacher multiclass 0.932); distilling the corrected ensemble into it
+[00:24:48][   1238s][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
+[00:26:51][   1361s][KD] soft labels for 6000 patches from 4 teachers in 123s
+[00:29:50][   1540s][KD] epoch 1/2 | patches=6000 | loss=0.0466 | 179s
+[00:32:49][   1719s][KD] epoch 2/2 | patches=6000 | loss=0.0459 | 179s
+[00:33:30][   1760s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 41s
+[00:33:35][   1765s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3159
+[00:33:39][   1769s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3248
+[00:33:44][   1774s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3628
+[00:33:48][   1778s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.3215
+[00:33:53][   1783s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.2480
+[00:33:59][   1789s][AL] round 1/6 | labeled=67 | selection_seconds=128.5000 round_seconds=679.8000 restart_student_dice=0.9398 restart_teacher_dice=0.9318 val_dice=0.8461 val_vi=0.0856 val_ece=0.0067
+[00:35:25][   1875s][AL] round 2: 1183/7983 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[00:35:36][   1886s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.819, uniform=0.253, top1=0.819, top2=0.819, top3=0.819 -> using mixture
+[00:35:36][   1886s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (650845 fg px): fly-mito=0.246, kasthuri=0.000, dros-vnc=0.024, multiclass=0.819 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[00:35:40][   1889s][CONS] round 2: on the 67 new patches student Dice=0.680 vs ensemble 0.689 -> pseudo-labels from the ensemble | consistency x1.00
+[00:35:48][   1898s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.5413
+[00:35:56][   1906s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.3978
+[00:36:05][   1914s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.3414
+[00:36:13][   1923s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.3627
+[00:36:22][   1931s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.3053
+[00:36:27][   1937s][AL] round 2/6 | labeled=134 | selection_seconds=86.6000 round_seconds=142.8000 pseudo_student_dice=0.6799 pseudo_ensemble_dice=0.6885 consistency_scale=1.0000 val_dice=0.8239 val_vi=0.0783 val_ece=0.0062
+[00:37:53][   2023s][AL] round 3: 1286/7916 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[00:38:08][   2038s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.826, uniform=0.199, top1=0.826, top2=0.826, top3=0.826 -> using mixture
+[00:38:08][   2038s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (1077649 fg px): fly-mito=0.271, kasthuri=0.000, dros-vnc=0.016, multiclass=0.826 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[00:38:12][   2042s][CONS] round 3: on the 67 new patches student Dice=0.843 vs ensemble 0.836 -> pseudo-labels from the ensemble | consistency x1.00
+[00:38:25][   2054s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.3340
+[00:38:37][   2067s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.3026
+[00:38:49][   2079s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.2820
+[00:39:02][   2092s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.2937
+[00:39:14][   2104s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2591
+[00:39:20][   2110s][AL] round 3/6 | labeled=201 | selection_seconds=85.8000 round_seconds=167.1000 pseudo_student_dice=0.8429 pseudo_ensemble_dice=0.8358 consistency_scale=1.0000 val_dice=0.8594 val_vi=0.0649 val_ece=0.0037
+[00:40:45][   2195s][AL] round 4: 1380/7849 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[00:41:05][   2215s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.841, uniform=0.192, top1=0.841, top2=0.841, top3=0.841 -> using mixture
+[00:41:05][   2215s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1455704 fg px): fly-mito=0.281, kasthuri=0.001, dros-vnc=0.023, multiclass=0.841 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[00:41:09][   2219s][CONS] round 4: on the 67 new patches student Dice=0.865 vs ensemble 0.884 -> pseudo-labels from the ensemble | consistency x1.00
+[00:41:25][   2235s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.2903
+[00:41:41][   2251s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.3076
+[00:41:58][   2267s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.2736
+[00:42:14][   2284s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.2825
+[00:42:30][   2300s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.2638
+[00:42:36][   2306s][AL] round 4/6 | labeled=268 | selection_seconds=84.9000 round_seconds=190.2000 pseudo_student_dice=0.8653 pseudo_ensemble_dice=0.8842 consistency_scale=1.0000 val_dice=0.8642 val_vi=0.0689 val_ece=0.0044
+[00:44:00][   2390s][AL] round 5: 1217/7782 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[00:44:25][   2415s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.849, uniform=0.205, top1=0.849, top2=0.849, top3=0.849 -> using mixture
+[00:44:25][   2415s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (1867412 fg px): fly-mito=0.284, kasthuri=0.001, dros-vnc=0.023, multiclass=0.849 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[00:44:29][   2419s][CONS] round 5: on the 67 new patches student Dice=0.875 vs ensemble 0.877 -> pseudo-labels from the ensemble | consistency x1.00
+[00:44:49][   2439s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2433
+[00:45:09][   2459s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.2352
+[00:45:29][   2479s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.2319
+[00:45:49][   2499s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.2302
+[00:46:10][   2519s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.2170
+[00:46:15][   2525s][AL] round 5/6 | labeled=335 | selection_seconds=84.2000 round_seconds=213.6000 pseudo_student_dice=0.8754 pseudo_ensemble_dice=0.8775 consistency_scale=1.0000 val_dice=0.8630 val_vi=0.0682 val_ece=0.0053
+[00:47:39][   2608s][AL] round 6: 1218/7715 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[00:48:09][   2638s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.851, uniform=0.205, top1=0.851, top2=0.851, top3=0.851 -> using mixture
+[00:48:09][   2638s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (2210520 fg px): fly-mito=0.279, kasthuri=0.000, dros-vnc=0.020, multiclass=0.851 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[00:48:13][   2642s][CONS] round 6: on the 67 new patches student Dice=0.864 vs ensemble 0.862 -> pseudo-labels from the ensemble | consistency x1.00
+[00:48:37][   2666s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2212
+[00:49:01][   2691s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.2089
+[00:49:25][   2715s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.1954
+[00:49:49][   2739s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.2278
+[00:50:14][   2763s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.2029
+[00:50:19][   2769s][AL] round 6/6 | labeled=402 | selection_seconds=83.5000 round_seconds=238.4000 pseudo_student_dice=0.8636 pseudo_ensemble_dice=0.8622 consistency_scale=1.0000 val_dice=0.8705 val_vi=0.0671 val_ece=0.0055
+[00:50:44][   2793s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.1935
+[00:51:08][   2818s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.1710
+[00:51:32][   2842s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.1776
+[00:51:56][   2866s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.1640
+[00:52:20][   2890s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.1549
+[00:52:45][   2915s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1496
+[00:53:09][   2939s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1495
+[00:53:33][   2963s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1429
+[00:53:58][   2987s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1361
+[00:54:22][   3012s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1372
+[00:54:28][   3017s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=242.4000 val_dice=0.8826 val_vi=0.0611 val_ece=0.0041
+[00:54:28][   3017s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[00:55:50][   3099s][TEST] raw student (no EMA, no TTA): patch_dice=0.9136 patch_vi=0.0429
+[00:55:59][   3109s][TEST] final model without TTA: patch_dice=0.9136 patch_vi=0.0429
+[00:55:59][   3109s][RESOURCES] {"phases": {"setup": {"seconds": 8.23, "peak_cpu_rss_mb": 2169.5, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 60.15, "peak_cpu_rss_mb": 4165.8, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 1020.9, "peak_cpu_rss_mb": 5938.9, "peak_gpu_alloc_mb": 3826.7, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 1914.27, "peak_cpu_rss_mb": 5938.9, "peak_gpu_alloc_mb": 3947.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 91.59, "peak_cpu_rss_mb": 5938.9, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 19.1, "seconds_per_image": 1.832}}, "peak_gpu_alloc_mb": 3947.0, "peak_cpu_rss_mb": 5938.9, "total_seconds": 3095.1}
+[00:55:59][   3109s][TEST] patch_dice=0.9262 patch_iou=0.8974 patch_precision=0.9610 patch_recall=0.9315 patch_vi=0.0406 image_dice=0.9555 image_iou=0.9153 image_precision=0.9706 image_recall=0.9415 image_vi=0.0454 cal_ece=0.0018 cal_nll=0.0168 cal_brier=0.0035 cal_mean_confidence=0.9974 cal_accuracy=0.9956 cal_predicted_prevalence=0.0439 cal_true_prevalence=0.0452 cal_mean_probability=0.0442
+[00:56:00][   3110s][DONE] lucchi-pp macs-v11-norestart-acq-laplace b5 s1 | test patch Dice=0.9262 VI=0.0406 | image Dice=0.9555 VI=0.0454 | 3109s

macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-laplace/budget-5/seed-1/result.json

