Upload folder using huggingface_hub

fes_cond_mask50_320/jepa/fes.log

@@ -0,0 +1,41 @@
+
+Target temperature 320 K | TICA lag 10 ns = 10 frames, dim 4
+Rollout 20 x 100 steps of 1 ns, last 10 frames as equilibrium samples
+
+/mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/platito/platito/data/mdcath_dataset.py:239: UserWarning: The given NumPy array is not writable, and PyTorch does not support non-writable tensors. This means writing to this tensor will result in undefined behavior. You may want to copy the array to protect its data or make it writable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at /__w/pytorch/pytorch/torch/csrc/utils/tensor_numpy.cpp:213.)
+  return torch.from_numpy(np.ascontiguousarray(arr[indices])).float()
+1balA00 (L=51): 2210 reference frames, 1176 pair features, CA-CA=0.384 nm, Rg=1.07 nm
+    [floor]      MAE=  0.938  RMSE=  1.352  Coverage= 0.637  CA-CA= 0.384   <- reference vs itself; no model can beat this
+    jepa_rollout reusing cached rollout
+    jepa_rollout MAE=  2.312  RMSE=  2.760  Coverage= 0.184  CA-CA= 0.353  Rg=  126%  <-- NON-PHYSICAL
+1em7A00 (L=56): 2220 reference frames, 1431 pair features, CA-CA=0.383 nm, Rg=1.03 nm
+    [floor]      MAE=  0.127  RMSE=  0.226  Coverage= 0.927  CA-CA= 0.383   <- reference vs itself; no model can beat this
+    jepa_rollout reusing cached rollout
+/mnt/iusers01/fatpou01/compsci01/n70579mp/.conda/envs/md/lib/python3.12/site-packages/numpy/_core/fromnumeric.py:3862: RuntimeWarning: Mean of empty slice
+  return _methods._mean(a, axis=axis, dtype=dtype,
+/mnt/iusers01/fatpou01/compsci01/n70579mp/.conda/envs/md/lib/python3.12/site-packages/numpy/_core/_methods.py:142: RuntimeWarning: invalid value encountered in scalar divide
+  ret = ret.dtype.type(ret / rcount)
+    jepa_rollout MAE=    nan  RMSE=    nan  Coverage= 0.000  CA-CA= 0.358  Rg=  130%  <-- NON-PHYSICAL
+1gjsA00 (L=65): 2150 reference frames, 1953 pair features, CA-CA=0.384 nm, Rg=1.22 nm
+    [floor]      MAE=  0.831  RMSE=  1.051  Coverage= 0.771  CA-CA= 0.384   <- reference vs itself; no model can beat this
+    jepa_rollout reusing cached rollout
+    jepa_rollout MAE=  1.251  RMSE=  1.582  Coverage= 0.079  CA-CA= 0.355  Rg=  127%  <-- NON-PHYSICAL
+
+========================================================================
+mean +/- std across 3 mdCATH held-out domains @ 320 K
+========================================================================
+  arm              n               MAE              RMSE          Coverage        CA-CA (nm)
+  jepa_rollout     3        nan+/-nan         nan+/-nan       0.088+/-0.075      0.355+/-0.002   3/3 NON-PHYSICAL
+  reference_floor   3      0.632+/-0.359      0.876+/-0.476      0.778+/-0.119      0.383+/-0.000
+
+Lower MAE/RMSE and higher Coverage are better. Comparable between arms,
+not to the paper's fast-folder table. `reference_floor` is the reference
+scored against itself: the best any model could achieve at this sample size.
+
+CA-CA is the mean consecutive backbone spacing of the SCORED frames.
+Real mdCATH is 0.383 nm (sd 0.008); an arm outside [0.35, 0.42] is flagged NON-PHYSICAL and its
+MAE/RMSE/Coverage should not be compared -- score_nonzero masks to bins the
+model reaches and minimizes over a constant offset, so a uniformly distorted
+ensemble can post a good Coverage without producing proteins.
+
+Wrote /mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/runs/fes_cond_mask50_320/jepa/mdcath_fes.json

