AlonSaguy commited on about 24 hours ago

Commit

94711f6

verified ·

1 Parent(s): afe0db2

Publish 2026_W42_unit

Browse files

Files changed (19) hide show

README.md +29 -22
autoencoder.pt +2 -2
checksums.json +45 -40
conditioning_summary.json +0 -183
config.json +6 -37
global_gmm.joblib → context_weights_subtypes.pt +2 -2
context_weight_model_bundle.pt → context_weights_types.pt +2 -2
ephysatlas_model.json +21 -22
example/expected_predictions.parquet +1 -1
global_gmm_summary.json +0 -37
hierarchy_summary.json +102 -0
knn_bank.npz → members.npz +2 -2
preprocessing/unit_stats.npz +2 -2
readout_summary.json +0 -64
results/summary.json +22 -35
component_feature_expectations.npz → subtype_parent.npy +2 -2
subtypes_gmm.joblib +3 -0
training_summary.json +1348 -0
types_gmm.joblib +3 -0

README.md CHANGED Viewed

@@ -9,27 +9,32 @@ tags:
   - international-brain-lab
 ---
-# Ephys Atlas unit-level interpolation model (2026_W41)
 Predicts **which kinds of neurons you expect to record at a brain position**, and their expected
 waveform phenotype. Trained by the
 [International Brain Laboratory](https://www.internationalbrainlab.com/) on spike-sorted units of
 the Ephys Atlas release `2026_W41` (56497 training units).
-The model has three stages:
 1. a multimodal autoencoder embeds each unit's multi-channel **waveform**, 3D
    **autocorrelogram** and **population coupling** (waveform, acg, stpc) into a
    60-d latent -- its phenotype;
-2. a **25-component Gaussian mixture** over the standardised latents, whose
-   components read as putative cell types. Component shapes are global; the **mixture weights
-   depend on the molecular and cellular context** (PCA volumes of the molecular and cellular
-   atlases) at the unit's position;
-3. a **context-local exemplar readout** maps the model back to 10 interpretable waveform features: `depolarisation_slope`, `recovery_slope`, `repolarisation_slope`, `spatial_spread_um`, `tip_val`, `spike_width_secs`, `predepolarisation_width_secs`, `spike_amplitude`, `peak_to_trough_ratio_log`, `polarity`. At a position, each component is represented by its training members whose molecular context resembles the position's (the 4096 nearest in a 3-d context key, shrunk toward all of the component's members -- the more so the farther the context lies from the training data, and entirely where there is no context). So the context sets the mixture weights **and** selects which training exemplars represent each component; the latent density depends on it only through the weights.
 > **The input is a position.** `predict` takes `x, y, z` (IBL/Allen frame, metres) and returns the
-> expected phenotype there: the local component weights times each component's expected features
-> there. `mixture_weights` returns the local putative cell-type composition itself.
 ## Quickstart
@@ -37,27 +42,29 @@ The model has three stages:
 import pandas as pd
 from ephysatlas import load_pretrained
-model = load_pretrained("int-brain-lab/ea-encoder-unit", revision="2026_W41")
 df = pd.read_parquet("example/positions_sample.parquet")   # positions (x, y, z)
 out = model.predict(df)                                     # one pred_<feature> column each
-weights = model.mixture_weights(df)                         # putative cell-type composition
-draws = model.sample(df, n_samples=16)                   # training units drawn from the predictive distribution
 model.selftest()                                            # reproduces the shipped outputs
 ```
 Units you recorded yourself can be embedded with `model.encode(waveform, acg, stpc)` (arrays shaped
-as in `config.json`), then assigned to a putative cell type with `model.assign(z)`.
 ## What ships here
 - `autoencoder.pt`, `shared_latent_scaler.joblib` -- the phenotype encoder and its latent scaler.
-- `global_gmm.joblib` -- the mixture's component means and covariances.
-- `context_transform.joblib`, `context_weight_model_bundle.pt` -- context -> mixture weights.
-- `knn_bank.npz` -- standardised latents, phenotype features, GMM components and context keys of the training units (the exemplars; part of the model, like the stored points of any nearest-neighbour model).
-- `component_feature_expectations.npz` -- each component's expected features over the whole
-  brain.
 - `agea_vol_pca.npy`, `merfish_vol_pca.npy` -- the molecular and cellular context volumes, the same frozen
-  volumes as the channel-level model `int-brain-lab/ea-encoder-channel` of this vintage.
 - `split.json`, `config.json`, `preprocessing/unit_stats.npz`, `results/` -- the probe split, the
   training configuration, the training-data statistics and held-out evaluation summaries.
@@ -69,12 +76,12 @@ is only needed to retrain the model or to reproduce the atlas-wide analyses.