@@ -0,0 +1,647 @@
+{
+  "target": "lucchi-pp",
+  "variant": "macs-v11-norestart-acq-laplace",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "mitochondria",
+    "target": "lucchi-pp",
+    "variant": "macs-v11-norestart-acq-laplace",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "fce9b15",
+  "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "lucchi-pp",
+      "--variants",
+      "macs-v11-norestart-acq-laplace",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--task",
+      "mitochondria",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "commit": "fce9b15",
+    "dirty": true
+  },
+  "sources": [
+    "fly-brain-mito-benchmark",
+    "kasthuri-pp",
+    "drosophila-vnc",
+    "multiclass"
+  ],
+  "alphas": [
+    0.03256907457032885,
+    0.031801008835420135,
+    0.015561893577512515,
+    0.9200680230167385
+  ],
+  "weights_all": {
+    "sources": [
+      "fly-brain-mito-benchmark",
+      "kasthuri-pp",
+      "drosophila-vnc",
+      "multiclass"
+    ],
+    "mixture": [
+      0.03256907457032885,
+      0.031801008835420135,
+      0.015561893577512515,
+      0.9200680230167385
+    ],
+    "per_source": [
+      0.013021514318439549,
+      0.021594227384204207,
+      0.00495364759466876,
+      0.9604306107026874
+    ],
+    "uniform": [
+      0.25,
+      0.25,
+      0.25,
+      0.25
+    ],
+    "mixture_info": {
+      "objective": 0.001413505929905978,
+      "mmd2_mixture": 0.005039697348754646,
+      "mmd2_uniform": 0.2092228336519747,
+      "mmd2_single_source": [
+        0.5680513520444281,
+        0.4644148125694437,
+        0.6002511462494531,
+        0.004698715465898062
+      ],
+      "median_squared_distance": 0.1892747764462741,
+      "gammas": [
+        1.3208310409546982,
+        2.6416620819093963,
+        5.283324163818793,
+        10.566648327637585,
+        21.13329665527517
+      ],
+      "max_weight": 0.9200680230167385,
+      "entropy": 0.36261914188486677,
+      "effective_sources": 1.4370884282110035,
+      "min_weight": 0.015561893577512515
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.6675354526109456,
+        0.5899267604637966,
+        0.8158227740743191,
+        0.007664790896468099
+      ],
+      "standardized": [
+        0.4800153933276398,
+        0.22710384776568787,
+        0.9632547653742088,
+        -1.670374006467536
+      ],
+      "tau": 0.5,
+      "max_weight": 0.9604306107026874,
+      "entropy": 0.2044174691373809,
+      "effective_sources": 1.2268102019632934,
+      "min_weight": 0.00495364759466876
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "multiclass",
+    "seconds": 10.6
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.24317352702666994,
+      0.16070093323689605,
+      0.16243146338599246,
+      0.8333945099982429
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.8341164139370107,
+      "mixture@raw": 0.8334634452984915,
+      "mixture@logit": 0.8348661307075531,
+      "per_source": 0.8335017830355866,
+      "per_source@raw": 0.8334355010464378,
+      "per_source@logit": 0.8337681001665089,
+      "uniform": 0.3312974448451058,
+      "uniform@raw": 0.18374038775198187,
+      "uniform@logit": 0.21963766526924644
+    },
+    "val_patches": 1024
+  },
+  "temperatures": [
+    1.3822230100631714,
+    2.5090832710266113,
+    2.767641305923462,
+    2.597146987915039
+  ],
+  "top_source": "multiclass",
+  "weight_stats": {
+    "max_weight": 0.9200680230167385,
+    "entropy": 0.36261914188486677,
+    "effective_sources": 1.4370884282110035,
+    "min_weight": 0.015561893577512515
+  },
+  "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": 1020.9,
+  "round0_val": {
+    "val_dice": 0.8827209277665129,
+    "val_vi": 0.06173844436601252,
+    "val_ece": 0.01930872331093795
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 67,
+      "selection_seconds": 128.5,
+      "round_seconds": 679.8,
+      "teacher_dice_labeled": [
+        0.2876389731008006,
+        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": 0.9398350397937785,
+      "restart_teacher_dice": 0.931847537434372,
+      "weights": [
+        5.6389282479936555e-05,
+        3.1020171205931166e-06,
+        1.907901148148134e-06,
+        0.9999386007992513
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.931847537434372,
+        "uniform": 0.3539312329080638,
+        "top1": 0.931847537434372,
+        "top2": 0.931847537434372,
+        "top3": 0.931847537434372
+      },
+      "val_dice": 0.8461434440443407,
+      "val_vi": 0.0855945925313224,
+      "val_ece": 0.006651857957124172
+    },
+    {
+      "round": 2,
+      "n_labeled": 134,
+      "selection_seconds": 86.6,
+      "round_seconds": 142.8,
+      "teacher_dice_labeled": [
+        0.24577679079476183,
+        1.5364617743675156e-06,
+        0.024486936416884684,
+        0.8192993927643852
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 650845,
+      "evidence_nonempty_patches": 122,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.00011432785797566688,
+        9.558707631533626e-06,
+        5.975296532053933e-06,
+        0.9998701381378607
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.8193014677864318,
+        "uniform": 0.2526532619810576,
+        "top1": 0.8192993927643852,
+        "top2": 0.8193014677864318,
+        "top3": 0.8193014677864318
+      },
+      "pseudo_student_dice": 0.6799073601500278,
+      "pseudo_ensemble_dice": 0.6885056543210284,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.823862031483791,
+      "val_vi": 0.07833448240855631,
+      "val_ece": 0.006199167541584895
+    },
+    {
+      "round": 3,
+      "n_labeled": 201,
+      "selection_seconds": 85.8,
+      "round_seconds": 167.1,
+      "teacher_dice_labeled": [
+        0.27051968258106773,
+        9.271819140187124e-07,
+        0.015514227110664483,
+        0.8257373465675559
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1077649,
+      "evidence_nonempty_patches": 186,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.00013729036451304996,
+        8.962465766820798e-06,
+        5.1217981496385855e-06,
+        0.9998486253715705
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.8257374489433349,
+        "uniform": 0.19946310223893535,
+        "top1": 0.8257373465675559,
+        "top2": 0.8257366798750658,
+        "top3": 0.8257374489433349
+      },
+      "pseudo_student_dice": 0.8428575797424998,
+      "pseudo_ensemble_dice": 0.8358479338209674,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.859391961227498,
+      "val_vi": 0.06487842490840803,
+      "val_ece": 0.003686388289243956
+    },
+    {
+      "round": 4,
+      "n_labeled": 268,
+      "selection_seconds": 84.9,
+      "round_seconds": 190.2,
+      "teacher_dice_labeled": [
+        0.2809361991370772,
+        0.0006554222291767896,
+        0.023371464715406718,
+        0.8408604006205874
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1455704,
+      "evidence_nonempty_patches": 246,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0001309794754778175,
+        7.755231705957663e-06,
+        4.7628931689371864e-06,
+        0.9998565023996473
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.8408612401217148,
+        "uniform": 0.19158436581242347,
+        "top1": 0.8408604006205874,
+        "top2": 0.8408612401217148,
+        "top3": 0.8408612401217148
+      },
+      "pseudo_student_dice": 0.865318273934081,
+      "pseudo_ensemble_dice": 0.8841677834896886,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.864204841298776,
+      "val_vi": 0.0689438616669614,
+      "val_ece": 0.004439818108043122
+    },
+    {
+      "round": 5,
+      "n_labeled": 335,
+      "selection_seconds": 84.2,
+      "round_seconds": 213.6,
+      "teacher_dice_labeled": [
+        0.2835080031881515,
+        0.0005110274916737301,
+        0.02308174871596876,
+        0.8489501264344202
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1867412,
+      "evidence_nonempty_patches": 310,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.00012394890757176728,
+        7.142297450423627e-06,
+        4.380088376626105e-06,
+        0.9998645287066013
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.8489512174331817,
+        "uniform": 0.20478316977482933,
+        "top1": 0.8489501264344202,
+        "top2": 0.8489512174331817,
+        "top3": 0.8489512174331817
+      },
+      "pseudo_student_dice": 0.8753591055736171,
+      "pseudo_ensemble_dice": 0.8774800102230823,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.862984788458472,
+      "val_vi": 0.06815227306475158,
+      "val_ece": 0.005252702164533228
+    },
+    {
+      "round": 6,
+      "n_labeled": 402,
+      "selection_seconds": 83.5,
+      "round_seconds": 238.4,
+      "teacher_dice_labeled": [
+        0.27886655669468124,
+        0.0004317570658965266,
+        0.019823472474504847,
+        0.8510156994214092
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 2210520,
+      "evidence_nonempty_patches": 377,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.0001159092629984249,
+        6.990796383176612e-06,
+        4.153032658780695e-06,
+        0.9998729469079597
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.851015583658806,
+        "uniform": 0.20455471975081585,
+        "top1": 0.8510156994214092,
+        "top2": 0.851015583658806,
+        "top3": 0.851015583658806
+      },
+      "pseudo_student_dice": 0.8635595641020113,
+      "pseudo_ensemble_dice": 0.8621746701279371,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8705280669376912,
+      "val_vi": 0.06705494363318357,
+      "val_ece": 0.005491697527176027
+    },
+    {
+      "round": 7,
+      "n_labeled": 402,
+      "selection_seconds": 0.0,
+      "round_seconds": 242.4,
+      "final_consolidation": true,
+      "val_dice": 0.8825650776052844,
+      "val_vi": 0.06109141014936791,
+      "val_ece": 0.004077543849137184
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.08900566294125338,
+    "auroc_top20_error": 0.471344873046875
+  },
+  "al_seconds": 1914.3,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    0.0001159092629984249,
+    6.990796383176612e-06,
+    4.153032658780695e-06,
+    0.9998729469079597
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9998729469079597,
+    "entropy": 0.0013119464075929815,
+    "effective_sources": 1.0013128073857591,
+    "min_weight": 4.153032658780695e-06
+  },
+  "top_source_final": "multiclass",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.91356207848983,
+    "patch_iou": 0.8831404748948558,
+    "patch_precision": 0.946976123955655,
+    "patch_recall": 0.930699016573354,
+    "patch_vi": 0.042884940354461186,
+    "n_patches": 1750,
+    "cal_ece": 0.0026736035516054433,
+    "cal_nll": 0.021422703820900835,
+    "cal_brier": 0.003958336581594274,
+    "cal_mean_confidence": 0.9978627776587483,
+    "cal_accuracy": 0.9951891741071428,
+    "cal_predicted_prevalence": 0.04423939732142857,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.04434355303823744
+  },
+  "test_notta": {
+    "patch_dice": 0.91356207848983,
+    "patch_iou": 0.8831404748948558,
+    "patch_precision": 0.946976123955655,
+    "patch_recall": 0.930699016573354,
+    "patch_vi": 0.042884940354461186,
+    "n_patches": 1750,
+    "cal_ece": 0.0026736035516054433,
+    "cal_nll": 0.021422703820900835,
+    "cal_brier": 0.003958336581594274,
+    "cal_mean_confidence": 0.9978627776587483,
+    "cal_accuracy": 0.9951891741071428,
+    "cal_predicted_prevalence": 0.04423939732142857,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.04434355303823744
+  },
+  "test": {
+    "patch_dice": 0.9261972077942084,
+    "patch_iou": 0.8974155611349277,
+    "patch_precision": 0.9609866432654087,
+    "patch_recall": 0.9315003450236838,
+    "patch_vi": 0.040560742745314356,
+    "n_patches": 1750,
+    "image_dice": 0.9555399724621937,
+    "image_iou": 0.9153154542048801,
+    "image_precision": 0.970564719786427,
+    "image_recall": 0.9415382349126573,
+    "image_vi": 0.04536973324369179,
+    "n_images": 50,
+    "cal_ece": 0.0018369080010505645,
+    "cal_nll": 0.01683470636319407,
+    "cal_brier": 0.0034860976003171663,
+    "cal_mean_confidence": 0.9973964473557894,
+    "cal_accuracy": 0.9955675223214285,
+    "cal_predicted_prevalence": 0.043865513392857144,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.04415598747573566
+  },
+  "eval_seconds": 91.6,
+  "total_seconds": 3108.8,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 8.23,
+        "peak_cpu_rss_mb": 2169.5,
+        "peak_gpu_alloc_mb": 592.4,
+        "peak_gpu_reserved_mb": 594.0
+      },
+      "features_and_weights": {
+        "seconds": 60.15,
+        "peak_cpu_rss_mb": 4165.8,
+        "peak_gpu_alloc_mb": 2298.2,
+        "peak_gpu_reserved_mb": 2640.0
+      },
+      "distillation": {
+        "seconds": 1020.9,
+        "peak_cpu_rss_mb": 5938.9,
+        "peak_gpu_alloc_mb": 3826.7,
+        "peak_gpu_reserved_mb": 4128.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 1914.27,
+        "peak_cpu_rss_mb": 5938.9,
+        "peak_gpu_alloc_mb": 3947.0,
+        "peak_gpu_reserved_mb": 4176.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 91.59,
+        "peak_cpu_rss_mb": 5938.9,
+        "peak_gpu_alloc_mb": 2149.5,
+        "peak_gpu_reserved_mb": 4176.0,
+        "patches": 1750,
+        "images": 50,
+        "patches_per_second": 19.1,
+        "seconds_per_image": 1.832
+      }
+    },
+    "peak_gpu_alloc_mb": 3947.0,
+    "peak_cpu_rss_mb": 5938.9,
+    "total_seconds": 3095.1
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-laplace_b5_s1.pth"
+}

macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-laplace/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
+1,67,128.5,679.8,"[0.2876389731008006, 2.8393119779215102e-06, 0.02286523788150808, 0.931847537434372]",,352197,62,True,kept,0.9398350397937785,0.931847537434372,"[5.6389282479936555e-05, 3.1020171205931166e-06, 1.907901148148134e-06, 0.9999386007992513]",mixture,"{'mixture': 0.931847537434372, 'uniform': 0.3539312329080638, 'top1': 0.931847537434372, 'top2': 0.931847537434372, 'top3': 0.931847537434372}",0.8461434440443407,0.0855945925313224,0.006651857957124172,,,,,
+2,134,86.6,142.8,"[0.24577679079476183, 1.5364617743675156e-06, 0.024486936416884684, 0.8192993927643852]",,650845,122,False,,,,"[0.00011432785797566688, 9.558707631533626e-06, 5.975296532053933e-06, 0.9998701381378607]",mixture,"{'mixture': 0.8193014677864318, 'uniform': 0.2526532619810576, 'top1': 0.8192993927643852, 'top2': 0.8193014677864318, 'top3': 0.8193014677864318}",0.823862031483791,0.07833448240855631,0.006199167541584895,0.6799073601500278,0.6885056543210284,ensemble,1.0,
+3,201,85.8,167.1,"[0.27051968258106773, 9.271819140187124e-07, 0.015514227110664483, 0.8257373465675559]",,1077649,186,False,,,,"[0.00013729036451304996, 8.962465766820798e-06, 5.1217981496385855e-06, 0.9998486253715705]",mixture,"{'mixture': 0.8257374489433349, 'uniform': 0.19946310223893535, 'top1': 0.8257373465675559, 'top2': 0.8257366798750658, 'top3': 0.8257374489433349}",0.859391961227498,0.06487842490840803,0.003686388289243956,0.8428575797424998,0.8358479338209674,ensemble,1.0,
+4,268,84.9,190.2,"[0.2809361991370772, 0.0006554222291767896, 0.023371464715406718, 0.8408604006205874]",,1455704,246,False,,,,"[0.0001309794754778175, 7.755231705957663e-06, 4.7628931689371864e-06, 0.9998565023996473]",mixture,"{'mixture': 0.8408612401217148, 'uniform': 0.19158436581242347, 'top1': 0.8408604006205874, 'top2': 0.8408612401217148, 'top3': 0.8408612401217148}",0.864204841298776,0.0689438616669614,0.004439818108043122,0.865318273934081,0.8841677834896886,ensemble,1.0,
+5,335,84.2,213.6,"[0.2835080031881515, 0.0005110274916737301, 0.02308174871596876, 0.8489501264344202]",,1867412,310,False,,,,"[0.00012394890757176728, 7.142297450423627e-06, 4.380088376626105e-06, 0.9998645287066013]",mixture,"{'mixture': 0.8489512174331817, 'uniform': 0.20478316977482933, 'top1': 0.8489501264344202, 'top2': 0.8489512174331817, 'top3': 0.8489512174331817}",0.862984788458472,0.06815227306475158,0.005252702164533228,0.8753591055736171,0.8774800102230823,ensemble,1.0,
+6,402,83.5,238.4,"[0.27886655669468124, 0.0004317570658965266, 0.019823472474504847, 0.8510156994214092]",,2210520,377,False,,,,"[0.0001159092629984249, 6.990796383176612e-06, 4.153032658780695e-06, 0.9998729469079597]",mixture,"{'mixture': 0.851015583658806, 'uniform': 0.20455471975081585, 'top1': 0.8510156994214092, 'top2': 0.851015583658806, 'top3': 0.851015583658806}",0.8705280669376912,0.06705494363318357,0.005491697527176027,0.8635595641020113,0.8621746701279371,ensemble,1.0,
+7,402,0.0,242.4,,,,,,,,,,,,0.8825650776052844,0.06109141014936791,0.004077543849137184,,,,,True

macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/log.txt

@@ -0,0 +1,107 @@
+[12:16:33][      0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-laplace budget=5.0% seed=2 | {"task": "mitochondria", "target": "lucchi-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": 8, "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}
+[12:16:33][      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.
+[12:16:33][      0s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches
+[12:16:35][      2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60
+[12:16:35][      2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s
+[12:16:36][      3s][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
+[12:16:36][      3s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass
+[12:17:14][     41s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833
+[12:17:14][     41s][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
+[12:17:14][     41s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920
+[12:17:14][     41s][INIT] student initialised from teacher multiclass (weight 0.920)
+[12:17:17][     44s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[12:17:17][     44s][KD] distilling the mixture ensemble (prob space) into the student
+[12:19:20][    167s][KD] soft labels for 6000 patches from 4 teachers in 123s
+[12:22:19][    346s][KD] epoch 1/5 | patches=6000 | loss=0.1187 | 179s
+[12:25:18][    525s][KD] epoch 2/5 | patches=6000 | loss=0.1069 | 179s
+[12:28:17][    704s][KD] epoch 3/5 | patches=6000 | loss=0.1048 | 179s
+[12:31:16][    883s][KD] epoch 4/5 | patches=6000 | loss=0.1037 | 179s
+[12:34:15][   1062s][KD] epoch 5/5 | patches=6000 | loss=0.1029 | 179s
+[12:34:21][   1068s][ROUND0] before any target label: val_dice=0.8804 val_vi=0.0672
+[12:34:21][   1068s][AL] patches=8050 (pool 8050 after dropping blank; free blank negatives: True) | budget=402 (5.0%) | rounds=6 | per_round=67 | acquisition=laplace | first_round=stratified
+[12:35:48][   1155s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': 0.09212662440628899, 'auroc_top20_error': 0.47728173828125}
+[12:36:29][   1197s][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
+[12:36:35][   1202s][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
+[12:36:36][   1203s][WEIGHTS] round 1: student kept (labeled Dice 0.948 vs best teacher multiclass 0.946); distilling the corrected ensemble into it
+[12:36:36][   1203s][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
+[12:38:39][   1327s][KD] soft labels for 6000 patches from 4 teachers in 123s
+[12:41:38][   1505s][KD] epoch 1/2 | patches=6000 | loss=0.0470 | 179s
+[12:44:37][   1684s][KD] epoch 2/2 | patches=6000 | loss=0.0482 | 179s
+[12:45:19][   1726s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 41s
+[12:45:23][   1730s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3215
+[12:45:28][   1735s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3800
+[12:45:33][   1740s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3388
+[12:45:38][   1745s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.3076
+[12:45:43][   1750s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.2970
+[12:45:48][   1755s][AL] round 1/6 | labeled=67 | selection_seconds=128.6000 round_seconds=681.7000 restart_student_dice=0.9485 restart_teacher_dice=0.9463 val_dice=0.8306 val_vi=0.0835 val_ece=0.0068
+[12:47:15][   1842s][AL] round 2: 1074/7983 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:47:26][   1853s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.837, uniform=0.226, top1=0.837, top2=0.837, top3=0.837 -> using mixture
+[12:47:26][   1853s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (646357 fg px): fly-mito=0.242, kasthuri=0.000, dros-vnc=0.013, multiclass=0.837 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[12:47:30][   1857s][CONS] round 2: on the 67 new patches student Dice=0.706 vs ensemble 0.732 -> pseudo-labels from the ensemble | consistency x1.00
+[12:47:38][   1865s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.4784
+[12:47:47][   1874s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.4344
+[12:47:55][   1883s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.4149
+[12:48:04][   1891s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.3680
+[12:48:13][   1900s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.3755
+[12:48:19][   1906s][AL] round 2/6 | labeled=134 | selection_seconds=86.7000 round_seconds=144.5000 pseudo_student_dice=0.7064 pseudo_ensemble_dice=0.7318 consistency_scale=1.0000 val_dice=0.8366 val_vi=0.0717 val_ece=0.0038
+[12:49:45][   1992s][AL] round 3: 1032/7916 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:50:01][   2008s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.817, uniform=0.202, top1=0.817, top2=0.817, top3=0.817 -> using mixture
+[12:50:01][   2008s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (961409 fg px): fly-mito=0.243, kasthuri=0.000, dros-vnc=0.012, multiclass=0.817 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[12:50:16][   2023s][CONS] round 3: on the 67 new patches student Dice=0.809 vs ensemble 0.777 -> pseudo-labels from the student | consistency x1.00
+[12:50:29][   2036s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.3748
+[12:50:42][   2049s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.3606
+[12:50:54][   2062s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.3386
+[12:51:07][   2074s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.3334
+[12:51:20][   2087s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2969
+[12:51:26][   2093s][AL] round 3/6 | labeled=201 | selection_seconds=86.9000 round_seconds=181.5000 pseudo_student_dice=0.8086 pseudo_ensemble_dice=0.7771 consistency_scale=1.0000 val_dice=0.8521 val_vi=0.0731 val_ece=0.0066
+[12:52:52][   2179s][AL] round 4: 1253/7849 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[12:54:27][   2274s][WEIGHTS] round 4: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.291, kasthuri=0.000, dros-vnc=0.023, multiclass=0.952 | blended with labeled Dice at 0.50
+[12:55:52][   2359s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.891, uniform=0.264, top1=0.891, top2=0.891, top3=0.891 -> using mixture
+[12:55:52][   2359s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1406676 fg px): fly-mito=0.277, kasthuri=0.000, dros-vnc=0.018, multiclass=0.891 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[12:56:06][   2373s][CONS] round 4: on the 67 new patches student Dice=0.854 vs ensemble 0.859 -> pseudo-labels from the student | consistency x1.00
+[12:56:23][   2390s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.3130
+[12:56:39][   2406s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.2841
+[12:56:56][   2423s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.3083
+[12:57:13][   2440s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.2707
+[12:57:29][   2456s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.2564
+[12:57:35][   2462s][AL] round 4/6 | labeled=268 | selection_seconds=85.8000 round_seconds=363.0000 pseudo_student_dice=0.8540 pseudo_ensemble_dice=0.8587 consistency_scale=1.0000 val_dice=0.8240 val_vi=0.0822 val_ece=0.0069
+[12:59:00][   2547s][AL] round 5: 1164/7782 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[13:00:37][   2644s][WEIGHTS] round 5: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.283, kasthuri=0.000, dros-vnc=0.022, multiclass=0.948 | blended with labeled Dice at 0.50
+[13:02:05][   2732s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.893, uniform=0.261, top1=0.893, top2=0.893, top3=0.893 -> using mixture
+[13:02:05][   2732s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (1792685 fg px): fly-mito=0.276, kasthuri=0.000, dros-vnc=0.017, multiclass=0.893 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[13:02:19][   2746s][CONS] round 5: on the 67 new patches student Dice=0.855 vs ensemble 0.872 -> pseudo-labels from the student | consistency x1.00
+[13:02:40][   2767s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2735
+[13:03:00][   2787s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.2802
+[13:03:21][   2808s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.2619
+[13:03:41][   2828s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.2487
+[13:04:02][   2849s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.2306
+[13:04:07][   2855s][AL] round 5/6 | labeled=335 | selection_seconds=84.9000 round_seconds=386.6000 pseudo_student_dice=0.8549 pseudo_ensemble_dice=0.8723 consistency_scale=1.0000 val_dice=0.8753 val_vi=0.0637 val_ece=0.0034
+[13:05:32][   2939s][AL] round 6: 1238/7715 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[13:07:11][   3038s][WEIGHTS] round 6: teacher agreement with the student on 2000 unlabeled patches: fly-mito=0.282, kasthuri=0.001, dros-vnc=0.018, multiclass=0.957 | blended with labeled Dice at 0.50
+[13:08:39][   3126s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.902, uniform=0.267, top1=0.902, top2=0.902, top3=0.902 -> using mixture
+[13:08:39][   3126s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (2180202 fg px): fly-mito=0.278, kasthuri=0.000, dros-vnc=0.015, multiclass=0.902 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[13:08:54][   3141s][CONS] round 6: on the 67 new patches student Dice=0.892 vs ensemble 0.889 -> pseudo-labels from the student | consistency x1.00
+[13:09:18][   3165s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2037
+[13:09:43][   3190s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.2282
+[13:10:07][   3214s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.2195
+[13:10:32][   3239s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.2240
+[13:10:56][   3263s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.2315
+[13:11:02][   3269s][AL] round 6/6 | labeled=402 | selection_seconds=84.4000 round_seconds=409.0000 pseudo_student_dice=0.8924 pseudo_ensemble_dice=0.8893 consistency_scale=1.0000 val_dice=0.8780 val_vi=0.0607 val_ece=0.0039
+[13:11:13][   3280s][CONS] final consolidation: pseudo-labels refreshed from the student
+[13:11:37][   3304s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.2030
+[13:12:02][   3329s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.1907
+[13:12:26][   3353s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.1952
+[13:12:51][   3378s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.1823
+[13:13:15][   3402s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.1797
+[13:13:40][   3427s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1618
+[13:14:04][   3451s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1515
+[13:14:29][   3476s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1343
+[13:14:53][   3500s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1320
+[13:15:18][   3525s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1291
+[13:15:24][   3531s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=255.7000 val_dice=0.8830 val_vi=0.0588 val_ece=0.0039
+[13:15:24][   3531s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[13:16:47][   3614s][TEST] raw student (no EMA, no TTA): patch_dice=0.9102 patch_vi=0.0424
+[13:16:56][   3624s][TEST] final model without TTA: patch_dice=0.9102 patch_vi=0.0424
+[13:16:56][   3624s][RESOURCES] {"phases": {"setup": {"seconds": 0.8, "peak_cpu_rss_mb": 2160.2, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 42.4, "peak_cpu_rss_mb": 4175.1, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 1017.99, "peak_cpu_rss_mb": 5948.8, "peak_gpu_alloc_mb": 3826.7, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 2463.01, "peak_cpu_rss_mb": 5948.8, "peak_gpu_alloc_mb": 3947.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 92.62, "peak_cpu_rss_mb": 5948.8, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 18.9, "seconds_per_image": 1.852}}, "peak_gpu_alloc_mb": 3947.0, "peak_cpu_rss_mb": 5948.8, "total_seconds": 3616.8}
+[13:16:56][   3624s][TEST] patch_dice=0.9270 patch_iou=0.8989 patch_precision=0.9584 patch_recall=0.9368 patch_vi=0.0392 image_dice=0.9589 image_iou=0.9214 image_precision=0.9696 image_recall=0.9490 image_vi=0.0428 cal_ece=0.0017 cal_nll=0.0162 cal_brier=0.0033 cal_mean_confidence=0.9974 cal_accuracy=0.9958 cal_predicted_prevalence=0.0443 cal_true_prevalence=0.0452 cal_mean_probability=0.0447
+[13:16:57][   3624s][DONE] lucchi-pp macs-v11-norestart-acq-laplace b5 s2 | test patch Dice=0.9270 VI=0.0392 | image Dice=0.9589 VI=0.0428 | 3624s

macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-laplace/budget-5/seed-2/result.json

@@ -0,0 +1,662 @@
+{
+  "target": "lucchi-pp",
+  "variant": "macs-v11-norestart-acq-laplace",
+  "budget": 5.0,
+  "seed": 2,
+  "config": {
+    "task": "mitochondria",
+    "target": "lucchi-pp",
+    "variant": "macs-v11-norestart-acq-laplace",
+    "budget": 5.0,
+    "seed": 2,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "lucchi-pp",
+      "--variants",
+      "macs-v11-norestart-acq-laplace",
+      "--budgets",
+      "5",
+      "--seeds",
+      "2",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--task",
+      "mitochondria",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "commit": "aecf380",
+    "dirty": true
+  },
+  "sources": [
+    "fly-brain-mito-benchmark",
+    "kasthuri-pp",
+    "drosophila-vnc",
+    "multiclass"
+  ],
+  "alphas": [
+    0.03256907457032885,
+    0.031801008835420135,
+    0.015561893577512515,
+    0.9200680230167385
+  ],
+  "weights_all": {
+    "sources": [
+      "fly-brain-mito-benchmark",
+      "kasthuri-pp",
+      "drosophila-vnc",
+      "multiclass"
+    ],
+    "mixture": [
+      0.03256907457032885,
+      0.031801008835420135,
+      0.015561893577512515,
+      0.9200680230167385
+    ],
+    "per_source": [
+      0.013021514318439549,
+      0.021594227384204207,
+      0.00495364759466876,
+      0.9604306107026874
+    ],
+    "uniform": [
+      0.25,
+      0.25,
+      0.25,
+      0.25
+    ],
+    "mixture_info": {
+      "objective": 0.001413505929905978,
+      "mmd2_mixture": 0.005039697348754646,
+      "mmd2_uniform": 0.2092228336519747,
+      "mmd2_single_source": [
+        0.5680513520444281,
+        0.4644148125694437,
+        0.6002511462494531,
+        0.004698715465898062
+      ],
+      "median_squared_distance": 0.1892747764462741,
+      "gammas": [
+        1.3208310409546982,
+        2.6416620819093963,
+        5.283324163818793,
+        10.566648327637585,
+        21.13329665527517
+      ],
+      "max_weight": 0.9200680230167385,
+      "entropy": 0.36261914188486677,
+      "effective_sources": 1.4370884282110035,
+      "min_weight": 0.015561893577512515
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.6675354526109456,
+        0.5899267604637966,
+        0.8158227740743191,
+        0.007664790896468099
+      ],
+      "standardized": [
+        0.4800153933276398,
+        0.22710384776568787,
+        0.9632547653742088,
+        -1.670374006467536
+      ],
+      "tau": 0.5,
+      "max_weight": 0.9604306107026874,
+      "entropy": 0.2044174691373809,
+      "effective_sources": 1.2268102019632934,
+      "min_weight": 0.00495364759466876
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "multiclass",
+    "seconds": 1.4
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.24317352702666994,
+      0.16070093323689605,
+      0.16243146338599246,
+      0.8333945099982429
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.8341164139370107,
+      "mixture@raw": 0.8334634452984915,
+      "mixture@logit": 0.8348661307075531,
+      "per_source": 0.8335017830355866,
+      "per_source@raw": 0.8334355010464378,
+      "per_source@logit": 0.8337681001665089,
+      "uniform": 0.3312974448451058,
+      "uniform@raw": 0.18374038775198187,
+      "uniform@logit": 0.21963766526924644
+    },
+    "val_patches": 1024
+  },
+  "temperatures": [
+    1.3822230100631714,
+    2.5090832710266113,
+    2.767641305923462,
+    2.597146987915039
+  ],
+  "top_source": "multiclass",
+  "weight_stats": {
+    "max_weight": 0.9200680230167385,
+    "entropy": 0.36261914188486677,
+    "effective_sources": 1.4370884282110035,
+    "min_weight": 0.015561893577512515
+  },
+  "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": 1018.0,
+  "round0_val": {
+    "val_dice": 0.8803879820113556,
+    "val_vi": 0.06716140467912256,
+    "val_ece": 0.01670684602633669
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 67,
+      "selection_seconds": 128.6,
+      "round_seconds": 681.7,
+      "teacher_dice_labeled": [
+        0.26456266329153644,
+        3.1912075848621877e-06,
+        0.015528068385600497,
+        0.9463065640641752
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 313095,
+      "evidence_nonempty_patches": 59,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.9484527016764064,
+      "restart_teacher_dice": 0.9463065640641752,
+      "weights": [
+        3.8742771162253374e-05,
+        2.684472576277311e-06,
+        1.5342785029993225e-06,
+        0.9999570384777584
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9463065640641752,
+        "uniform": 0.3201075901937694,
+        "top1": 0.9463065640641752,
+        "top2": 0.9463065640641752,
+        "top3": 0.9463065640641752
+      },
+      "val_dice": 0.8306197011240201,
+      "val_vi": 0.08348743124097768,
+      "val_ece": 0.006788817784942072
+    },
+    {
+      "round": 2,
+      "n_labeled": 134,
+      "selection_seconds": 86.7,
+      "round_seconds": 144.5,
+      "teacher_dice_labeled": [
+        0.24219815146629306,
+        1.5464961809276811e-06,
+        0.012802129960691319,
+        0.8367222328254639
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 646357,
+      "evidence_nonempty_patches": 119,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        9.267327807422955e-05,
+        8.03051714291749e-06,
+        4.466396604498123e-06,
+        0.9998948298081783
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.8367218370151065,
+        "uniform": 0.22632976686687015,
+        "top1": 0.8367222328254639,
+        "top2": 0.8367218370151065,
+        "top3": 0.8367218370151065
+      },
+      "pseudo_student_dice": 0.7063903969579743,
+      "pseudo_ensemble_dice": 0.7317769245256655,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8365702659977134,
+      "val_vi": 0.07170842372260706,
+      "val_ece": 0.003773437115060449
+    },
+    {
+      "round": 3,
+      "n_labeled": 201,
+      "selection_seconds": 86.9,
+      "round_seconds": 181.5,
+      "teacher_dice_labeled": [
+        0.2432412176643748,
+        1.0398523409675826e-06,
+        0.011922198046396285,
+        0.8167738623605225
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 961409,
+      "evidence_nonempty_patches": 179,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        0.00011431641553317873,
+        9.803144569781586e-06,
+        5.404554281086763e-06,
+        0.9998704758856161
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.8167700660471827,
+        "uniform": 0.2017762310498385,
+        "top1": 0.8167738623605225,
+        "top2": 0.8167700660471827,
+        "top3": 0.8167700660471827
+      },
+      "pseudo_student_dice": 0.8085989086638157,
+      "pseudo_ensemble_dice": 0.7770533684262554,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8520933668114993,
+      "val_vi": 0.07309227793276327,
+      "val_ece": 0.006567630992236721
+    },
+    {
+      "round": 4,
+      "n_labeled": 268,
+      "selection_seconds": 85.8,
+      "round_seconds": 363.0,
+      "teacher_dice_labeled": [
+        0.2774407873320246,
+        0.00021757778155051012,
+        0.017681862132876883,
+        0.8908474538075571
+      ],
+      "teacher_dice_pool": [
+        0.2914352298181363,
+        0.00043444480174767376,
+        0.022986299099895986,
+        0.9522409389197227
+      ],
+      "evidence_fg_pixels": 1406676,
+      "evidence_nonempty_patches": 245,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        7.6728663456923e-05,
+        4.684119979198459e-06,
+        2.729580942504572e-06,
+        0.9999158576356213
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.8908468921531416,
+        "uniform": 0.26357952088220327,
+        "top1": 0.8908474538075571,
+        "top2": 0.8908466768504242,
+        "top3": 0.8908466745253185
+      },
+      "pseudo_student_dice": 0.8540059637638082,
+      "pseudo_ensemble_dice": 0.8586939563953454,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8240250559424885,
+      "val_vi": 0.08222283438241113,
+      "val_ece": 0.006853722569206671
+    },
+    {
+      "round": 5,
+      "n_labeled": 335,
+      "selection_seconds": 84.9,
+      "round_seconds": 386.6,
+      "teacher_dice_labeled": [
+        0.27581865697076663,
+        0.00021098453816418617,
+        0.01700586078043835,
+        0.8932543310747327
+      ],
+      "teacher_dice_pool": [
+        0.2830548899311145,
+        0.00042141133660207754,
+        0.0218039269162802,
+        0.9478203037556284
+      ],
+      "evidence_fg_pixels": 1792685,
+      "evidence_nonempty_patches": 307,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        7.369894582484202e-05,
+        4.572438172123279e-06,
+        2.646723698279208e-06,
+        0.9999190818923047
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.8932545092013533,
+        "uniform": 0.26145255935947603,
+        "top1": 0.8932543310747327,
+        "top2": 0.8932540626415855,
+        "top3": 0.8932542293319554
+      },
+      "pseudo_student_dice": 0.854906715658334,
+      "pseudo_ensemble_dice": 0.8722805399151158,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8753142915497427,
+      "val_vi": 0.06372080616488156,
+      "val_ece": 0.0033989463018334635
+    },
+    {
+      "round": 6,
+      "n_labeled": 402,
+      "selection_seconds": 84.4,
+      "round_seconds": 409.0,
+      "teacher_dice_labeled": [
+        0.2778182744095054,
+        0.00028432021766136274,
+        0.01471529608171681,
+        0.9022960378346366
+      ],
+      "teacher_dice_pool": [
+        0.2821117277570612,
+        0.0005681818181818182,
+        0.01807491548267771,
+        0.9568074274579239
+      ],
+      "evidence_fg_pixels": 2180202,
+      "evidence_nonempty_patches": 371,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        6.86879160649449e-05,
+        4.180239560132354e-06,
+        2.3631737087619997e-06,
+        0.9999247686706662
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9022957193488081,
+        "uniform": 0.26732216329589775,
+        "top1": 0.9022960378346366,
+        "top2": 0.9022954553953105,
+        "top3": 0.9022956474829681
+      },
+      "pseudo_student_dice": 0.8923665379571899,
+      "pseudo_ensemble_dice": 0.8892548845825352,
+      "pseudo_source": "student",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8780199357605633,
+      "val_vi": 0.06069790674314873,
+      "val_ece": 0.0038901085719583125
+    },
+    {
+      "round": 7,
+      "n_labeled": 402,
+      "selection_seconds": 0.0,
+      "round_seconds": 255.7,
+      "final_consolidation": true,
+      "val_dice": 0.8830426902129573,
+      "val_vi": 0.058771422313978854,
+      "val_ece": 0.003914845060847729
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": 0.09212662440628899,
+    "auroc_top20_error": 0.47728173828125
+  },
+  "al_seconds": 2463.0,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    6.86879160649449e-05,
+    4.180239560132354e-06,
+    2.3631737087619997e-06,
+    0.9999247686706662
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9999247686706662,
+    "entropy": 0.0008160555233599509,
+    "effective_sources": 1.0008163885872619,
+    "min_weight": 2.3631737087619997e-06
+  },
+  "top_source_final": "multiclass",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.9101973027773687,
+    "patch_iou": 0.8799479153075589,
+    "patch_precision": 0.9401215968693091,
+    "patch_recall": 0.9321027238495903,
+    "patch_vi": 0.042363949100457855,
+    "n_patches": 1750,
+    "cal_ece": 0.0028486549461974936,
+    "cal_nll": 0.022192469784087544,
+    "cal_brier": 0.003855580196129815,
+    "cal_mean_confidence": 0.9981672933390547,
+    "cal_accuracy": 0.9953186383928572,
+    "cal_predicted_prevalence": 0.04432533482142857,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.044275469432547426
+  },
+  "test_notta": {
+    "patch_dice": 0.9101973027773687,
+    "patch_iou": 0.8799479153075589,
+    "patch_precision": 0.9401215968693091,
+    "patch_recall": 0.9321027238495903,
+    "patch_vi": 0.042363949100457855,
+    "n_patches": 1750,
+    "cal_ece": 0.0028486549461974936,
+    "cal_nll": 0.022192469784087544,
+    "cal_brier": 0.003855580196129815,
+    "cal_mean_confidence": 0.9981672933390547,
+    "cal_accuracy": 0.9953186383928572,
+    "cal_predicted_prevalence": 0.04432533482142857,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.044275469432547426
+  },
+  "test": {
+    "patch_dice": 0.9270490015694021,
+    "patch_iou": 0.8989149479160026,
+    "patch_precision": 0.9583615572217189,
+    "patch_recall": 0.9367646860294109,
+    "patch_vi": 0.03922821826799574,
+    "n_patches": 1750,
+    "image_dice": 0.9588872499883614,
+    "image_iou": 0.9213606570440477,
+    "image_precision": 0.9695966441984472,
+    "image_recall": 0.9489955116436768,
+    "image_vi": 0.0427693149732126,
+    "n_images": 50,
+    "cal_ece": 0.0016793533075367766,
+    "cal_nll": 0.01617886206972107,
+    "cal_brier": 0.0033182671865588133,
+    "cal_mean_confidence": 0.9974321434861084,
+    "cal_accuracy": 0.9957527901785714,
+    "cal_predicted_prevalence": 0.044307477678571426,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.044659475159475076
+  },
+  "eval_seconds": 92.6,
+  "total_seconds": 3623.5,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 0.8,
+        "peak_cpu_rss_mb": 2160.2,
+        "peak_gpu_alloc_mb": 592.4,
+        "peak_gpu_reserved_mb": 594.0
+      },
+      "features_and_weights": {
+        "seconds": 42.4,
+        "peak_cpu_rss_mb": 4175.1,
+        "peak_gpu_alloc_mb": 2298.2,
+        "peak_gpu_reserved_mb": 2640.0
+      },
+      "distillation": {
+        "seconds": 1017.99,
+        "peak_cpu_rss_mb": 5948.8,
+        "peak_gpu_alloc_mb": 3826.7,
+        "peak_gpu_reserved_mb": 4128.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 2463.01,
+        "peak_cpu_rss_mb": 5948.8,
+        "peak_gpu_alloc_mb": 3947.0,
+        "peak_gpu_reserved_mb": 4176.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 92.62,
+        "peak_cpu_rss_mb": 5948.8,
+        "peak_gpu_alloc_mb": 2149.5,
+        "peak_gpu_reserved_mb": 4176.0,
+        "patches": 1750,
+        "images": 50,
+        "patches_per_second": 18.9,
+        "seconds_per_image": 1.852
+      }
+    },
+    "peak_gpu_alloc_mb": 3947.0,
+    "peak_cpu_rss_mb": 5948.8,
+    "total_seconds": 3616.8
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-laplace_b5_s2.pth"
+}