fes_cond_mask50_320/jepa/holdout_from_split.yaml

@@ -0,0 +1,3 @@
+- 1balA00
+- 1em7A00
+- 1gjsA00

fes_cond_mask50_320/jepa/mdcath_fes.json

@@ -0,0 +1,97 @@
+{
+  "1balA00": {
+    "length": 51,
+    "n_reference_frames": 2210,
+    "n_pair_features": 1176,
+    "reference_floor": {
+      "mae": 0.937743057160997,
+      "rmse": 1.3522564780819946,
+      "coverage": 0.6366734832992502,
+      "ca_ca_nm": 0.3835442286064416
+    },
+    "reference_ca_ca_nm": 0.3835442286064416,
+    "jepa_rollout": {
+      "mae": 2.3122283123285863,
+      "rmse": 2.75981976528379,
+      "coverage": 0.1835278858625162,
+      "ca_ca_nm": 0.3528202835483495,
+      "rg_nm": 1.356909042166069,
+      "rg_frac": 1.2630627439972804,
+      "geometry_valid": false
+    }
+  },
+  "1em7A00": {
+    "length": 56,
+    "n_reference_frames": 2220,
+    "n_pair_features": 1431,
+    "reference_floor": {
+      "mae": 0.12737728299172452,
+      "rmse": 0.22566002535153618,
+      "coverage": 0.9274084124830394,
+      "ca_ca_nm": 0.38292506315074654
+    },
+    "reference_ca_ca_nm": 0.38292506315074654,
+    "jepa_rollout": {
+      "mae": NaN,
+      "rmse": NaN,
+      "coverage": 0.0,
+      "ca_ca_nm": 0.3583827841303657,
+      "rg_nm": 1.3430799120711154,
+      "rg_frac": 1.297663538341803,
+      "geometry_valid": false
+    }
+  },
+  "1gjsA00": {
+    "length": 65,
+    "n_reference_frames": 2150,
+    "n_pair_features": 1953,
+    "reference_floor": {
+      "mae": 0.8308398762114027,
+      "rmse": 1.0505297542011245,
+      "coverage": 0.77129750982962,
+      "ca_ca_nm": 0.3837785387567121
+    },
+    "reference_ca_ca_nm": 0.3837785387567121,
+    "jepa_rollout": {
+      "mae": 1.2505835104159726,
+      "rmse": 1.5822507973264202,
+      "coverage": 0.07934238741958542,
+      "ca_ca_nm": 0.35482734278851324,
+      "rg_nm": 1.5541268721969528,
+      "rg_frac": 1.2734020736696419,
+      "geometry_valid": false
+    }
+  },
+  "_summary": {
+    "jepa_rollout": {
+      "n_domains": 3,
+      "n_failed": 0,
+      "mae_mean": NaN,
+      "mae_std": NaN,
+      "rmse_mean": NaN,
+      "rmse_std": NaN,
+      "coverage_mean": 0.0876234244273672,
+      "coverage_std": 0.07515341155589404,
+      "ca_ca_nm_mean": 0.35534347015574275,
+      "ca_ca_nm_std": 0.002300020836954176,
+      "rg_frac_mean": 1.2780427853362417,
+      "rg_frac_std": 0.014501859556562005,
+      "n_non_physical": 3
+    },
+    "reference_floor": {
+      "n_domains": 3,
+      "n_failed": 0,
+      "mae_mean": 0.6319867387880415,
+      "mae_std": 0.35947192721507176,
+      "rmse_mean": 0.8761487525448851,
+      "rmse_std": 0.4761732472746924,
+      "coverage_mean": 0.7784598018706365,
+      "coverage_std": 0.11880003810980404,
+      "ca_ca_nm_mean": 0.38341594350463337,
+      "ca_ca_nm_std": 0.00036004439819187065,
+      "rg_frac_mean": null,
+      "rg_frac_std": null,
+      "n_non_physical": 0
+    }
+  }
+}

fes_cond_mask50_320/jepa/rollout_manifest_jepa_rollout.json

@@ -0,0 +1,35 @@
+{
+  "arm": "jepa_rollout",
+  "mode": "rollout",
+  "feedback": "coord",
+  "stage2_checkpoint": "/mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/runs/stage2_cond_mask50_snapshot.ckpt",
+  "stage3_checkpoint": "/mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/runs/pjepa_stage3_tito_platito_mask50_fmdecay/last_tito_mask50_fmdecay.ckpt",
+  "temperature": "320",
+  "n_traj": 20,
+  "n_steps": 100,
+  "step_ns": 1,
+  "ode_steps": 50,
+  "seed": 12345,
+  "renorm_pred": true,
+  "rescale_rg": false,
+  "window": 2,
+  "stage2_flow": true,
+  "stage2_ode_steps": 10,
+  "stage2_ode_method": "heun",
+  "stage2_flow_renorm": "layernorm",
+  "domains": [
+    "1balA00",
+    "1em7A00",
+    "1gjsA00"
+  ],
+  "files": [
+    "/mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/runs/fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1balA00_320K_20x100_step1_ode50_seed12345.npy",
+    "/mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/runs/fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1em7A00_320K_20x100_step1_ode50_seed12345.npy",
+    "/mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/runs/fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1gjsA00_320K_20x100_step1_ode50_seed12345.npy"
+  ],
+  "ca_ca_nm": {
+    "1balA00": 0.3523999750614166,
+    "1em7A00": 0.36201563477516174,
+    "1gjsA00": 0.35668808221817017
+  }
+}

fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1balA00_320K_20x100_step1_ode50_seed12345.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70163dd66e62df14177e1ce58c8c19c02b5c1e93f7297b3dc7d119715591746f
+size 1236368

fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1balA00_320K_20x100_step1_ode50_seed12345.prov.json

@@ -0,0 +1,28 @@
+{
+  "feedback": "coord",
+  "mode": "rollout",
+  "predictor_shape": [
+    192,
+    3,
+    6
+  ],
+  "renorm_pred": true,
+  "rescale_rg": false,
+  "stage2": {
+    "path": "/net/scratch/n70579mp/md/mdjepa/runs/stage2_cond_mask50_snapshot.ckpt",
+    "size": 37406061
+  },
+  "stage2_cond": [
+    "residue_ids",
+    "temp"
+  ],
+  "stage2_flow": true,
+  "stage2_flow_renorm": "layernorm",
+  "stage2_ode_method": "heun",
+  "stage2_ode_steps": 10,
+  "stage3": {
+    "path": "/net/scratch/n70579mp/md/mdjepa/runs/pjepa_stage3_tito_platito_mask50_fmdecay/last_tito_mask50_fmdecay.ckpt",
+    "size": 73149523
+  },
+  "window": 2
+}

fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1em7A00_320K_20x100_step1_ode50_seed12345.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:945fecff6b13b1f855de8bfb4470d375b834e66d4873d506197e8b5532cdfdbc
+size 1357568

fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1em7A00_320K_20x100_step1_ode50_seed12345.prov.json

@@ -0,0 +1,28 @@
+{
+  "feedback": "coord",
+  "mode": "rollout",
+  "predictor_shape": [
+    192,
+    3,
+    6
+  ],
+  "renorm_pred": true,
+  "rescale_rg": false,
+  "stage2": {
+    "path": "/net/scratch/n70579mp/md/mdjepa/runs/stage2_cond_mask50_snapshot.ckpt",
+    "size": 37406061
+  },
+  "stage2_cond": [
+    "residue_ids",
+    "temp"
+  ],
+  "stage2_flow": true,
+  "stage2_flow_renorm": "layernorm",
+  "stage2_ode_method": "heun",
+  "stage2_ode_steps": 10,
+  "stage3": {
+    "path": "/net/scratch/n70579mp/md/mdjepa/runs/pjepa_stage3_tito_platito_mask50_fmdecay/last_tito_mask50_fmdecay.ckpt",
+    "size": 73149523
+  },
+  "window": 2
+}

fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1gjsA00_320K_20x100_step1_ode50_seed12345.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ced57c39bdfc150a639086668bc18642bb4084dc7e9e10566a3d1d59db212d8f
+size 1575728

fes_cond_mask50_320/jepa/rollouts/jepa_rollout__1gjsA00_320K_20x100_step1_ode50_seed12345.prov.json

@@ -0,0 +1,28 @@
+{
+  "feedback": "coord",
+  "mode": "rollout",
+  "predictor_shape": [
+    192,
+    3,
+    6
+  ],
+  "renorm_pred": true,
+  "rescale_rg": false,
+  "stage2": {
+    "path": "/net/scratch/n70579mp/md/mdjepa/runs/stage2_cond_mask50_snapshot.ckpt",
+    "size": 37406061
+  },
+  "stage2_cond": [
+    "residue_ids",
+    "temp"
+  ],
+  "stage2_flow": true,
+  "stage2_flow_renorm": "layernorm",
+  "stage2_ode_method": "heun",
+  "stage2_ode_steps": 10,
+  "stage3": {
+    "path": "/net/scratch/n70579mp/md/mdjepa/runs/pjepa_stage3_tito_platito_mask50_fmdecay/last_tito_mask50_fmdecay.ckpt",
+    "size": 73149523
+  },
+  "window": 2
+}