 ## Reproducibility
-**Pin the revision.** `revision="2026_W41"` is an immutable tag; omitting it resolves to `main`,
 which tracks the currently recommended model. `ephysatlas_model.json` records the training-time
 environment and random seed; `model.selftest()` verifies your install reproduces the shipped
 output (no IBL account needed).
 ## Citation
-Please cite the International Brain Laboratory Ephys Atlas. Model id `2026_W41_unit`, vintage
-`2026_W41`.

   - international-brain-lab
 ---
+# Ephys Atlas unit-level interpolation model (2026_W42)
 Predicts **which kinds of neurons you expect to record at a brain position**, and their expected
 waveform phenotype. Trained by the
 [International Brain Laboratory](https://www.internationalbrainlab.com/) on spike-sorted units of
 the Ephys Atlas release `2026_W41` (56497 training units).
+The model has four stages:
 1. a multimodal autoencoder embeds each unit's multi-channel **waveform**, 3D
    **autocorrelogram** and **population coupling** (waveform, acg, stpc) into a
    60-d latent -- its phenotype;
+2. a **hierarchical Gaussian mixture** over the standardised latents: 30 putative cell
+   types, each split into at most 10 sub-types (300 in all). Assignment is
+   top-down -- the most likely type, then the most likely of its sub-types -- so a unit's
+   sub-type always lies inside its type. The types and sub-types are global;
+3. the **sub-type weights depend on the molecular and cellular context** (PCA volumes of the
+   molecular and cellular atlases) at the position: one network predicts the type weights, a
+   second the shares of the sub-types within each type;
+4. a **members readout** maps the model back to 10 interpretable waveform features:
+   `depolarisation_slope`, `recovery_slope`, `repolarisation_slope`, `spatial_spread_um`, `tip_val`, `spike_width_secs`, `predepolarisation_width_secs`, `spike_amplitude`, `peak_to_trough_ratio_log`, `polarity`. Each sub-type is represented by its training units, from anywhere in the brain,
+   weighted by their posterior of the sub-type.
 > **The input is a position.** `predict` takes `x, y, z` (IBL/Allen frame, metres) and returns the
+> expected phenotype there: the local sub-type weights times each sub-type's expected features.
+> `mixture_weights` returns the local putative cell-type composition itself.
 ## Quickstart
 import pandas as pd
 from ephysatlas import load_pretrained
+model = load_pretrained("int-brain-lab/ea-encoder-unit", revision="2026_W42")
 df = pd.read_parquet("example/positions_sample.parquet")   # positions (x, y, z)
 out = model.predict(df)                                     # one pred_<feature> column each
+types = model.mixture_weights(df, level="type")             # putative cell-type composition
+subtypes = model.mixture_weights(df)                        # ... and of the sub-types
+draws = model.sample(df, n_samples=16)                      # training units drawn at each position
 model.selftest()                                            # reproduces the shipped outputs
 ```
 Units you recorded yourself can be embedded with `model.encode(waveform, acg, stpc)` (arrays shaped
+as in `config.json`), then assigned to a putative cell type and sub-type with `model.assign(z)`.
 ## What ships here
 - `autoencoder.pt`, `shared_latent_scaler.joblib` -- the phenotype encoder and its latent scaler.
+- `types_gmm.joblib`, `subtypes_gmm.joblib`, `subtype_parent.npy` -- the types, the sub-types and
+  the type of each sub-type.
+- `context_transform.joblib`, `context_weights_types.pt`, `context_weights_subtypes.pt` -- context
+  -> type and sub-type weights.
+- `members.npz` -- the phenotype features of the training units and their sub-type posteriors
+  (the readout; part of the model, like the stored points of any nearest-neighbour model).
 - `agea_vol_pca.npy`, `merfish_vol_pca.npy` -- the molecular and cellular context volumes, the same frozen
+  volumes as the channel-level model `int-brain-lab/ea-encoder-channel` of release `2026_W41`.
 - `split.json`, `config.json`, `preprocessing/unit_stats.npz`, `results/` -- the probe split, the
   training configuration, the training-data statistics and held-out evaluation summaries.
 ## Reproducibility
+**Pin the revision.** `revision="2026_W42"` is an immutable tag; omitting it resolves to `main`,
 which tracks the currently recommended model. `ephysatlas_model.json` records the training-time
 environment and random seed; `model.selftest()` verifies your install reproduces the shipped
 output (no IBL account needed).
 ## Citation
+Please cite the International Brain Laboratory Ephys Atlas. Model id `2026_W42_unit`, vintage
+`2026_W42`.