macs_v2_r26/runs/mitochondria/lucchi-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
+1,67,128.6,681.7,"[0.26456266329153644, 3.1912075848621877e-06, 0.015528068385600497, 0.9463065640641752]",,313095,59,True,kept,0.9484527016764064,0.9463065640641752,"[3.8742771162253374e-05, 2.684472576277311e-06, 1.5342785029993225e-06, 0.9999570384777584]",mixture,"{'mixture': 0.9463065640641752, 'uniform': 0.3201075901937694, 'top1': 0.9463065640641752, 'top2': 0.9463065640641752, 'top3': 0.9463065640641752}",0.8306197011240201,0.08348743124097768,0.006788817784942072,,,,,
+2,134,86.7,144.5,"[0.24219815146629306, 1.5464961809276811e-06, 0.012802129960691319, 0.8367222328254639]",,646357,119,False,,,,"[9.267327807422955e-05, 8.03051714291749e-06, 4.466396604498123e-06, 0.9998948298081783]",mixture,"{'mixture': 0.8367218370151065, 'uniform': 0.22632976686687015, 'top1': 0.8367222328254639, 'top2': 0.8367218370151065, 'top3': 0.8367218370151065}",0.8365702659977134,0.07170842372260706,0.003773437115060449,0.7063903969579743,0.7317769245256655,ensemble,1.0,
+3,201,86.9,181.5,"[0.2432412176643748, 1.0398523409675826e-06, 0.011922198046396285, 0.8167738623605225]",,961409,179,False,,,,"[0.00011431641553317873, 9.803144569781586e-06, 5.404554281086763e-06, 0.9998704758856161]",mixture,"{'mixture': 0.8167700660471827, 'uniform': 0.2017762310498385, 'top1': 0.8167738623605225, 'top2': 0.8167700660471827, 'top3': 0.8167700660471827}",0.8520933668114993,0.07309227793276327,0.006567630992236721,0.8085989086638157,0.7770533684262554,student,1.0,
+4,268,85.8,363.0,"[0.2774407873320246, 0.00021757778155051012, 0.017681862132876883, 0.8908474538075571]","[0.2914352298181363, 0.00043444480174767376, 0.022986299099895986, 0.9522409389197227]",1406676,245,False,,,,"[7.6728663456923e-05, 4.684119979198459e-06, 2.729580942504572e-06, 0.9999158576356213]",mixture,"{'mixture': 0.8908468921531416, 'uniform': 0.26357952088220327, 'top1': 0.8908474538075571, 'top2': 0.8908466768504242, 'top3': 0.8908466745253185}",0.8240250559424885,0.08222283438241113,0.006853722569206671,0.8540059637638082,0.8586939563953454,student,1.0,
+5,335,84.9,386.6,"[0.27581865697076663, 0.00021098453816418617, 0.01700586078043835, 0.8932543310747327]","[0.2830548899311145, 0.00042141133660207754, 0.0218039269162802, 0.9478203037556284]",1792685,307,False,,,,"[7.369894582484202e-05, 4.572438172123279e-06, 2.646723698279208e-06, 0.9999190818923047]",mixture,"{'mixture': 0.8932545092013533, 'uniform': 0.26145255935947603, 'top1': 0.8932543310747327, 'top2': 0.8932540626415855, 'top3': 0.8932542293319554}",0.8753142915497427,0.06372080616488156,0.0033989463018334635,0.854906715658334,0.8722805399151158,student,1.0,
+6,402,84.4,409.0,"[0.2778182744095054, 0.00028432021766136274, 0.01471529608171681, 0.9022960378346366]","[0.2821117277570612, 0.0005681818181818182, 0.01807491548267771, 0.9568074274579239]",2180202,371,False,,,,"[6.86879160649449e-05, 4.180239560132354e-06, 2.3631737087619997e-06, 0.9999247686706662]",mixture,"{'mixture': 0.9022957193488081, 'uniform': 0.26732216329589775, 'top1': 0.9022960378346366, 'top2': 0.9022954553953105, 'top3': 0.9022956474829681}",0.8780199357605633,0.06069790674314873,0.0038901085719583125,0.8923665379571899,0.8892548845825352,student,1.0,
+7,402,0.0,255.7,,,,,,,,,,,,0.8830426902129573,0.058771422313978854,0.003914845060847729,,,,,True

macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-1/log.txt

@@ -0,0 +1,98 @@
+[07:06:35][      0s][START] target=lucchi-pp variant=macs-v11-norestart-acq-random budget=5.0% seed=1 | {"task": "mitochondria", "target": "lucchi-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": 8, "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}
+[07:06:35][      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.
+[07:06:35][      0s][DATA] lucchi-pp: train=8050 val=1750 test=1750 patches
+[07:06:36][      2s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, dros-vnc=2.77, multiclass=2.60
+[07:06:36][      2s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 0s
+[07:06:38][      3s][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
+[07:06:38][      3s][WEIGHTS] per-source: fly-mito=0.013, kasthuri=0.022, dros-vnc=0.005, multiclass=0.960 | min-MMD source (NeuroADDA rule) = multiclass
+[07:07:15][     41s][TEACHERS] target-val Dice: fly-mito=0.243, kasthuri=0.161, dros-vnc=0.162, multiclass=0.833
+[07:07:15][     41s][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
+[07:07:15][     41s][WEIGHTS] using mixture: fly-mito=0.033, kasthuri=0.032, dros-vnc=0.016, multiclass=0.920
+[07:07:15][     41s][INIT] student initialised from teacher multiclass (weight 0.920)
+[07:07:19][     44s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[07:07:19][     44s][KD] distilling the mixture ensemble (prob space) into the student
+[07:09:21][    167s][KD] soft labels for 6000 patches from 4 teachers in 123s
+[07:12:19][    345s][KD] epoch 1/5 | patches=6000 | loss=0.1185 | 178s
+[07:15:17][    523s][KD] epoch 2/5 | patches=6000 | loss=0.1064 | 178s
+[07:18:15][    701s][KD] epoch 3/5 | patches=6000 | loss=0.1043 | 178s
+[07:21:13][    879s][KD] epoch 4/5 | patches=6000 | loss=0.1035 | 178s
+[07:24:11][   1057s][KD] epoch 5/5 | patches=6000 | loss=0.1029 | 178s
+[07:24:17][   1063s][ROUND0] before any target label: val_dice=0.8825 val_vi=0.0631
+[07:24:17][   1063s][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
+[07:24:59][   1104s][AL] round-1 uncertainty quality on 8000 patches: {'spearman_uncertainty_error': -0.0022443099490930996, 'auroc_top20_error': 0.49312246093750006}
+[07:25:40][   1145s][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
+[07:25:45][   1151s][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
+[07:25:46][   1152s][WEIGHTS] round 1: student kept (labeled Dice 0.940 vs best teacher multiclass 0.932); distilling the corrected ensemble into it
+[07:25:46][   1152s][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
+[07:27:49][   1274s][KD] soft labels for 6000 patches from 4 teachers in 123s
+[07:30:47][   1452s][KD] epoch 1/2 | patches=6000 | loss=0.0471 | 178s
+[07:33:45][   1630s][KD] epoch 2/2 | patches=6000 | loss=0.0459 | 178s
+[07:34:26][   1672s][CONS] pseudo-labels for 2000 unlabeled patches from the current ensemble in 41s
+[07:34:31][   1676s][FT] round 1/6 epoch 1/5 | patches=67 | loss=0.3095
+[07:34:36][   1681s][FT] round 1/6 epoch 2/5 | patches=67 | loss=0.3610
+[07:34:40][   1686s][FT] round 1/6 epoch 3/5 | patches=67 | loss=0.3744
+[07:34:45][   1690s][FT] round 1/6 epoch 4/5 | patches=67 | loss=0.2857
+[07:34:50][   1695s][FT] round 1/6 epoch 5/5 | patches=67 | loss=0.3504
+[07:34:56][   1701s][AL] round 1/6 | labeled=67 | selection_seconds=82.7000 round_seconds=632.7000 restart_student_dice=0.9398 restart_teacher_dice=0.9318 val_dice=0.7964 val_vi=0.0972 val_ece=0.0086
+[07:35:06][   1712s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.931, uniform=0.316, top1=0.931, top2=0.931, top3=0.931 -> using mixture
+[07:35:06][   1712s][WEIGHTS] round 2: teacher Dice on 134 labeled patches + 0 blank (686915 fg px): fly-mito=0.280, kasthuri=0.000, dros-vnc=0.031, multiclass=0.931 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[07:35:10][   1716s][CONS] round 2: on the 67 new patches student Dice=0.922 vs ensemble 0.930 -> pseudo-labels from the ensemble | consistency x1.00
+[07:35:19][   1724s][FT] round 2/6 epoch 1/5 | patches=134 | loss=0.3237
+[07:35:28][   1733s][FT] round 2/6 epoch 2/5 | patches=134 | loss=0.2841
+[07:35:36][   1741s][FT] round 2/6 epoch 3/5 | patches=134 | loss=0.2896
+[07:35:45][   1750s][FT] round 2/6 epoch 4/5 | patches=134 | loss=0.2877
+[07:35:53][   1759s][FT] round 2/6 epoch 5/5 | patches=134 | loss=0.2835
+[07:35:59][   1764s][AL] round 2/6 | labeled=134 | selection_seconds=0.2000 round_seconds=57.6000 pseudo_student_dice=0.9218 pseudo_ensemble_dice=0.9298 consistency_scale=1.0000 val_dice=0.8360 val_vi=0.0763 val_ece=0.0050
+[07:36:15][   1780s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.936, uniform=0.309, top1=0.936, top2=0.936, top3=0.936 -> using mixture
+[07:36:15][   1780s][WEIGHTS] round 3: teacher Dice on 201 labeled patches + 0 blank (1044085 fg px): fly-mito=0.285, kasthuri=0.000, dros-vnc=0.028, multiclass=0.936 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[07:36:19][   1784s][CONS] round 3: on the 67 new patches student Dice=0.949 vs ensemble 0.946 -> pseudo-labels from the ensemble | consistency x1.00
+[07:36:31][   1797s][FT] round 3/6 epoch 1/5 | patches=201 | loss=0.2540
+[07:36:44][   1809s][FT] round 3/6 epoch 2/5 | patches=201 | loss=0.3283
+[07:36:56][   1822s][FT] round 3/6 epoch 3/5 | patches=201 | loss=0.2696
+[07:37:09][   1834s][FT] round 3/6 epoch 4/5 | patches=201 | loss=0.2373
+[07:37:22][   1847s][FT] round 3/6 epoch 5/5 | patches=201 | loss=0.2316
+[07:37:27][   1853s][AL] round 3/6 | labeled=201 | selection_seconds=0.2000 round_seconds=82.7000 pseudo_student_dice=0.9488 pseudo_ensemble_dice=0.9456 consistency_scale=1.0000 val_dice=0.8718 val_vi=0.0733 val_ece=0.0063
+[07:37:48][   1873s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.934, uniform=0.297, top1=0.934, top2=0.934, top3=0.934 -> using mixture
+[07:37:48][   1873s][WEIGHTS] round 4: teacher Dice on 268 labeled patches + 0 blank (1398805 fg px): fly-mito=0.289, kasthuri=0.000, dros-vnc=0.027, multiclass=0.934 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[07:37:52][   1877s][CONS] round 4: on the 67 new patches student Dice=0.938 vs ensemble 0.929 -> pseudo-labels from the ensemble | consistency x1.00
+[07:38:08][   1894s][FT] round 4/6 epoch 1/5 | patches=268 | loss=0.2372
+[07:38:25][   1910s][FT] round 4/6 epoch 2/5 | patches=268 | loss=0.2139
+[07:38:41][   1927s][FT] round 4/6 epoch 3/5 | patches=268 | loss=0.3157
+[07:38:58][   1943s][FT] round 4/6 epoch 4/5 | patches=268 | loss=0.2718
+[07:39:14][   1959s][FT] round 4/6 epoch 5/5 | patches=268 | loss=0.2351
+[07:39:20][   1965s][AL] round 4/6 | labeled=268 | selection_seconds=0.2000 round_seconds=106.7000 pseudo_student_dice=0.9379 pseudo_ensemble_dice=0.9288 consistency_scale=1.0000 val_dice=0.8609 val_vi=0.0737 val_ece=0.0064
+[07:39:45][   1991s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.935, uniform=0.314, top1=0.935, top2=0.935, top3=0.935 -> using mixture
+[07:39:45][   1991s][WEIGHTS] round 5: teacher Dice on 335 labeled patches + 0 blank (1751052 fg px): fly-mito=0.285, kasthuri=0.000, dros-vnc=0.024, multiclass=0.935 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[07:39:49][   1995s][CONS] round 5: on the 67 new patches student Dice=0.944 vs ensemble 0.937 -> pseudo-labels from the ensemble | consistency x1.00
+[07:40:09][   2015s][FT] round 5/6 epoch 1/5 | patches=335 | loss=0.2320
+[07:40:30][   2035s][FT] round 5/6 epoch 2/5 | patches=335 | loss=0.2251
+[07:40:50][   2055s][FT] round 5/6 epoch 3/5 | patches=335 | loss=0.2079
+[07:41:10][   2076s][FT] round 5/6 epoch 4/5 | patches=335 | loss=0.2125
+[07:41:30][   2096s][FT] round 5/6 epoch 5/5 | patches=335 | loss=0.2060
+[07:41:36][   2102s][AL] round 5/6 | labeled=335 | selection_seconds=0.2000 round_seconds=130.5000 pseudo_student_dice=0.9436 pseudo_ensemble_dice=0.9366 consistency_scale=1.0000 val_dice=0.8758 val_vi=0.0692 val_ece=0.0052
+[07:42:06][   2132s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.934, uniform=0.324, top1=0.934, top2=0.934, top3=0.934 -> using mixture
+[07:42:06][   2132s][WEIGHTS] round 6: teacher Dice on 402 labeled patches + 0 blank (2057337 fg px): fly-mito=0.282, kasthuri=0.000, dros-vnc=0.023, multiclass=0.934 | weights: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[07:42:11][   2136s][CONS] round 6: on the 67 new patches student Dice=0.934 vs ensemble 0.929 -> pseudo-labels from the ensemble | consistency x1.00
+[07:42:35][   2160s][FT] round 6/6 epoch 1/5 | patches=402 | loss=0.2464
+[07:42:59][   2184s][FT] round 6/6 epoch 2/5 | patches=402 | loss=0.2427
+[07:43:23][   2209s][FT] round 6/6 epoch 3/5 | patches=402 | loss=0.2298
+[07:43:48][   2233s][FT] round 6/6 epoch 4/5 | patches=402 | loss=0.2270
+[07:44:12][   2257s][FT] round 6/6 epoch 5/5 | patches=402 | loss=0.2242
+[07:44:18][   2263s][AL] round 6/6 | labeled=402 | selection_seconds=0.2000 round_seconds=155.8000 pseudo_student_dice=0.9343 pseudo_ensemble_dice=0.9295 consistency_scale=1.0000 val_dice=0.8795 val_vi=0.0632 val_ece=0.0040
+[07:44:42][   2288s][FT] final 1/10 epoch 1/1 | patches=402 | loss=0.2052
+[07:45:07][   2312s][FT] final 2/10 epoch 1/1 | patches=402 | loss=0.2055
+[07:45:31][   2336s][FT] final 3/10 epoch 1/1 | patches=402 | loss=0.2056
+[07:45:55][   2361s][FT] final 4/10 epoch 1/1 | patches=402 | loss=0.2401
+[07:46:20][   2385s][FT] final 5/10 epoch 1/1 | patches=402 | loss=0.2020
+[07:46:44][   2410s][FT] final 6/10 epoch 1/1 | patches=402 | loss=0.1953
+[07:47:09][   2434s][FT] final 7/10 epoch 1/1 | patches=402 | loss=0.1838
+[07:47:33][   2458s][FT] final 8/10 epoch 1/1 | patches=402 | loss=0.1807
+[07:47:57][   2483s][FT] final 9/10 epoch 1/1 | patches=402 | loss=0.1836
+[07:48:22][   2507s][FT] final 10/10 epoch 1/1 | patches=402 | loss=0.1761
+[07:48:28][   2513s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=243.8000 val_dice=0.8844 val_vi=0.0651 val_ece=0.0045
+[07:48:28][   2513s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, dros-vnc=0.000, multiclass=1.000
+[07:49:50][   2595s][TEST] raw student (no EMA, no TTA): patch_dice=0.9146 patch_vi=0.0435
+[07:49:59][   2605s][TEST] final model without TTA: patch_dice=0.9146 patch_vi=0.0435
+[07:49:59][   2605s][RESOURCES] {"phases": {"setup": {"seconds": 0.75, "peak_cpu_rss_mb": 2174.8, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 42.33, "peak_cpu_rss_mb": 4176.2, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 1012.81, "peak_cpu_rss_mb": 5951.0, "peak_gpu_alloc_mb": 3826.7, "peak_gpu_reserved_mb": 4128.0, "kd_epochs": 5}, "active_learning": {"seconds": 1450.45, "peak_cpu_rss_mb": 5951.0, "peak_gpu_alloc_mb": 3947.0, "peak_gpu_reserved_mb": 4176.0, "rounds": 6}, "inference": {"seconds": 91.64, "peak_cpu_rss_mb": 5951.0, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4176.0, "patches": 1750, "images": 50, "patches_per_second": 19.1, "seconds_per_image": 1.833}}, "peak_gpu_alloc_mb": 3947.0, "peak_cpu_rss_mb": 5951.0, "total_seconds": 2598.0}
+[07:49:59][   2605s][TEST] patch_dice=0.9217 patch_iou=0.8940 patch_precision=0.9546 patch_recall=0.9358 patch_vi=0.0413 image_dice=0.9519 image_iou=0.9087 image_precision=0.9614 image_recall=0.9437 image_vi=0.0484 cal_ece=0.0017 cal_nll=0.0169 cal_brier=0.0036 cal_mean_confidence=0.9970 cal_accuracy=0.9953 cal_predicted_prevalence=0.0446 cal_true_prevalence=0.0452 cal_mean_probability=0.0448
+[07:50:00][   2605s][DONE] lucchi-pp macs-v11-norestart-acq-random b5 s1 | test patch Dice=0.9217 VI=0.0413 | image Dice=0.9519 VI=0.0484 | 2605s

macs_v2_r26/runs/mitochondria/lucchi-pp/macs-v11-norestart-acq-random/budget-5/seed-1/result.json