fes_cond_mask50_320/tito/fes.log

@@ -0,0 +1,36 @@
+
+Target temperature 320 K | TICA lag 10 ns = 10 frames, dim 4
+Rollout 20 x 100 steps of 1 ns, last 10 frames as equilibrium samples
+
+/mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/platito/platito/data/mdcath_dataset.py:239: UserWarning: The given NumPy array is not writable, and PyTorch does not support non-writable tensors. This means writing to this tensor will result in undefined behavior. You may want to copy the array to protect its data or make it writable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at /__w/pytorch/pytorch/torch/csrc/utils/tensor_numpy.cpp:213.)
+  return torch.from_numpy(np.ascontiguousarray(arr[indices])).float()
+1balA00 (L=51): 2210 reference frames, 1176 pair features, CA-CA=0.384 nm, Rg=1.07 nm
+    [floor]      MAE=  0.938  RMSE=  1.352  Coverage= 0.637  CA-CA= 0.384   <- reference vs itself; no model can beat this
+/mnt/iusers01/fatpou01/compsci01/n70579mp/.conda/envs/md/lib/python3.12/site-packages/lightning/pytorch/utilities/parsing.py:209: Attribute 'velocity_net' is an instance of `nn.Module` and is already saved during checkpointing. It is recommended to ignore them using `self.save_hyperparameters(ignore=['velocity_net'])`.
+/mnt/iusers01/fatpou01/compsci01/n70579mp/.conda/envs/md/lib/python3.12/site-packages/lightning/pytorch/utilities/parsing.py:209: Attribute 'condition_net' is an instance of `nn.Module` and is already saved during checkpointing. It is recommended to ignore them using `self.save_hyperparameters(ignore=['condition_net'])`.
+    tito_3M      MAE=  1.848  RMSE=  2.265  Coverage= 0.331  CA-CA= 0.380  Rg=  105%
+1em7A00 (L=56): 2220 reference frames, 1431 pair features, CA-CA=0.383 nm, Rg=1.03 nm
+    [floor]      MAE=  0.127  RMSE=  0.226  Coverage= 0.927  CA-CA= 0.383   <- reference vs itself; no model can beat this
+    tito_3M      MAE=  0.514  RMSE=  0.677  Coverage= 0.966  CA-CA= 0.383  Rg=  101%
+1gjsA00 (L=65): 2150 reference frames, 1953 pair features, CA-CA=0.384 nm, Rg=1.22 nm
+    [floor]      MAE=  0.831  RMSE=  1.051  Coverage= 0.771  CA-CA= 0.384   <- reference vs itself; no model can beat this
+    tito_3M      MAE=  1.450  RMSE=  1.739  Coverage= 0.836  CA-CA= 0.382  Rg=  103%
+
+========================================================================
+mean +/- std across 3 mdCATH held-out domains @ 320 K
+========================================================================
+  arm              n               MAE              RMSE          Coverage        CA-CA (nm)
+  reference_floor   3      0.632+/-0.359      0.876+/-0.476      0.778+/-0.119      0.383+/-0.000
+  tito_3M          3      1.271+/-0.559      1.560+/-0.661      0.711+/-0.274      0.382+/-0.001
+
+Lower MAE/RMSE and higher Coverage are better. Comparable between arms,
+not to the paper's fast-folder table. `reference_floor` is the reference
+scored against itself: the best any model could achieve at this sample size.
+
+CA-CA is the mean consecutive backbone spacing of the SCORED frames.
+Real mdCATH is 0.383 nm (sd 0.008); an arm outside [0.35, 0.42] is flagged NON-PHYSICAL and its
+MAE/RMSE/Coverage should not be compared -- score_nonzero masks to bins the
+model reaches and minimizes over a constant offset, so a uniformly distorted
+ensemble can post a good Coverage without producing proteins.
+
+Wrote /mnt/iusers01/fatpou01/compsci01/n70579mp/scratch/md/mdjepa/runs/fes_cond_mask50_320/tito/mdcath_fes.json