autoencoder.pt CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a378b78c8882ed5a67e47ce5c7874952053effba468bd404cfda2efc41eeba7
-size 22723365

 version https://git-lfs.github.com/spec/v1
+oid sha256:e80529fbe476721edbc8564ad7568ced63e5d5aa861628ced2168984ba1c76df
+size 22722341

checksums.json CHANGED Viewed

@@ -8,23 +8,13 @@
     },
     {
       "path": "autoencoder.pt",
-      "hash": "a422fc57e96292f4eda04f4987d4b942e141184f",
-      "bytes": 22723365
-    },
-    {
-      "path": "component_feature_expectations.npz",
-      "hash": "e9e2bce209a1cafe397b3418033d1ac145b85a98",
-      "bytes": 1982
-    },
-    {
-      "path": "conditioning_summary.json",
-      "hash": "0320fcce083af6390afbecf3a5c36257304e6e95",
-      "bytes": 4641
     },
     {
       "path": "config.json",
-      "hash": "960f06662503129eaa1452ce3816dc9f5dbd9d13",
-      "bytes": 3143
     },
     {
       "path": "context_transform.joblib",
@@ -32,14 +22,19 @@
       "bytes": 3049
     },
     {
-      "path": "context_weight_model_bundle.pt",
-      "hash": "c2e3fea134bbfd3ca385d8f8f3906402321016fb",
-      "bytes": 69189
     },
     {
       "path": "ephysatlas_model.json",
-      "hash": "ce8afb84467171502a630dca9168b3ea35b378a9",
-      "bytes": 3419
     },
     {
       "path": "example/expected_latents.npy",
@@ -48,7 +43,7 @@
     },
     {
       "path": "example/expected_predictions.parquet",
-      "hash": "713d2560e759736fea3f7009bb59af5b5a917f78",
       "bytes": 32897
     },
     {
@@ -62,19 +57,14 @@
       "bytes": 303532
     },
     {
-      "path": "global_gmm.joblib",
-      "hash": "c4c976013ba9471ab89ae20c2a7b962214e796aa",
-      "bytes": 2174220
-    },
-    {
-      "path": "global_gmm_summary.json",
-      "hash": "fdca9c0e029ec048fa836a979125c1275e39c9c2",
-      "bytes": 951
     },
     {
-      "path": "knn_bank.npz",
-      "hash": "b6bdbf674ff7516293425290cda972c95a30fd04",
-      "bytes": 14528458
     },
     {
       "path": "merfish_vol_pca.npy",
@@ -83,18 +73,13 @@
     },
     {
       "path": "preprocessing/unit_stats.npz",
-      "hash": "fffe170d76a8209c3327e037474f3b30141cb9d6",
-      "bytes": 4684
-    },
-    {
-      "path": "readout_summary.json",
-      "hash": "c89dc533ed778f2e9f16bf90dc2e1e531a205e3e",
-      "bytes": 1207
     },
     {
       "path": "results/summary.json",
-      "hash": "4541c7aa7ccefe25d0f73d708cf83aa7705c5ed2",
-      "bytes": 2792
     },
     {
       "path": "shared_latent_scaler.joblib",
@@ -105,6 +90,26 @@
       "path": "split.json",
       "hash": "4929803af5ab981187146b79725bb2262516ab07",
       "bytes": 36198
     }
   ]
 }

     },
     {
       "path": "autoencoder.pt",
+      "hash": "b063707006c8b7b6caa6707d41f16758457bdcf7",
+      "bytes": 22722341
     },
     {
       "path": "config.json",
+      "hash": "96da460f3f683a86285d48c879ea5b9441b059ce",
+      "bytes": 2161
     },
     {
       "path": "context_transform.joblib",
       "bytes": 3049
     },
     {
+      "path": "context_weights_subtypes.pt",
+      "hash": "7921606bdf8c5de484575b62871069db3ccb3a5c",
+      "bytes": 211105
+    },
+    {
+      "path": "context_weights_types.pt",
+      "hash": "6652d8695760ea463b1edf26d506d521c3effcb5",
+      "bytes": 71613
     },
     {
       "path": "ephysatlas_model.json",
+      "hash": "22479fc3735728bfd970417e0b3c72fb7fe4a5f8",
+      "bytes": 3280
     },
     {
       "path": "example/expected_latents.npy",
     },
     {
       "path": "example/expected_predictions.parquet",
+      "hash": "f95ce00efc35becc9e522b1fc41143d1afdec9c1",
       "bytes": 32897
     },
     {
       "bytes": 303532
     },
     {
+      "path": "hierarchy_summary.json",
+      "hash": "89bfc46659f51e2a2ab686442b41f590e1a6df13",
+      "bytes": 1575
     },
     {
+      "path": "members.npz",
+      "hash": "4ec553b93c0b1a7ff6e7bf9e4939e01f25f3f975",
+      "bytes": 2941047
     },
     {
       "path": "merfish_vol_pca.npy",
     },
     {
       "path": "preprocessing/unit_stats.npz",
+      "hash": "3aa15fa2747ccfb0f2d19bd18a96f954ec5acd20",
+      "bytes": 5931
     },
     {
       "path": "results/summary.json",
+      "hash": "132ae9aaec9c4cc8326916a203de1e9124a02195",
+      "bytes": 2193
     },
     {
       "path": "shared_latent_scaler.joblib",
       "path": "split.json",
       "hash": "4929803af5ab981187146b79725bb2262516ab07",
       "bytes": 36198
+    },
+    {
+      "path": "subtype_parent.npy",
+      "hash": "548469f659f96181124fc77bca7468d2398fc1bd",
+      "bytes": 2528
+    },
+    {
+      "path": "subtypes_gmm.joblib",
+      "hash": "fea4d2750bf102c2af435a33d7675748f81ec50f",
+      "bytes": 17427276
+    },
+    {
+      "path": "training_summary.json",
+      "hash": "da346c135e36b53d77ade91c32c80b92a772102c",
+      "bytes": 27582
+    },
+    {
+      "path": "types_gmm.joblib",
+      "hash": "ed1f6c578389e569690c146f140d73e168c7defa",
+      "bytes": 2608443
     }
   ]
 }