@@ -0,0 +1,647 @@
+{
+  "target": "lucchi-pp",
+  "variant": "macs-v11-norestart-acq-random",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "mitochondria",
+    "target": "lucchi-pp",
+    "variant": "macs-v11-norestart-acq-random",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "lucchi-pp",
+      "--variants",
+      "macs-v11-norestart-acq-random",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--task",
+      "mitochondria",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "commit": "aecf380",
+    "dirty": true
+  },
+  "sources": [
+    "fly-brain-mito-benchmark",
+    "kasthuri-pp",
+    "drosophila-vnc",
+    "multiclass"
+  ],
+  "alphas": [
+    0.03256907457032885,
+    0.031801008835420135,
+    0.015561893577512515,
+    0.9200680230167385
+  ],
+  "weights_all": {
+    "sources": [
+      "fly-brain-mito-benchmark",
+      "kasthuri-pp",
+      "drosophila-vnc",
+      "multiclass"
+    ],
+    "mixture": [
+      0.03256907457032885,
+      0.031801008835420135,
+      0.015561893577512515,
+      0.9200680230167385
+    ],
+    "per_source": [
+      0.013021514318439549,
+      0.021594227384204207,
+      0.00495364759466876,
+      0.9604306107026874
+    ],
+    "uniform": [
+      0.25,
+      0.25,
+      0.25,
+      0.25
+    ],
+    "mixture_info": {
+      "objective": 0.001413505929905978,
+      "mmd2_mixture": 0.005039697348754646,
+      "mmd2_uniform": 0.2092228336519747,
+      "mmd2_single_source": [
+        0.5680513520444281,
+        0.4644148125694437,
+        0.6002511462494531,
+        0.004698715465898062
+      ],
+      "median_squared_distance": 0.1892747764462741,
+      "gammas": [
+        1.3208310409546982,
+        2.6416620819093963,
+        5.283324163818793,
+        10.566648327637585,
+        21.13329665527517
+      ],
+      "max_weight": 0.9200680230167385,
+      "entropy": 0.36261914188486677,
+      "effective_sources": 1.4370884282110035,
+      "min_weight": 0.015561893577512515
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.6675354526109456,
+        0.5899267604637966,
+        0.8158227740743191,
+        0.007664790896468099
+      ],
+      "standardized": [
+        0.4800153933276398,
+        0.22710384776568787,
+        0.9632547653742088,
+        -1.670374006467536
+      ],
+      "tau": 0.5,
+      "max_weight": 0.9604306107026874,
+      "entropy": 0.2044174691373809,
+      "effective_sources": 1.2268102019632934,
+      "min_weight": 0.00495364759466876
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "multiclass",
+    "seconds": 1.3
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.24317352702666994,
+      0.16070093323689605,
+      0.16243146338599246,
+      0.8333945099982429
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.8341164139370107,
+      "mixture@raw": 0.8334634452984915,
+      "mixture@logit": 0.8348661307075531,
+      "per_source": 0.8335017830355866,
+      "per_source@raw": 0.8334355010464378,
+      "per_source@logit": 0.8337681001665089,
+      "uniform": 0.3312974448451058,
+      "uniform@raw": 0.18374038775198187,
+      "uniform@logit": 0.21963766526924644
+    },
+    "val_patches": 1024
+  },
+  "temperatures": [
+    1.3822230100631714,
+    2.5090832710266113,
+    2.767641305923462,
+    2.597146987915039
+  ],
+  "top_source": "multiclass",
+  "weight_stats": {
+    "max_weight": 0.9200680230167385,
+    "entropy": 0.36261914188486677,
+    "effective_sources": 1.4370884282110035,
+    "min_weight": 0.015561893577512515
+  },
+  "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": 1012.8,
+  "round0_val": {
+    "val_dice": 0.8824999818736274,
+    "val_vi": 0.06309425262480922,
+    "val_ece": 0.01823718431407162
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 67,
+      "selection_seconds": 82.7,
+      "round_seconds": 632.7,
+      "teacher_dice_labeled": [
+        0.2876389731008006,
+        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": 0.939764804154994,
+      "restart_teacher_dice": 0.931847537434372,
+      "weights": [
+        5.6389282479936555e-05,
+        3.1020171205931166e-06,
+        1.907901148148134e-06,
+        0.9999386007992513
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.931847537434372,
+        "uniform": 0.3539312329080638,
+        "top1": 0.931847537434372,
+        "top2": 0.931847537434372,
+        "top3": 0.931847537434372
+      },
+      "val_dice": 0.7963996417830616,
+      "val_vi": 0.09719983582607203,
+      "val_ece": 0.008608905565021063
+    },
+    {
+      "round": 2,
+      "n_labeled": 134,
+      "selection_seconds": 0.2,
+      "round_seconds": 57.6,
+      "teacher_dice_labeled": [
+        0.2797322410747025,
+        1.455782075246464e-06,
+        0.03129283038951154,
+        0.9308365557494995
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 686915,
+      "evidence_nonempty_patches": 121,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        5.2632044600482846e-05,
+        3.1335047376057656e-06,
+        2.0967594661903174e-06,
+        0.9999421376911957
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9308365557494995,
+        "uniform": 0.3157631136821172,
+        "top1": 0.9308365557494995,
+        "top2": 0.9308365557494995,
+        "top3": 0.9308365557494995
+      },
+      "pseudo_student_dice": 0.9217522272049676,
+      "pseudo_ensemble_dice": 0.9297682266068903,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8360258919249423,
+      "val_vi": 0.07626112770829221,
+      "val_ece": 0.005044878277267725
+    },
+    {
+      "round": 3,
+      "n_labeled": 201,
+      "selection_seconds": 0.2,
+      "round_seconds": 82.7,
+      "teacher_dice_labeled": [
+        0.285088893636434,
+        9.573124792382856e-07,
+        0.027690813420591064,
+        0.9358991276509686
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1044085,
+      "evidence_nonempty_patches": 179,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        5.278706224823984e-05,
+        2.978803091277857e-06,
+        1.922732452354128e-06,
+        0.9999423114022081
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9358986801938234,
+        "uniform": 0.3092815616297828,
+        "top1": 0.9358991276509686,
+        "top2": 0.9358986801938234,
+        "top3": 0.9358986801938234
+      },
+      "pseudo_student_dice": 0.9488361135853891,
+      "pseudo_ensemble_dice": 0.9455641796027374,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8717971151028363,
+      "val_vi": 0.07333970560102024,
+      "val_ece": 0.006264443101983306
+    },
+    {
+      "round": 4,
+      "n_labeled": 268,
+      "selection_seconds": 0.2,
+      "round_seconds": 106.7,
+      "teacher_dice_labeled": [
+        0.2887129617740495,
+        7.146374179864233e-07,
+        0.02675218913568767,
+        0.9341140653396488
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1398805,
+      "evidence_nonempty_patches": 243,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        5.572085136061948e-05,
+        3.0324375408485426e-06,
+        1.9390704386544167e-06,
+        0.99993930764066
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9341141124811376,
+        "uniform": 0.2974809266161287,
+        "top1": 0.9341140653396488,
+        "top2": 0.9341141124811376,
+        "top3": 0.9341141124811376
+      },
+      "pseudo_student_dice": 0.9379381073761351,
+      "pseudo_ensemble_dice": 0.9288095752421666,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8609243697760304,
+      "val_vi": 0.0737337951427876,
+      "val_ece": 0.006416487514190574
+    },
+    {
+      "round": 5,
+      "n_labeled": 335,
+      "selection_seconds": 0.2,
+      "round_seconds": 130.5,
+      "teacher_dice_labeled": [
+        0.2851837942410718,
+        5.709202892510553e-07,
+        0.02400631314910587,
+        0.9346082435626396
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 1751052,
+      "evidence_nonempty_patches": 304,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        5.352363032039484e-05,
+        3.017491387355586e-06,
+        1.8772546836238203e-06,
+        0.9999415816236087
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.934608280971694,
+        "uniform": 0.3136637816728857,
+        "top1": 0.9346082435626396,
+        "top2": 0.934608280971694,
+        "top3": 0.934608280971694
+      },
+      "pseudo_student_dice": 0.9435884035290092,
+      "pseudo_ensemble_dice": 0.9365794422929289,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8758075852599518,
+      "val_vi": 0.06919363862118208,
+      "val_ece": 0.005191076904368195
+    },
+    {
+      "round": 6,
+      "n_labeled": 402,
+      "selection_seconds": 0.2,
+      "round_seconds": 155.8,
+      "teacher_dice_labeled": [
+        0.28191093592370076,
+        4.857738693491748e-07,
+        0.022959482419344,
+        0.9338568269067726
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 2057337,
+      "evidence_nonempty_patches": 357,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        5.219100403170781e-05,
+        3.040252132571167e-06,
+        1.8717196545786043e-06,
+        0.9999428970241812
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9338566313420587,
+        "uniform": 0.32391720725548095,
+        "top1": 0.9338568269067726,
+        "top2": 0.9338566313420587,
+        "top3": 0.9338566313420587
+      },
+      "pseudo_student_dice": 0.9342727392589453,
+      "pseudo_ensemble_dice": 0.9294967729090275,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.8795111878237105,
+      "val_vi": 0.063227044280279,
+      "val_ece": 0.00398630717600138
+    },
+    {
+      "round": 7,
+      "n_labeled": 402,
+      "selection_seconds": 0.0,
+      "round_seconds": 243.8,
+      "final_consolidation": true,
+      "val_dice": 0.884373415141364,
+      "val_vi": 0.06514450746984593,
+      "val_ece": 0.004541055743000292
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": -0.0022443099490930996,
+    "auroc_top20_error": 0.49312246093750006
+  },
+  "al_seconds": 1450.4,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    5.219100403170781e-05,
+    3.040252132571167e-06,
+    1.8717196545786043e-06,
+    0.9999428970241812
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9999428970241812,
+    "entropy": 0.0006350434984520711,
+    "effective_sources": 1.0006352451812648,
+    "min_weight": 1.8717196545786043e-06
+  },
+  "top_source_final": "multiclass",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.9146401557969808,
+    "patch_iou": 0.8843900144666924,
+    "patch_precision": 0.9444387414189249,
+    "patch_recall": 0.9370968354478443,
+    "patch_vi": 0.0434695779629698,
+    "n_patches": 1750,
+    "cal_ece": 0.002688002960785389,
+    "cal_nll": 0.021318272008663725,
+    "cal_brier": 0.004021202493275474,
+    "cal_mean_confidence": 0.997700837782214,
+    "cal_accuracy": 0.9950128348214285,
+    "cal_predicted_prevalence": 0.04469475446428572,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.04464832570282386
+  },
+  "test_notta": {
+    "patch_dice": 0.9146401557969808,
+    "patch_iou": 0.8843900144666924,
+    "patch_precision": 0.9444387414189249,
+    "patch_recall": 0.9370968354478443,
+    "patch_vi": 0.0434695779629698,
+    "n_patches": 1750,
+    "cal_ece": 0.002688002960785389,
+    "cal_nll": 0.021318272008663725,
+    "cal_brier": 0.004021202493275474,
+    "cal_mean_confidence": 0.997700837782214,
+    "cal_accuracy": 0.9950128348214285,
+    "cal_predicted_prevalence": 0.04469475446428572,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.04464832570282386
+  },
+  "test": {
+    "patch_dice": 0.9216756956926461,
+    "patch_iou": 0.8940452932960153,
+    "patch_precision": 0.9545525749986802,
+    "patch_recall": 0.9358326668206619,
+    "patch_vi": 0.04133018567270993,
+    "n_patches": 1750,
+    "image_dice": 0.9518754282319007,
+    "image_iou": 0.908654900675551,
+    "image_precision": 0.9613938476386131,
+    "image_recall": 0.9436737891974423,
+    "image_vi": 0.04840384507907384,
+    "n_images": 50,
+    "cal_ece": 0.0017047703042317487,
+    "cal_nll": 0.016949773001781545,
+    "cal_brier": 0.0036152859965237925,
+    "cal_mean_confidence": 0.9969826720899462,
+    "cal_accuracy": 0.9952779017857143,
+    "cal_predicted_prevalence": 0.044606026785714285,
+    "cal_true_prevalence": 0.04515513392857143,
+    "cal_mean_probability": 0.04480511960019529
+  },
+  "eval_seconds": 91.6,
+  "total_seconds": 2604.6,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 0.75,
+        "peak_cpu_rss_mb": 2174.8,
+        "peak_gpu_alloc_mb": 592.4,
+        "peak_gpu_reserved_mb": 594.0
+      },
+      "features_and_weights": {
+        "seconds": 42.33,
+        "peak_cpu_rss_mb": 4176.2,
+        "peak_gpu_alloc_mb": 2298.2,
+        "peak_gpu_reserved_mb": 2640.0
+      },
+      "distillation": {
+        "seconds": 1012.81,
+        "peak_cpu_rss_mb": 5951.0,
+        "peak_gpu_alloc_mb": 3826.7,
+        "peak_gpu_reserved_mb": 4128.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 1450.45,
+        "peak_cpu_rss_mb": 5951.0,
+        "peak_gpu_alloc_mb": 3947.0,
+        "peak_gpu_reserved_mb": 4176.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 91.64,
+        "peak_cpu_rss_mb": 5951.0,
+        "peak_gpu_alloc_mb": 2149.5,
+        "peak_gpu_reserved_mb": 4176.0,
+        "patches": 1750,
+        "images": 50,
+        "patches_per_second": 19.1,
+        "seconds_per_image": 1.833
+      }
+    },
+    "peak_gpu_alloc_mb": 3947.0,
+    "peak_cpu_rss_mb": 5951.0,
+    "total_seconds": 2598.0
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/lucchi-pp/macs-v11-norestart-acq-random_b5_s1.pth"
+}

macs_v2_r26/runs/mitochondria/lucchi-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
+1,67,82.7,632.7,"[0.2876389731008006, 2.8393119779215102e-06, 0.02286523788150808, 0.931847537434372]",,352197,62,True,kept,0.939764804154994,0.931847537434372,"[5.6389282479936555e-05, 3.1020171205931166e-06, 1.907901148148134e-06, 0.9999386007992513]",mixture,"{'mixture': 0.931847537434372, 'uniform': 0.3539312329080638, 'top1': 0.931847537434372, 'top2': 0.931847537434372, 'top3': 0.931847537434372}",0.7963996417830616,0.09719983582607203,0.008608905565021063,,,,,
+2,134,0.2,57.6,"[0.2797322410747025, 1.455782075246464e-06, 0.03129283038951154, 0.9308365557494995]",,686915,121,False,,,,"[5.2632044600482846e-05, 3.1335047376057656e-06, 2.0967594661903174e-06, 0.9999421376911957]",mixture,"{'mixture': 0.9308365557494995, 'uniform': 0.3157631136821172, 'top1': 0.9308365557494995, 'top2': 0.9308365557494995, 'top3': 0.9308365557494995}",0.8360258919249423,0.07626112770829221,0.005044878277267725,0.9217522272049676,0.9297682266068903,ensemble,1.0,
+3,201,0.2,82.7,"[0.285088893636434, 9.573124792382856e-07, 0.027690813420591064, 0.9358991276509686]",,1044085,179,False,,,,"[5.278706224823984e-05, 2.978803091277857e-06, 1.922732452354128e-06, 0.9999423114022081]",mixture,"{'mixture': 0.9358986801938234, 'uniform': 0.3092815616297828, 'top1': 0.9358991276509686, 'top2': 0.9358986801938234, 'top3': 0.9358986801938234}",0.8717971151028363,0.07333970560102024,0.006264443101983306,0.9488361135853891,0.9455641796027374,ensemble,1.0,
+4,268,0.2,106.7,"[0.2887129617740495, 7.146374179864233e-07, 0.02675218913568767, 0.9341140653396488]",,1398805,243,False,,,,"[5.572085136061948e-05, 3.0324375408485426e-06, 1.9390704386544167e-06, 0.99993930764066]",mixture,"{'mixture': 0.9341141124811376, 'uniform': 0.2974809266161287, 'top1': 0.9341140653396488, 'top2': 0.9341141124811376, 'top3': 0.9341141124811376}",0.8609243697760304,0.0737337951427876,0.006416487514190574,0.9379381073761351,0.9288095752421666,ensemble,1.0,
+5,335,0.2,130.5,"[0.2851837942410718, 5.709202892510553e-07, 0.02400631314910587, 0.9346082435626396]",,1751052,304,False,,,,"[5.352363032039484e-05, 3.017491387355586e-06, 1.8772546836238203e-06, 0.9999415816236087]",mixture,"{'mixture': 0.934608280971694, 'uniform': 0.3136637816728857, 'top1': 0.9346082435626396, 'top2': 0.934608280971694, 'top3': 0.934608280971694}",0.8758075852599518,0.06919363862118208,0.005191076904368195,0.9435884035290092,0.9365794422929289,ensemble,1.0,
+6,402,0.2,155.8,"[0.28191093592370076, 4.857738693491748e-07, 0.022959482419344, 0.9338568269067726]",,2057337,357,False,,,,"[5.219100403170781e-05, 3.040252132571167e-06, 1.8717196545786043e-06, 0.9999428970241812]",mixture,"{'mixture': 0.9338566313420587, 'uniform': 0.32391720725548095, 'top1': 0.9338568269067726, 'top2': 0.9338566313420587, 'top3': 0.9338566313420587}",0.8795111878237105,0.063227044280279,0.00398630717600138,0.9342727392589453,0.9294967729090275,ensemble,1.0,
+7,402,0.0,243.8,,,,,,,,,,,,0.884373415141364,0.06514450746984593,0.004541055743000292,,,,,True

macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-laplace/budget-5/seed-1/log.txt