fes_cond_mask50_320/tito/holdout_from_split.yaml

@@ -0,0 +1,3 @@
+- 1balA00
+- 1em7A00
+- 1gjsA00

fes_cond_mask50_320/tito/mdcath_fes.json

@@ -0,0 +1,97 @@
+{
+  "1balA00": {
+    "length": 51,
+    "n_reference_frames": 2210,
+    "n_pair_features": 1176,
+    "reference_floor": {
+      "mae": 0.937743057160997,
+      "rmse": 1.3522564780819946,
+      "coverage": 0.6366734832992502,
+      "ca_ca_nm": 0.3835442286064416
+    },
+    "reference_ca_ca_nm": 0.3835442286064416,
+    "tito_3M": {
+      "mae": 1.84790516021449,
+      "rmse": 2.2649354903776957,
+      "coverage": 0.33073929961089493,
+      "ca_ca_nm": 0.38049063848325987,
+      "rg_nm": 1.1310655278110946,
+      "rg_frac": 1.0528389780034861,
+      "geometry_valid": true
+    }
+  },
+  "1em7A00": {
+    "length": 56,
+    "n_reference_frames": 2220,
+    "n_pair_features": 1431,
+    "reference_floor": {
+      "mae": 0.12737728299172452,
+      "rmse": 0.22566002535153618,
+      "coverage": 0.9274084124830394,
+      "ca_ca_nm": 0.38292506315074654
+    },
+    "reference_ca_ca_nm": 0.38292506315074654,
+    "tito_3M": {
+      "mae": 0.5140662401350976,
+      "rmse": 0.6767793701776513,
+      "coverage": 0.9661870503597122,
+      "ca_ca_nm": 0.38268347034134564,
+      "rg_nm": 1.0459862379721598,
+      "rg_frac": 1.0106161148152988,
+      "geometry_valid": true
+    }
+  },
+  "1gjsA00": {
+    "length": 65,
+    "n_reference_frames": 2150,
+    "n_pair_features": 1953,
+    "reference_floor": {
+      "mae": 0.8308398762114027,
+      "rmse": 1.0505297542011245,
+      "coverage": 0.77129750982962,
+      "ca_ca_nm": 0.3837785387567121
+    },
+    "reference_ca_ca_nm": 0.3837785387567121,
+    "tito_3M": {
+      "mae": 1.4502437084149655,
+      "rmse": 1.7393888905726407,
+      "coverage": 0.8355968548963545,
+      "ca_ca_nm": 0.38235562792708705,
+      "rg_nm": 1.2519071096844319,
+      "rg_frac": 1.0257728233347827,
+      "geometry_valid": true
+    }
+  },
+  "_summary": {
+    "tito_3M": {
+      "n_domains": 3,
+      "n_failed": 0,
+      "mae_mean": 1.2707383695881844,
+      "mae_std": 0.5591351602729153,
+      "rmse_mean": 1.5603679170426625,
+      "rmse_std": 0.6606039394649568,
+      "coverage_mean": 0.7108410682889872,
+      "coverage_std": 0.2740090822119466,
+      "ca_ca_nm_mean": 0.38184324558389754,
+      "ca_ca_nm_std": 0.0009657569185767443,
+      "rg_frac_mean": 1.0297426387178559,
+      "rg_frac_std": 0.017464480516825382,
+      "n_non_physical": 0
+    },
+    "reference_floor": {
+      "n_domains": 3,
+      "n_failed": 0,
+      "mae_mean": 0.6319867387880415,
+      "mae_std": 0.35947192721507176,
+      "rmse_mean": 0.8761487525448851,
+      "rmse_std": 0.4761732472746924,
+      "coverage_mean": 0.7784598018706365,
+      "coverage_std": 0.11880003810980404,
+      "ca_ca_nm_mean": 0.38341594350463337,
+      "ca_ca_nm_std": 0.00036004439819187065,
+      "rg_frac_mean": null,
+      "rg_frac_std": null,
+      "n_non_physical": 0
+    }
+  }
+}

fes_cond_mask50_320/tito/rollouts/tito_3M__1balA00_320K_20x100_step1_ode50_seed12345.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:372288b10730d4f5926f3344f075e540eda82ab30799469e440cba018106a510
+size 1236368

fes_cond_mask50_320/tito/rollouts/tito_3M__1em7A00_320K_20x100_step1_ode50_seed12345.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ede48002c31dbca437bd51ad51dfbc46141eca526b4e00b405fec112c8d09d4
+size 1357568

fes_cond_mask50_320/tito/rollouts/tito_3M__1gjsA00_320K_20x100_step1_ode50_seed12345.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c406b7b9330cad51a187365814b3af05a12cce289e6491841b34c6054373136
+size 1575728