conditioning_summary.json DELETED Viewed

@@ -1,183 +0,0 @@
-{
-  "best_val_balanced_ce": 2.502358913421631,
-  "rare_component_class_weights": [
-    1.0959323644638062,
-    0.8443834185600281,
-    0.7508723139762878,
-    1.2009798288345337,
-    0.7173991799354553,
-    1.6520248651504517,
-    1.4514827728271484,
-    1.766238808631897,
-    1.5663753747940063,
-    1.136484146118164,
-    0.784567654132843,
-    1.3841673135757446,
-    1.2993239164352417,
-    1.0953930616378784,
-    0.880633533000946,
-    1.443649172782898,
-    1.6870092153549194,
-    0.942762017250061,
-    0.8432743549346924,
-    1.1054006814956665,
-    1.243672490119934,
-    0.9290009140968323,
-    1.2716392278671265,
-    1.2723108530044556,
-    0.6863182187080383
-  ],
-  "history": [
-    {
-      "epoch": 1,
-      "train_balanced_ce": 3.1044461077596557,
-      "val_balanced_ce": 2.940312147140503
-    },
-    {
-      "epoch": 2,
-      "train_balanced_ce": 2.8550544814358565,
-      "val_balanced_ce": 2.7653393745422363
-    },
-    {
-      "epoch": 3,
-      "train_balanced_ce": 2.717028480602673,
-      "val_balanced_ce": 2.665975332260132
-    },
-    {
-      "epoch": 4,
-      "train_balanced_ce": 2.6381118051380823,
-      "val_balanced_ce": 2.60945463180542
-    },
-    {
-      "epoch": 5,
-      "train_balanced_ce": 2.5907921370467073,
-      "val_balanced_ce": 2.574721574783325
-    },
-    {
-      "epoch": 6,
-      "train_balanced_ce": 2.5612764647803012,
-      "val_balanced_ce": 2.5545220375061035
-    },
-    {
-      "epoch": 7,
-      "train_balanced_ce": 2.539769241716456,
-      "val_balanced_ce": 2.5401904582977295
-    },
-    {
-      "epoch": 8,
-      "train_balanced_ce": 2.524711161564394,
-      "val_balanced_ce": 2.5301058292388916
-    },
-    {
-      "epoch": 9,
-      "train_balanced_ce": 2.5107240215293882,
-      "val_balanced_ce": 2.5225393772125244
-    },
-    {
-      "epoch": 10,
-      "train_balanced_ce": 2.5011777977737735,
-      "val_balanced_ce": 2.518281936645508
-    },
-    {
-      "epoch": 11,
-      "train_balanced_ce": 2.4927975566462672,
-      "val_balanced_ce": 2.5139763355255127
-    },
-    {
-      "epoch": 12,
-      "train_balanced_ce": 2.485523729676122,
-      "val_balanced_ce": 2.5117905139923096
-    },
-    {
-      "epoch": 13,
-      "train_balanced_ce": 2.4786812922156005,
-      "val_balanced_ce": 2.509345769882202
-    },
-    {
-      "epoch": 14,
-      "train_balanced_ce": 2.473920933611214,
-      "val_balanced_ce": 2.5084235668182373
-    },
-    {
-      "epoch": 15,
-      "train_balanced_ce": 2.466722688242459,
-      "val_balanced_ce": 2.506657838821411
-    },
-    {
-      "epoch": 16,
-      "train_balanced_ce": 2.4628473176030643,
-      "val_balanced_ce": 2.506051778793335
-    },
-    {
-      "epoch": 17,
-      "train_balanced_ce": 2.457756945894324,
-      "val_balanced_ce": 2.5049221515655518
-    },
-    {
-      "epoch": 18,
-      "train_balanced_ce": 2.4537503601886623,
-      "val_balanced_ce": 2.5040204524993896
-    },
-    {
-      "epoch": 19,
-      "train_balanced_ce": 2.4498772444420265,
-      "val_balanced_ce": 2.503654718399048
-    },
-    {
-      "epoch": 20,
-      "train_balanced_ce": 2.44552511771527,
-      "val_balanced_ce": 2.502358913421631
-    },
-    {
-      "epoch": 21,
-      "train_balanced_ce": 2.4419240016296824,
-      "val_balanced_ce": 2.5036613941192627
-    },
-    {
-      "epoch": 22,
-      "train_balanced_ce": 2.438449048670186,
-      "val_balanced_ce": 2.5030152797698975
-    },
-    {
-      "epoch": 23,
-      "train_balanced_ce": 2.434962947273629,
-      "val_balanced_ce": 2.502431631088257
-    },
-    {
-      "epoch": 24,
-      "train_balanced_ce": 2.433007692690201,
-      "val_balanced_ce": 2.503234386444092
-    },
-    {
-      "epoch": 25,
-      "train_balanced_ce": 2.4299673003884696,
-      "val_balanced_ce": 2.503336191177368
-    },
-    {
-      "epoch": 26,
-      "train_balanced_ce": 2.42602004544816,
-      "val_balanced_ce": 2.5042073726654053
-    },
-    {
-      "epoch": 27,
-      "train_balanced_ce": 2.4244782100102014,
-      "val_balanced_ce": 2.503185749053955
-    },
-    {
-      "epoch": 28,
-      "train_balanced_ce": 2.4208548767878075,
-      "val_balanced_ce": 2.5045764446258545
-    },
-    {
-      "epoch": 29,
-      "train_balanced_ce": 2.4174641966830315,
-      "val_balanced_ce": 2.5043716430664062
-    },
-    {
-      "epoch": 30,
-      "train_balanced_ce": 2.415899703259455,
-      "val_balanced_ce": 2.50522780418396
-    }
-  ],
-  "context_definition": "50 MERFISH PCs + 50 AGEA PCs; TRAIN-only StandardScaler; mixture weights gamma only"
-}