@@ -0,0 +1,103 @@
+[23:14:51][      0s][START] target=multiclass variant=macs-v11-norestart-acq-laplace budget=5.0% seed=1 | {"task": "mitochondria", "target": "multiclass", "variant": "macs-v11-norestart-acq-laplace", "budget": 5.0, "seed": 1, "sources": "", "run_tag": "", "arch": "unet", "teacher_ckpt": "model_best.pth", "device": "cuda:0", "num_workers": 8, "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}
+[23:14:51][      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.
+[23:14:51][      0s][DATA] multiclass: train=1855 val=350 test=175 patches
+[23:14:55][      4s][CALIB] source-val temperatures: fly-mito=1.38, kasthuri=2.51, lucchi=3.80, dros-vnc=2.77
+[23:14:55][      4s][FEATURES] 20 domain x encoder feature sets (20 cached) | n=512/domain | 1s
+[23:14:57][      6s][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
+[23:14:57][      6s][WEIGHTS] per-source: fly-mito=0.014, kasthuri=0.019, lucchi=0.962, dros-vnc=0.005 | min-MMD source (NeuroADDA rule) = lucchi
+[23:15:10][     19s][TEACHERS] target-val Dice: fly-mito=0.258, kasthuri=0.175, lucchi=0.968, dros-vnc=0.176
+[23:15:10][     19s][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
+[23:15:10][     19s][WEIGHTS] using mixture: fly-mito=0.012, kasthuri=0.012, lucchi=0.968, dros-vnc=0.008
+[23:15:10][     19s][INIT] student initialised from teacher lucchi (weight 0.968)
+[23:15:13][     22s][ADABN] init: BatchNorm statistics re-estimated on 512 unlabeled target patches
+[23:15:13][     22s][KD] distilling the mixture ensemble (prob space) into the student
+[23:15:50][     60s][KD] soft labels for 1855 patches from 4 teachers in 37s
+[23:16:45][    114s][KD] epoch 1/5 | patches=1855 | loss=0.1031 | 55s
+[23:17:40][    169s][KD] epoch 2/5 | patches=1855 | loss=0.0736 | 55s
+[23:18:35][    224s][KD] epoch 3/5 | patches=1855 | loss=0.0710 | 55s
+[23:19:30][    279s][KD] epoch 4/5 | patches=1855 | loss=0.0699 | 55s
+[23:20:25][    334s][KD] epoch 5/5 | patches=1855 | loss=0.0687 | 55s
+[23:20:27][    336s][ROUND0] before any target label: val_dice=0.9624 val_vi=0.0434
+[23:20:27][    336s][AL] patches=1855 (pool 1855 after dropping blank; free blank negatives: True) | budget=92 (5.0%) | rounds=6 | per_round=15 | acquisition=laplace | first_round=stratified
+[23:20:47][    356s][AL] round-1 uncertainty quality on 1855 patches: {'spearman_uncertainty_error': -0.14785451464746693, 'auroc_top20_error': 0.2738500882731163}
+[23:21:26][    395s][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
+[23:21:27][    396s][WEIGHTS] round 1: labeled Dice of candidate ensembles: mixture=0.968, uniform=0.178, top1=0.968, top2=0.968, top3=0.968 -> using mixture
+[23:21:28][    397s][WEIGHTS] round 1: student kept (labeled Dice 0.966 vs best teacher lucchi 0.968); distilling the corrected ensemble into it
+[23:21:28][    397s][WEIGHTS] round 1: teacher Dice on 15 labeled patches + 0 blank (73632 fg px): fly-mito=0.294, kasthuri=0.000, lucchi=0.968, dros-vnc=0.018 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000
+[23:22:05][    434s][KD] soft labels for 1840 patches from 4 teachers in 38s
+[23:23:00][    489s][KD] epoch 1/2 | patches=1840 | loss=0.0453 | 55s
+[23:23:55][    544s][KD] epoch 2/2 | patches=1840 | loss=0.0448 | 55s
+[23:24:33][    582s][CONS] pseudo-labels for 1840 unlabeled patches from the current ensemble in 38s
+[23:24:34][    583s][FT] round 1/6 epoch 1/5 | patches=15 | loss=0.2333
+[23:24:36][    585s][FT] round 1/6 epoch 2/5 | patches=15 | loss=0.1945
+[23:24:37][    586s][FT] round 1/6 epoch 3/5 | patches=15 | loss=0.1615
+[23:24:38][    588s][FT] round 1/6 epoch 4/5 | patches=15 | loss=0.1698
+[23:24:40][    589s][FT] round 1/6 epoch 5/5 | patches=15 | loss=0.1832
+[23:24:42][    591s][AL] round 1/6 | labeled=15 | selection_seconds=58.3000 round_seconds=252.7000 restart_student_dice=0.9665 restart_teacher_dice=0.9683 val_dice=0.9368 val_vi=0.0490 val_ece=0.0022
+[23:25:02][    611s][AL] round 2: 272/1840 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:25:05][    614s][WEIGHTS] round 2: labeled Dice of candidate ensembles: mixture=0.968, uniform=0.130, top1=0.968, top2=0.968, top3=0.968 -> using mixture
+[23:25:05][    614s][WEIGHTS] round 2: teacher Dice on 30 labeled patches + 0 blank (155821 fg px): fly-mito=0.300, kasthuri=0.000, lucchi=0.968, dros-vnc=0.011 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000
+[23:25:07][    616s][CONS] round 2: on the 15 new patches student Dice=0.959 vs ensemble 0.968 -> pseudo-labels from the ensemble | consistency x1.00
+[23:25:09][    618s][FT] round 2/6 epoch 1/5 | patches=30 | loss=0.1966
+[23:25:11][    621s][FT] round 2/6 epoch 2/5 | patches=30 | loss=0.1494
+[23:25:14][    623s][FT] round 2/6 epoch 3/5 | patches=30 | loss=0.1582
+[23:25:16][    625s][FT] round 2/6 epoch 4/5 | patches=30 | loss=0.1536
+[23:25:19][    628s][FT] round 2/6 epoch 5/5 | patches=30 | loss=0.1538
+[23:25:21][    630s][AL] round 2/6 | labeled=30 | selection_seconds=20.2000 round_seconds=36.4000 pseudo_student_dice=0.9591 pseudo_ensemble_dice=0.9683 consistency_scale=1.0000 val_dice=0.9536 val_vi=0.0476 val_ece=0.0014
+[23:25:40][    650s][AL] round 3: 270/1825 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:25:45][    654s][WEIGHTS] round 3: labeled Dice of candidate ensembles: mixture=0.966, uniform=0.120, top1=0.966, top2=0.966, top3=0.966 -> using mixture
+[23:25:45][    654s][WEIGHTS] round 3: teacher Dice on 45 labeled patches + 0 blank (240801 fg px): fly-mito=0.311, kasthuri=0.000, lucchi=0.966, dros-vnc=0.019 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000
+[23:25:46][    655s][CONS] round 3: on the 15 new patches student Dice=0.966 vs ensemble 0.962 -> pseudo-labels from the ensemble | consistency x1.00
+[23:25:49][    658s][FT] round 3/6 epoch 1/5 | patches=45 | loss=0.1997
+[23:25:53][    662s][FT] round 3/6 epoch 2/5 | patches=45 | loss=0.1722
+[23:25:56][    665s][FT] round 3/6 epoch 3/5 | patches=45 | loss=0.1707
+[23:25:59][    668s][FT] round 3/6 epoch 4/5 | patches=45 | loss=0.1597
+[23:26:02][    671s][FT] round 3/6 epoch 5/5 | patches=45 | loss=0.1385
+[23:26:05][    674s][AL] round 3/6 | labeled=45 | selection_seconds=20.0000 round_seconds=41.7000 pseudo_student_dice=0.9658 pseudo_ensemble_dice=0.9617 consistency_scale=1.0000 val_dice=0.9543 val_vi=0.0476 val_ece=0.0014
+[23:26:24][    693s][AL] round 4: 282/1810 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:26:29][    699s][WEIGHTS] round 4: labeled Dice of candidate ensembles: mixture=0.966, uniform=0.148, top1=0.966, top2=0.966, top3=0.966 -> using mixture
+[23:26:29][    699s][WEIGHTS] round 4: teacher Dice on 60 labeled patches + 0 blank (328081 fg px): fly-mito=0.318, kasthuri=0.000, lucchi=0.966, dros-vnc=0.033 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000
+[23:26:31][    700s][CONS] round 4: on the 15 new patches student Dice=0.971 vs ensemble 0.966 -> pseudo-labels from the ensemble | consistency x1.00
+[23:26:35][    704s][FT] round 4/6 epoch 1/5 | patches=60 | loss=0.1675
+[23:26:39][    708s][FT] round 4/6 epoch 2/5 | patches=60 | loss=0.1548
+[23:26:43][    713s][FT] round 4/6 epoch 3/5 | patches=60 | loss=0.1342
+[23:26:48][    717s][FT] round 4/6 epoch 4/5 | patches=60 | loss=0.1252
+[23:26:52][    721s][FT] round 4/6 epoch 5/5 | patches=60 | loss=0.1480
+[23:26:54][    723s][AL] round 4/6 | labeled=60 | selection_seconds=19.8000 round_seconds=47.3000 pseudo_student_dice=0.9707 pseudo_ensemble_dice=0.9663 consistency_scale=1.0000 val_dice=0.9217 val_vi=0.0488 val_ece=0.0021
+[23:27:13][    743s][AL] round 5: 255/1795 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:27:20][    749s][WEIGHTS] round 5: labeled Dice of candidate ensembles: mixture=0.964, uniform=0.157, top1=0.964, top2=0.964, top3=0.964 -> using mixture
+[23:27:20][    749s][WEIGHTS] round 5: teacher Dice on 75 labeled patches + 0 blank (412459 fg px): fly-mito=0.323, kasthuri=0.000, lucchi=0.964, dros-vnc=0.033 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000
+[23:27:21][    750s][CONS] round 5: on the 15 new patches student Dice=0.965 vs ensemble 0.955 -> pseudo-labels from the ensemble | consistency x1.00
+[23:27:26][    755s][FT] round 5/6 epoch 1/5 | patches=75 | loss=0.1604
+[23:27:32][    761s][FT] round 5/6 epoch 2/5 | patches=75 | loss=0.1526
+[23:27:37][    766s][FT] round 5/6 epoch 3/5 | patches=75 | loss=0.1478
+[23:27:42][    771s][FT] round 5/6 epoch 4/5 | patches=75 | loss=0.1327
+[23:27:47][    776s][FT] round 5/6 epoch 5/5 | patches=75 | loss=0.1382
+[23:27:49][    778s][AL] round 5/6 | labeled=75 | selection_seconds=19.7000 round_seconds=53.0000 pseudo_student_dice=0.9648 pseudo_ensemble_dice=0.9555 consistency_scale=1.0000 val_dice=0.9241 val_vi=0.0486 val_ece=0.0021
+[23:28:09][    798s][AL] round 6: 270/1780 candidates skipped as agreed-empty (predicted prevalence < 0.005)
+[23:28:16][    805s][WEIGHTS] round 6: labeled Dice of candidate ensembles: mixture=0.963, uniform=0.160, top1=0.963, top2=0.963, top3=0.963 -> using mixture
+[23:28:16][    805s][WEIGHTS] round 6: teacher Dice on 90 labeled patches + 0 blank (495033 fg px): fly-mito=0.313, kasthuri=0.000, lucchi=0.963, dros-vnc=0.033 | weights: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000
+[23:28:18][    807s][CONS] round 6: on the 15 new patches student Dice=0.965 vs ensemble 0.961 -> pseudo-labels from the ensemble | consistency x1.00
+[23:28:24][    813s][FT] round 6/6 epoch 1/5 | patches=90 | loss=0.1348
+[23:28:30][    819s][FT] round 6/6 epoch 2/5 | patches=90 | loss=0.1244
+[23:28:36][    825s][FT] round 6/6 epoch 3/5 | patches=90 | loss=0.1252
+[23:28:42][    831s][FT] round 6/6 epoch 4/5 | patches=90 | loss=0.1216
+[23:28:48][    837s][FT] round 6/6 epoch 5/5 | patches=90 | loss=0.1199
+[23:28:50][    839s][AL] round 6/6 | labeled=90 | selection_seconds=19.7000 round_seconds=58.5000 pseudo_student_dice=0.9648 pseudo_ensemble_dice=0.9607 consistency_scale=1.0000 val_dice=0.9269 val_vi=0.0477 val_ece=0.0019
+[23:28:56][    845s][FT] final 1/10 epoch 1/1 | patches=90 | loss=0.1197
+[23:29:02][    851s][FT] final 2/10 epoch 1/1 | patches=90 | loss=0.1198
+[23:29:08][    857s][FT] final 3/10 epoch 1/1 | patches=90 | loss=0.1106
+[23:29:14][    863s][FT] final 4/10 epoch 1/1 | patches=90 | loss=0.1230
+[23:29:20][    869s][FT] final 5/10 epoch 1/1 | patches=90 | loss=0.1152
+[23:29:26][    875s][FT] final 6/10 epoch 1/1 | patches=90 | loss=0.1010
+[23:29:32][    881s][FT] final 7/10 epoch 1/1 | patches=90 | loss=0.1047
+[23:29:38][    887s][FT] final 8/10 epoch 1/1 | patches=90 | loss=0.1098
+[23:29:44][    893s][FT] final 9/10 epoch 1/1 | patches=90 | loss=0.1180
+[23:29:50][    899s][FT] final 10/10 epoch 1/1 | patches=90 | loss=0.0981
+[23:29:52][    901s][AL] final consolidation (10 epochs, cosine lr) | selection_seconds=0.0000 round_seconds=60.1000 val_dice=0.9394 val_vi=0.0461 val_ece=0.0019
+[23:29:52][    901s][WEIGHTS] final: fly-mito=0.000, kasthuri=0.000, lucchi=1.000, dros-vnc=0.000
+[23:30:01][    910s][TEST] raw student (no EMA, no TTA): patch_dice=0.9013 patch_vi=0.0835
+[23:30:02][    911s][TEST] final model without TTA: patch_dice=0.9013 patch_vi=0.0835
+[23:30:02][    911s][RESOURCES] {"phases": {"setup": {"seconds": 2.66, "peak_cpu_rss_mb": 2160.1, "peak_gpu_alloc_mb": 592.4, "peak_gpu_reserved_mb": 594.0}, "features_and_weights": {"seconds": 18.41, "peak_cpu_rss_mb": 4176.4, "peak_gpu_alloc_mb": 2298.2, "peak_gpu_reserved_mb": 2640.0}, "distillation": {"seconds": 312.05, "peak_cpu_rss_mb": 4656.7, "peak_gpu_alloc_mb": 3826.7, "peak_gpu_reserved_mb": 4148.0, "kd_epochs": 5}, "active_learning": {"seconds": 564.86, "peak_cpu_rss_mb": 4793.3, "peak_gpu_alloc_mb": 3947.0, "peak_gpu_reserved_mb": 4156.0, "rounds": 6}, "inference": {"seconds": 10.07, "peak_cpu_rss_mb": 4793.3, "peak_gpu_alloc_mb": 2149.5, "peak_gpu_reserved_mb": 4156.0, "patches": 175, "images": 5, "patches_per_second": 17.4, "seconds_per_image": 2.014}}, "peak_gpu_alloc_mb": 3947.0, "peak_cpu_rss_mb": 4793.3, "total_seconds": 908.1}
+[23:30:02][    911s][TEST] patch_dice=0.8905 patch_iou=0.8441 patch_precision=0.9614 patch_recall=0.8780 patch_vi=0.0924 image_dice=0.9038 image_iou=0.8275 image_precision=0.9575 image_recall=0.8651 image_vi=0.1166 cal_ece=0.0076 cal_nll=0.0696 cal_brier=0.0107 cal_mean_confidence=0.9948 cal_accuracy=0.9872 cal_predicted_prevalence=0.0829 cal_true_prevalence=0.0886 cal_mean_probability=0.0827
+[23:30:03][    912s][DONE] multiclass macs-v11-norestart-acq-laplace b5 s1 | test patch Dice=0.8905 VI=0.0924 | image Dice=0.9038 VI=0.1166 | 911s

macs_v2_r26/runs/mitochondria/multiclass/macs-v11-norestart-acq-laplace/budget-5/seed-1/result.json