config.json CHANGED Viewed

@@ -35,15 +35,15 @@
   "feature_fidelity_hidden_dim": 128,
   "lambda_feature_continuous": 0.25,
   "lambda_feature_polarity": 0.1,
-  "gmm_components": 25,
-  "gmm_covariance_type": "full",
   "gmm_reg_covar": 0.0001,
-  "gmm_n_init": 10,
   "gmm_max_iter": 500,
-  "gmm_min_component_fraction": 0.001,
   "n_cell_pcs": 50,
   "n_gene_pcs": 50,
-  "context_atlas_subdir": "context_atlas",
   "context_hidden_dim": 128,
   "context_layers": 2,
   "context_dropout": 0.1,
@@ -54,25 +54,12 @@
   "context_weight_patience": 10,
   "rare_component_power": 0.5,
   "rare_component_weight_cap": 10.0,
-  "knn_decoder_k": 20,
-  "readout_key_dim": 3,
-  "readout_key_alphas": [
-    100.0,
-    1000.0,
-    10000.0,
-    30000.0,
-    100000.0
-  ],
-  "readout_neighbours": 4096,
-  "readout_shrinkage": 10.0,
-  "readout_off_data_quantile": 0.99,
   "mirror_x_to_single_hemisphere": true,
   "mirror_x_sign": -1.0,
   "region_gaussian_variance_floor": 0.001,
   "kde_neighbors": 128,
   "kde_spatial_bandwidth_um": 350.0,
   "kde_latent_bandwidth": 0.55,
-  "kde_chunk_size": 256,
   "waveform_sampling_rate_hz": 30000.0,
   "diagnostic_examples": 16,
   "diagnostic_hist_bins": 40,
@@ -82,30 +69,12 @@
     0.5,
     0.8
   ],
-  "diagnostic_spatial_pcs": 3,
   "diagnostic_voxel_size_um": 100.0,
-  "diagnostic_slice_cmap": "seismic",
   "diagnostic_feature_slice_cmap": "seismic",
-  "diagnostic_save_voxel_predictions": false,
   "max_regions_in_distribution_plot": 16,
   "feature_slice_count": 5,
   "feature_slice_seed": 20260831,
-  "feature_slice_reference_max_units": 20000,
-  "feature_slice_component_mc_samples": 128,
-  "feature_slice_regional_mc_samples": 128,
-  "feature_slice_display_quantiles": [
-    0.02,
-    0.98
-  ],
-  "feature_slice_display_padding_fraction": 0.03,
-  "feature_slice_center": "median",
   "feature_nll_samples_per_test_unit": 4,
   "feature_nll_min_bandwidth_fraction": 0.02,
-  "feature_categorical_alpha": 1.0,
-  "latent_fidelity_test_units": 8000,
-  "latent_fidelity_sliced_wasserstein_projections": 128,
-  "merfish_correlation_voxel_um": 200,
-  "merfish_correlation_cmap": "inferno",
-  "merfish_correlation_vmin": 0.0,
-  "merfish_correlation_vmax": 0.5
 }

   "feature_fidelity_hidden_dim": 128,
   "lambda_feature_continuous": 0.25,
   "lambda_feature_polarity": 0.1,
+  "n_types": 30,
+  "max_subtypes": 10,
+  "min_units_per_subtype": 20,
+  "type_n_init": 1,
+  "subtype_n_init": 3,
   "gmm_reg_covar": 0.0001,
   "gmm_max_iter": 500,
   "n_cell_pcs": 50,
   "n_gene_pcs": 50,
   "context_hidden_dim": 128,
   "context_layers": 2,
   "context_dropout": 0.1,
   "context_weight_patience": 10,
   "rare_component_power": 0.5,
   "rare_component_weight_cap": 10.0,
   "mirror_x_to_single_hemisphere": true,
   "mirror_x_sign": -1.0,
   "region_gaussian_variance_floor": 0.001,
   "kde_neighbors": 128,
   "kde_spatial_bandwidth_um": 350.0,
   "kde_latent_bandwidth": 0.55,
   "waveform_sampling_rate_hz": 30000.0,
   "diagnostic_examples": 16,
   "diagnostic_hist_bins": 40,
     0.5,
     0.8
   ],
   "diagnostic_voxel_size_um": 100.0,
   "diagnostic_feature_slice_cmap": "seismic",
   "max_regions_in_distribution_plot": 16,
   "feature_slice_count": 5,
   "feature_slice_seed": 20260831,
   "feature_nll_samples_per_test_unit": 4,
   "feature_nll_min_bandwidth_fraction": 0.02,
+  "feature_categorical_alpha": 1.0
 }

global_gmm.joblib → context_weights_subtypes.pt RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2cf14fedcfa5dca6254a53e5cbe3070f9ac90f1fa071651d0c989c9636dc5f8
-size 2174220

 version https://git-lfs.github.com/spec/v1
+oid sha256:43c2a0dd38a8e20d52bb5bf398c70d97eefefb246dcfa715dafd5875aa07f4af
+size 211105

context_weight_model_bundle.pt → context_weights_types.pt RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0264ccc23b040f549c1d55808196d3b97fedb9201ca6ca0fb923d0c08640bd4
-size 69189

 version https://git-lfs.github.com/spec/v1
+oid sha256:7b8a1107449b17cc60862bf88f1f82e073f330f362093a7c348122c2f274e02a
+size 71613

ephysatlas_model.json CHANGED Viewed

@@ -1,8 +1,8 @@
 {
-  "model_id": "2026_W41_unit",
   "task": "unit-encoding",
   "model_class": "UnitAutoencoder",
-  "vintage": "2026_W41",
   "training": {
     "random_seed": 0,
     "training_size": 572,
@@ -22,17 +22,19 @@
     "ephysatlas": "0.9.0",
     "iblatlas": "1.2.0"
   },
-  "method": "gmm",
   "granularity": "unit",
   "artifacts": {
     "autoencoder": "autoencoder.pt",
     "config": "config.json",
     "scaler": "shared_latent_scaler.joblib",
-    "gmm": "global_gmm.joblib",
     "context_transform": "context_transform.joblib",
-    "context_weights": "context_weight_model_bundle.pt",
-    "knn_bank": "knn_bank.npz",
-    "component_features": "component_feature_expectations.npz",
     "stats": "preprocessing/unit_stats.npz",
     "split": "split.json",
     "context": [
@@ -85,31 +87,28 @@
       ],
       "modality_latent_dim": 20,
       "latent_dim": 60,
-      "gmm_components": 25,
-      "gmm_covariance_type": "full",
-      "knn_k": 20
     },
     "context": {
       "n_cell_pcs": 50,
       "n_gene_pcs": 50,
-      "conditions": "mixture weights, and which training exemplars represent each component in the phenotype readout; component means/covariances are global",
-      "source_model": "int-brain-lab/ea-encoder-channel"
     },
     "preprocessing": {
       "waveforms": "per-unit max-abs normalized, 20 channels centred on the peak channel",
       "stats_file": "preprocessing/unit_stats.npz",
       "mirror_x_to_single_hemisphere": true,
       "mirror_x_sign": -1.0
-    },
-    "readout": {
-      "method": "context_local_members",
-      "description": "the molecular context sets the mixture weights gamma_k(x) and selects which training exemplars represent each component: its members nearest to x in a context key, shrunk toward all of its members -- further as the key leaves the training data, and entirely without context",
-      "key_dim": 3,
-      "key_ridge_alpha": 1000.0,
-      "neighbours": 4096,
-      "shrinkage": 10.0,
-      "off_data_quantile": 0.99,
-      "absent_context": "global member means"
     }
   },
   "data_source": {

 {
+  "model_id": "2026_W42_unit",
   "task": "unit-encoding",
   "model_class": "UnitAutoencoder",
+  "vintage": "2026_W42",
   "training": {
     "random_seed": 0,
     "training_size": 572,
     "ephysatlas": "0.9.0",
     "iblatlas": "1.2.0"
   },
+  "method": "hierarchical_gmm",
   "granularity": "unit",
   "artifacts": {
     "autoencoder": "autoencoder.pt",
     "config": "config.json",
     "scaler": "shared_latent_scaler.joblib",
+    "types": "types_gmm.joblib",
+    "subtypes": "subtypes_gmm.joblib",
+    "subtype_parent": "subtype_parent.npy",
     "context_transform": "context_transform.joblib",
+    "type_weights": "context_weights_types.pt",
+    "subtype_weights": "context_weights_subtypes.pt",
+    "members": "members.npz",
     "stats": "preprocessing/unit_stats.npz",
     "split": "split.json",
     "context": [
       ],
       "modality_latent_dim": 20,
       "latent_dim": 60,
+      "density": "hierarchical Gaussian mixture, full covariances, top-down assignment",
+      "n_types": 30,
+      "n_subtypes": 300,
+      "max_subtypes_per_type": 10,
+      "min_units_per_subtype": 20
     },
     "context": {
       "n_cell_pcs": 50,
       "n_gene_pcs": 50,
+      "conditions": "the sub-type weights gamma_k(x) gamma_j|k(x) only; the type and sub-type means and covariances are global",
+      "source_model": "int-brain-lab/ea-encoder-channel",
+      "data_vintage": "2026_W41"
+    },
+    "readout": {
+      "method": "members",
+      "description": "each sub-type is represented by its training units, from anywhere in the brain, weighted by their top-down posterior of the sub-type"
     },
     "preprocessing": {
       "waveforms": "per-unit max-abs normalized, 20 channels centred on the peak channel",
       "stats_file": "preprocessing/unit_stats.npz",
       "mirror_x_to_single_hemisphere": true,
       "mirror_x_sign": -1.0
     }
   },
   "data_source": {

example/expected_predictions.parquet CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb33586780bd88be3409d10775315260daba26fdb6ef94c9efcd6f038fbebd1e
 size 32897

 version https://git-lfs.github.com/spec/v1
+oid sha256:84bb33f879a1c23b317468c2f066ddd4cd4f72aa0f5bf2d68638850f77ae3fec
 size 32897

global_gmm_summary.json DELETED Viewed

@@ -1,37 +0,0 @@
-{
-  "n_components": 25,
-  "covariance_type": "full",
-  "converged": true,
-  "n_iter": 54,
-  "component_mass_train": [
-    0.03589438647031784,
-    0.06046649068593979,
-    0.07646486163139343,
-    0.029889777302742004,
-    0.0837668851017952,
-    0.01579650305211544,
-    0.020463047549128532,
-    0.013819593004882336,
-    0.017571240663528442,
-    0.03337854519486427,
-    0.07003793865442276,
-    0.022501781582832336,
-    0.025536373257637024,
-    0.03592975065112114,
-    0.05559089034795761,
-    0.020685726776719093,
-    0.015148136764764786,
-    0.04850538820028305,
-    0.06062563508749008,
-    0.03528212010860443,
-    0.027872895821928978,
-    0.04995303228497505,
-    0.0266603734344244,
-    0.026632236316800117,
-    0.09152570366859436
-  ],
-  "min_component_mass": 0.013819593004882336,
-  "max_component_mass": 0.09152570366859436,
-  "rare_component_indices": [],
-  "rare_component_threshold": 0.001
-}