@@ -0,0 +1,647 @@
+{
+  "target": "multiclass",
+  "variant": "macs-v11-norestart-acq-laplace",
+  "budget": 5.0,
+  "seed": 1,
+  "config": {
+    "task": "mitochondria",
+    "target": "multiclass",
+    "variant": "macs-v11-norestart-acq-laplace",
+    "budget": 5.0,
+    "seed": 1,
+    "sources": "",
+    "run_tag": "",
+    "arch": "unet",
+    "teacher_ckpt": "model_best.pth",
+    "models_dir": "/mnt/sdc1/connectomics/MACS-2/models",
+    "cache_dir": "/mnt/sdc1/connectomics/MACS-2/cache/macs_v2",
+    "output_dir": "experiments/macs_v2/outputs-r26",
+    "checkpoint_dir": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2",
+    "device": "cuda:0",
+    "num_workers": 8,
+    "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": "fce9b15",
+  "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": "bayzid-system-2",
+    "python": "3.12.11",
+    "torch": "2.5.1+cu118",
+    "argv": [
+      "experiments/macs_v2/run.py",
+      "sweep",
+      "--targets",
+      "multiclass",
+      "--variants",
+      "macs-v11-norestart-acq-laplace",
+      "--budgets",
+      "5",
+      "--seeds",
+      "1",
+      "--num-workers",
+      "8",
+      "--output-dir",
+      "experiments/macs_v2/outputs-v3",
+      "--task",
+      "mitochondria",
+      "--output-dir",
+      "experiments/macs_v2/outputs-r26"
+    ],
+    "cwd": "/mnt/sdc1/connectomics/MACS-2",
+    "gpu": "NVIDIA GeForce RTX 3070",
+    "cuda": "11.8",
+    "commit": "fce9b15",
+    "dirty": true
+  },
+  "sources": [
+    "fly-brain-mito-benchmark",
+    "kasthuri-pp",
+    "lucchi-pp",
+    "drosophila-vnc"
+  ],
+  "alphas": [
+    0.01152117817089438,
+    0.011843492157429523,
+    0.9683960236342841,
+    0.008239306037391987
+  ],
+  "weights_all": {
+    "sources": [
+      "fly-brain-mito-benchmark",
+      "kasthuri-pp",
+      "lucchi-pp",
+      "drosophila-vnc"
+    ],
+    "mixture": [
+      0.01152117817089438,
+      0.011843492157429523,
+      0.9683960236342841,
+      0.008239306037391987
+    ],
+    "per_source": [
+      0.014321460439305007,
+      0.018762755437246332,
+      0.9619826418368663,
+      0.0049331422865822525
+    ],
+    "uniform": [
+      0.25,
+      0.25,
+      0.25,
+      0.25
+    ],
+    "mixture_info": {
+      "objective": 0.004128238346556883,
+      "mmd2_mixture": 0.005874251861260116,
+      "mmd2_uniform": 0.2484069500788153,
+      "mmd2_single_source": [
+        0.6121996102822926,
+        0.5667987013145535,
+        0.00484548842698207,
+        0.6689345615903355
+      ],
+      "median_squared_distance": 0.20474058073505946,
+      "gammas": [
+        1.2210573942031921,
+        2.4421147884063843,
+        4.884229576812769,
+        9.768459153625537,
+        19.536918307251074
+      ],
+      "max_weight": 0.9683960236342841,
+      "entropy": 0.17460135147032324,
+      "effective_sources": 1.1907714227038573,
+      "min_weight": 0.008239306037391987
+    },
+    "per_source_info": {
+      "mmd2_own_rkhs": [
+        0.6529838206918517,
+        0.6112545706593259,
+        0.0030196534926445473,
+        0.8176336137299548
+      ],
+      "standardized": [
+        0.426446119561544,
+        0.2913887851649038,
+        -1.6771725126169683,
+        0.9593376078905211
+      ],
+      "tau": 0.5,
+      "max_weight": 0.9619826418368663,
+      "entropy": 0.19889648186407677,
+      "effective_sources": 1.2200556614747875,
+      "min_weight": 0.0049331422865822525
+    },
+    "lam": 0.01,
+    "tau": 0.5,
+    "single_mmd_source": "lucchi-pp",
+    "seconds": 1.9
+  },
+  "teacher_diagnostics": {
+    "teacher_val_dice": [
+      0.25842427994689,
+      0.17499716869513837,
+      0.9675528820923397,
+      0.17625340082547544
+    ],
+    "ensemble_val_dice": {
+      "mixture": 0.9673725087302072,
+      "mixture@raw": 0.9675092802728925,
+      "mixture@logit": 0.9667406060014452,
+      "per_source": 0.9673224803379604,
+      "per_source@raw": 0.9674927929469518,
+      "per_source@logit": 0.9664724666731698,
+      "uniform": 0.2475083000162392,
+      "uniform@raw": 0.20821212102935532,
+      "uniform@logit": 0.19508060271550287
+    },
+    "val_patches": 350
+  },
+  "temperatures": [
+    1.3822230100631714,
+    2.5090832710266113,
+    3.80090069770813,
+    2.767641305923462
+  ],
+  "top_source": "lucchi-pp",
+  "weight_stats": {
+    "max_weight": 0.9683960236342841,
+    "entropy": 0.17460135147032324,
+    "effective_sources": 1.1907714227038573,
+    "min_weight": 0.008239306037391987
+  },
+  "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": 312.0,
+  "round0_val": {
+    "val_dice": 0.9623507293600193,
+    "val_vi": 0.043359066853027754,
+    "val_ece": 0.009826597372024078
+  },
+  "rounds": [
+    {
+      "round": 1,
+      "n_labeled": 15,
+      "selection_seconds": 58.3,
+      "round_seconds": 252.7,
+      "teacher_dice_labeled": [
+        0.2938186560944717,
+        1.358086727418413e-05,
+        0.9683060184264144,
+        0.01782471245039349
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 73632,
+      "evidence_nonempty_patches": 14,
+      "first_correction": true,
+      "restart": "kept",
+      "restart_student_dice": 0.9664782881707965,
+      "restart_teacher_dice": 0.9683060184264144,
+      "weights": [
+        1.4001540225714672e-05,
+        7.623939953124306e-07,
+        0.9999846022791299,
+        6.337866489948326e-07
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9683060184264144,
+        "uniform": 0.1778269851676831,
+        "top1": 0.9683060184264144,
+        "top2": 0.9683060184264144,
+        "top3": 0.9683060184264144
+      },
+      "val_dice": 0.9367825440230815,
+      "val_vi": 0.04902821843243505,
+      "val_ece": 0.002242055302164851
+    },
+    {
+      "round": 2,
+      "n_labeled": 30,
+      "selection_seconds": 20.2,
+      "round_seconds": 36.4,
+      "teacher_dice_labeled": [
+        0.30025932706211406,
+        6.417579032485785e-06,
+        0.9682767700219854,
+        0.011387269224389809
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 155821,
+      "evidence_nonempty_patches": 29,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.493736268096847e-05,
+        7.625617057448709e-07,
+        0.9999837056298168,
+        5.944457965627549e-07
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9682767700219854,
+        "uniform": 0.12951527045112196,
+        "top1": 0.9682767700219854,
+        "top2": 0.9682767700219854,
+        "top3": 0.9682767700219854
+      },
+      "pseudo_student_dice": 0.9590947745807612,
+      "pseudo_ensemble_dice": 0.968250747149762,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.9536202317207382,
+      "val_vi": 0.04758055893308673,
+      "val_ece": 0.0013883460739741694
+    },
+    {
+      "round": 3,
+      "n_labeled": 45,
+      "selection_seconds": 20.0,
+      "round_seconds": 41.7,
+      "teacher_dice_labeled": [
+        0.31087552052835477,
+        4.15278942865923e-06,
+        0.965947077844261,
+        0.018605052939869852
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 240801,
+      "evidence_nonempty_patches": 43,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.7001853221602358e-05,
+        7.805162522607558e-07,
+        0.9999815636339252,
+        6.539966008991305e-07
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.965947077844261,
+        "uniform": 0.11951632802861115,
+        "top1": 0.965947077844261,
+        "top2": 0.965947077844261,
+        "top3": 0.965947077844261
+      },
+      "pseudo_student_dice": 0.9657906130425116,
+      "pseudo_ensemble_dice": 0.9616518292208995,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.954270906544748,
+      "val_vi": 0.047638874112841144,
+      "val_ece": 0.0013604008995015291
+    },
+    {
+      "round": 4,
+      "n_labeled": 60,
+      "selection_seconds": 19.8,
+      "round_seconds": 47.3,
+      "teacher_dice_labeled": [
+        0.317738364665974,
+        3.048018483184082e-06,
+        0.9660351508741329,
+        0.033080268430627946
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 328081,
+      "evidence_nonempty_patches": 58,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.819357973396529e-05,
+        7.798195075771067e-07,
+        0.9999802714085584,
+        7.551922000094123e-07
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9660351508741329,
+        "uniform": 0.14769137834183466,
+        "top1": 0.9660351508741329,
+        "top2": 0.9660351508741329,
+        "top3": 0.9660351508741329
+      },
+      "pseudo_student_dice": 0.9706596671377888,
+      "pseudo_ensemble_dice": 0.9662802457660312,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.9216831759004065,
+      "val_vi": 0.04882421978576393,
+      "val_ece": 0.0020669694208124826
+    },
+    {
+      "round": 5,
+      "n_labeled": 75,
+      "selection_seconds": 19.7,
+      "round_seconds": 53.0,
+      "teacher_dice_labeled": [
+        0.32291000827099386,
+        2.4244775250933425e-06,
+        0.9639021682139222,
+        0.03308238053915913
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 412459,
+      "evidence_nonempty_patches": 73,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.957226710843179e-05,
+        7.96625492397049e-07,
+        0.9999788596188555,
+        7.714885436697631e-07
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9639021682139222,
+        "uniform": 0.156949700850309,
+        "top1": 0.9639021682139222,
+        "top2": 0.9639021682139222,
+        "top3": 0.9639021682139222
+      },
+      "pseudo_student_dice": 0.964805861328671,
+      "pseudo_ensemble_dice": 0.955485177998267,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.9241157053196762,
+      "val_vi": 0.04862066420187986,
+      "val_ece": 0.0020597238850812863
+    },
+    {
+      "round": 6,
+      "n_labeled": 90,
+      "selection_seconds": 19.7,
+      "round_seconds": 58.5,
+      "teacher_dice_labeled": [
+        0.31313687709842153,
+        2.0200632683815656e-06,
+        0.9633628632939749,
+        0.03338318269293788
+      ],
+      "teacher_dice_pool": null,
+      "evidence_fg_pixels": 495033,
+      "evidence_nonempty_patches": 88,
+      "first_correction": false,
+      "restart": "",
+      "weights": [
+        1.7845958603297553e-05,
+        8.009314735799591e-07,
+        0.9999805751114135,
+        7.779985096301431e-07
+      ],
+      "ensemble_choice": "mixture",
+      "candidate_dice": {
+        "mixture": 0.9633628632939749,
+        "uniform": 0.15983339962159931,
+        "top1": 0.9633628632939749,
+        "top2": 0.9633628632939749,
+        "top3": 0.9633628632939749
+      },
+      "pseudo_student_dice": 0.9647581094503258,
+      "pseudo_ensemble_dice": 0.9606947095384917,
+      "pseudo_source": "ensemble",
+      "consistency_scale": 1.0,
+      "val_dice": 0.9269402183550002,
+      "val_vi": 0.04773342905549056,
+      "val_ece": 0.0019010978224997363
+    },
+    {
+      "round": 7,
+      "n_labeled": 90,
+      "selection_seconds": 0.0,
+      "round_seconds": 60.1,
+      "final_consolidation": true,
+      "val_dice": 0.939400585836005,
+      "val_vi": 0.04612817190652949,
+      "val_ece": 0.001865632105289674
+    }
+  ],
+  "uncertainty_quality": {
+    "spearman_uncertainty_error": -0.14785451464746693,
+    "auroc_top20_error": 0.2738500882731163
+  },
+  "al_seconds": 564.9,
+  "source_archs": [
+    "unet",
+    "unet",
+    "unet",
+    "unet"
+  ],
+  "alphas_final": [
+    1.7845958603297553e-05,
+    8.009314735799591e-07,
+    0.9999805751114135,
+    7.779985096301431e-07
+  ],
+  "weight_stats_final": {
+    "max_weight": 0.9999805751114135,
+    "entropy": 0.00023673447107835246,
+    "effective_sources": 1.0002367624948947,
+    "min_weight": 7.779985096301431e-07
+  },
+  "top_source_final": "lucchi-pp",
+  "ensemble_choice_final": "",
+  "restart": "kept",
+  "test_raw_student": {
+    "patch_dice": 0.9013441843656629,
+    "patch_iou": 0.8612735447008448,
+    "patch_precision": 0.9643553459614856,
+    "patch_recall": 0.8939486028232443,
+    "patch_vi": 0.08351672149892893,
+    "n_patches": 175,
+    "cal_ece": 0.008238096063656893,
+    "cal_nll": 0.07850860294000966,
+    "cal_brier": 0.01023246901159236,
+    "cal_mean_confidence": 0.9966647752568322,
+    "cal_accuracy": 0.9884933035714286,
+    "cal_predicted_prevalence": 0.0837890625,
+    "cal_true_prevalence": 0.08861049107142857,
+    "cal_mean_probability": 0.08348922600161458
+  },
+  "test_notta": {
+    "patch_dice": 0.9013441843656629,
+    "patch_iou": 0.8612735447008448,
+    "patch_precision": 0.9643553459614856,
+    "patch_recall": 0.8939486028232443,
+    "patch_vi": 0.08351672149892893,
+    "n_patches": 175,
+    "cal_ece": 0.008238096063656893,
+    "cal_nll": 0.07850860294000966,
+    "cal_brier": 0.01023246901159236,
+    "cal_mean_confidence": 0.9966647752568322,
+    "cal_accuracy": 0.9884933035714286,
+    "cal_predicted_prevalence": 0.0837890625,
+    "cal_true_prevalence": 0.08861049107142857,
+    "cal_mean_probability": 0.08348922600161458
+  },
+  "test": {
+    "patch_dice": 0.8904517688434722,
+    "patch_iou": 0.8440724660048682,
+    "patch_precision": 0.9614157225591229,
+    "patch_recall": 0.8779824872614105,
+    "patch_vi": 0.09238620068438104,
+    "n_patches": 175,
+    "image_dice": 0.9037671893357683,
+    "image_iou": 0.827489132989274,
+    "image_precision": 0.9575323651068441,
+    "image_recall": 0.8650652426532572,
+    "image_vi": 0.11661794111904884,
+    "n_images": 5,
+    "cal_ece": 0.0076183893408824175,
+    "cal_nll": 0.06957390569200572,
+    "cal_brier": 0.01065475153991808,
+    "cal_mean_confidence": 0.9947722827666771,
+    "cal_accuracy": 0.9871819196428572,
+    "cal_predicted_prevalence": 0.08293526785714286,
+    "cal_true_prevalence": 0.08861049107142857,
+    "cal_mean_probability": 0.08272788942355813
+  },
+  "eval_seconds": 10.1,
+  "total_seconds": 911.3,
+  "resources": {
+    "phases": {
+      "setup": {
+        "seconds": 2.66,
+        "peak_cpu_rss_mb": 2160.1,
+        "peak_gpu_alloc_mb": 592.4,
+        "peak_gpu_reserved_mb": 594.0
+      },
+      "features_and_weights": {
+        "seconds": 18.41,
+        "peak_cpu_rss_mb": 4176.4,
+        "peak_gpu_alloc_mb": 2298.2,
+        "peak_gpu_reserved_mb": 2640.0
+      },
+      "distillation": {
+        "seconds": 312.05,
+        "peak_cpu_rss_mb": 4656.7,
+        "peak_gpu_alloc_mb": 3826.7,
+        "peak_gpu_reserved_mb": 4148.0,
+        "kd_epochs": 5
+      },
+      "active_learning": {
+        "seconds": 564.86,
+        "peak_cpu_rss_mb": 4793.3,
+        "peak_gpu_alloc_mb": 3947.0,
+        "peak_gpu_reserved_mb": 4156.0,
+        "rounds": 6
+      },
+      "inference": {
+        "seconds": 10.07,
+        "peak_cpu_rss_mb": 4793.3,
+        "peak_gpu_alloc_mb": 2149.5,
+        "peak_gpu_reserved_mb": 4156.0,
+        "patches": 175,
+        "images": 5,
+        "patches_per_second": 17.4,
+        "seconds_per_image": 2.014
+      }
+    },
+    "peak_gpu_alloc_mb": 3947.0,
+    "peak_cpu_rss_mb": 4793.3,
+    "total_seconds": 908.1
+  },
+  "student_parameters": 31042369,
+  "teacher_parameters": [
+    31042369,
+    31042369,
+    31042369,
+    31042369
+  ],
+  "checkpoint": "/mnt/sdc1/connectomics/MACS-2/models/experiments/macs_v2/multiclass/macs-v11-norestart-acq-laplace_b5_s1.pth"
+}