hierarchy_summary.json ADDED Viewed

	@@ -0,0 +1,102 @@

+{
+  "n_types": 30,
+  "n_subtypes": 300,
+  "subtypes_per_type": [
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10,
+    10
+  ],
+  "type_weights": [
+    0.04381398293679264,
+    0.037172569501938674,
+    0.02427903521234171,
+    0.0069111436405017565,
+    0.017799655051780185,
+    0.08154949179254226,
+    0.029276411132847943,
+    0.03978187889420966,
+    0.008872282071696878,
+    0.021841680566053713,
+    0.03599554057292672,
+    0.053692554384372676,
+    0.01754854426983168,
+    0.011899622430156388,
+    0.03709573105577985,
+    0.04909497542907621,
+    0.030531050409611953,
+    0.03478716229065524,
+    0.03012711696831558,
+    0.019278565200257378,
+    0.03976877612734015,
+    0.05570118528050738,
+    0.017418595054987426,
+    0.028776960521673715,
+    0.04041748391924934,
+    0.023711837517380444,
+    0.021711902004824212,
+    0.019385190425204944,
+    0.09577338399742967,
+    0.025985691339713582
+  ],
+  "type_converged": true,
+  "type_n_iter": 43,
+  "train_units_per_type": [
+    2486,
+    2103,
+    1373,
+    391,
+    1004,
+    4608,
+    1659,
+    2233,
+    502,
+    1229,
+    2023,
+    3055,
+    990,
+    672,
+    2100,
+    2769,
+    1709,
+    1968,
+    1710,
+    1092,
+    2246,
+    3158,
+    983,
+    1629,
+    2259,
+    1332,
+    1225,
+    1094,
+    5424,
+    1471
+  ]
+}

knn_bank.npz → members.npz RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c5e979bb32760c5dc82cee934dd13892909d08f74e0fa1d4b424c0bce6359203
-size 14528458

 version https://git-lfs.github.com/spec/v1
+oid sha256:8f0a9655f4c4d38db3156ef8529eddeae2dc6b27bb6f856da2612a4cb72232b5
+size 2941047

preprocessing/unit_stats.npz CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca5c928ed76ff8f6a2cdca816b408cf1ae54235af91e2cbaf4b2cf4376953c1c
-size 4684

 version https://git-lfs.github.com/spec/v1
+oid sha256:9db8b415b11accc45d314d5c8472c95228d344d2878df028959cea83d599b156
+size 5931

readout_summary.json DELETED Viewed

@@ -1,64 +0,0 @@
-{
-  "method": "context_local_members",
-  "key_features": [
-    "depolarisation_slope",
-    "recovery_slope",
-    "repolarisation_slope",
-    "spatial_spread_um",
-    "tip_val",
-    "spike_width_secs",
-    "predepolarisation_width_secs",
-    "spike_amplitude",
-    "peak_to_trough_ratio_log"
-  ],
-  "key_alpha": 1000.0,
-  "key_alpha_validation_mse": {
-    "100.0": 1.1037312015628236,
-    "1000.0": 1.1033603851820213,
-    "10000.0": 1.1033679139310546,
-    "30000.0": 1.1048441456682687,
-    "100000.0": 1.1105265257126138
-  },
-  "key_dim": 3,
-  "key_singular_values": [
-    95.02516531228356,
-    62.48198561115676,
-    51.015579869285794,
-    35.38328421883545,
-    21.195708629676567,
-    16.722950989014983,
-    10.308581820566737,
-    10.023965043087022,
-    4.122605997586523
-  ],
-  "neighbours": 4096,
-  "shrinkage": 10.0,
-  "off_data_quantile": 0.99,
-  "train_members_per_component": [
-    2036,
-    3412,
-    4319,
-    1685,
-    4732,
-    889,
-    1159,
-    780,
-    991,
-    1885,
-    3953,
-    1277,
-    1444,
-    2027,
-    3144,
-    1173,
-    855,
-    2740,
-    3417,
-    1989,
-    1571,
-    2826,
-    1503,
-    1505,
-    5185
-  ]
-}

results/summary.json CHANGED Viewed

@@ -1,6 +1,7 @@
 {
-  "model_name": "context_weights_k25_knn20",
-  "vintage": "2026_W41",
   "architecture": {
     "modalities": [
       "waveform",
@@ -9,22 +10,15 @@
     ],
     "modality_latent_dim": 20,
     "joint_latent_dim": 60,
-    "gmm_components": 25,
-    "gmm_covariance_type": "full",
-    "gmm_geometry": "global means and covariance matrices",
-    "context_conditioning": "mixture weights gamma, and the TRAIN exemplars representing each component",
     "context_definition": "50 MERFISH PCs + 50 AGEA PCs",
-    "knn_k": 20,
-    "knn_policy": "distance-weighted empirical TRAIN-exemplar projection in standardized latent space",
-    "phenotype_readout": {
-      "method": "context-local members: gamma_k(x) weights the mean of component k's TRAIN members nearest to x in a context key, shrunk toward all its members",
-      "key_dim": 3,
-      "key_ridge_alpha": 1000.0,
-      "neighbours": 4096,
-      "shrinkage": 10.0,
-      "off_data_quantile": 0.99,
-      "absent_context": "global member means"
-    }
   },
   "split_counts": {
     "train_units": 56497,
@@ -35,7 +29,7 @@
     "test_pids": 149
   },
   "heldout_basic_test": {
-    "model": "context_weights_k25_knn20",
     "split": "held-out test PIDs",
     "reconstruction": {
       "waveform_mse": 0.000788584933616221,
@@ -44,30 +38,23 @@
       "n_test_units": 4096
     },
     "latent_nll": {
-      "conditional_k25": -26.648546443872384,
-      "unconditional_k25": -26.095386854718793,
-      "conditional_improvement": 0.5531595891535908,
       "lower_is_better": true
     },
-    "knn20": {
-      "k": 20,
-      "nearest_distance_mean": 3.551875114440918,
-      "nearest_distance_median": 3.415438652038574,
-      "kth_distance_mean": 4.589000225067139,
-      "kth_distance_median": 4.397944450378418,
-      "effective_neighbors_mean": 19.91122817993164,
-      "effective_neighbors_median": 19.925037384033203
-    },
     "sanity_checks": {
       "all_metrics_finite": true,
-      "conditional_beats_unconditional": true
     }
   },
   "invariants": [
-    "TEST PIDs are never used for AE/GMM/context training or as kNN retrieval exemplars.",
-    "The GMM means and covariance matrices are global.",
-    "Molecular context sets the GMM mixture weights gamma_k(x) and selects which TRAIN exemplars represent each component in the phenotype readout; the conditional latent density depends on it only through gamma_k(x).",
-    "The kNN stage changes phenotype projection only; it does not change the conditional latent density.",
     "All spatial modeling uses the configured single-hemisphere mirroring convention."
   ]
 }

 {
+  "model_name": "hierarchical_gmm_30_types_300_subtypes",
+  "release_tag": "2026_W42",
+  "data_vintage": "2026_W41",
   "architecture": {
     "modalities": [
       "waveform",
     ],
     "modality_latent_dim": 20,
     "joint_latent_dim": 60,
+    "density": "hierarchical Gaussian mixture, full covariances",
+    "n_types": 30,
+    "n_subtypes": 300,
+    "max_subtypes_per_type": 10,
+    "min_units_per_subtype": 20,
+    "assignment": "top-down: P(k, j | z) = P(k | z) P(j | k, z), P(j | k, z) among the sub-types of k",
+    "context_conditioning": "sub-type weights gamma_k(x) gamma_j|k(x), from two context networks",
     "context_definition": "50 MERFISH PCs + 50 AGEA PCs",
+    "phenotype_readout": "members: each sub-type's TRAIN units from anywhere in the brain, weighted by their top-down posterior of the sub-type"
   },
   "split_counts": {
     "train_units": 56497,
     "test_pids": 149
   },
   "heldout_basic_test": {
+    "model": "hierarchical_gmm_30_types_300_subtypes",
     "split": "held-out test PIDs",
     "reconstruction": {
       "waveform_mse": 0.000788584933616221,
       "n_test_units": 4096
     },
     "latent_nll": {
+      "context_weights": -29.065819952667763,
+      "brain_wide_weights": -28.336594620975855,
+      "context_improvement": 0.7292253316919073,
+      "n_test_units": 15880,
       "lower_is_better": true
     },
     "sanity_checks": {
       "all_metrics_finite": true,
+      "context_beats_brain_wide": true,
+      "subtypes_inside_types": true
     }
   },
   "invariants": [
+    "TEST PIDs are never used for AE/mixture/context training or as readout members.",
+    "The type and sub-type means and covariance matrices are global.",
+    "Molecular context changes only the sub-type weights gamma_k(x) gamma_j|k(x).",
+    "A unit's sub-type always lies inside its type (top-down assignment).",
     "All spatial modeling uses the configured single-hemisphere mirroring convention."
   ]
 }

component_feature_expectations.npz → subtype_parent.npy RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1dabd6400099ea571918ca34a9b579de46ebd9ec6cc7187ef48641096221d60
-size 1982

 version https://git-lfs.github.com/spec/v1
+oid sha256:f6935879013bedeb209c8a0d152dbb98e195b1d8a21555451d4bd33150ff5ba8
+size 2528

subtypes_gmm.joblib ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:c2601d34df695e9f18c4be160e592d90d42c0e1e8701e3cf3024d3494269066a
+size 17427276

training_summary.json ADDED Viewed

	@@ -0,0 +1,1348 @@

+{
+  "autoencoder": "C:\\Users\\bodic\\.cache\\ephysatlas\\models\\models--int-brain-lab--ea-encoder-unit\\snapshots\\afe0db21c8e53286ecfd36513e4b6e4c202f41d9\\autoencoder.pt",
+  "n_types": 30,
+  "n_subtypes": 300,
+  "train_units_per_type": [
+    2486,
+    2103,
+    1373,
+    391,
+    1004,
+    4608,
+    1659,
+    2233,
+    502,
+    1229,
+    2023,
+    3055,
+    990,
+    672,
+    2100,
+    2769,
+    1709,
+    1968,
+    1710,
+    1092,
+    2246,
+    3158,
+    983,
+    1629,
+    2259,
+    1332,
+    1225,
+    1094,
+    5424,
+    1471
+  ],
+  "train_units_per_subtype": [
+    267,
+    134,
+    337,
+    265,
+    230,
+    249,
+    255,
+    261,
+    159,
+    329,
+    166,
+    161,
+    140,
+    460,
+    241,
+    129,
+    134,
+    268,
+    218,
+    186,
+    146,
+    155,
+    111,
+    166,
+    167,
+    194,
+    112,
+    52,
+    160,
+    110,
+    26,
+    44,
+    43,
+    42,
+    36,
+    33,
+    39,
+    27,
+    41,
+    60,
+    69,
+    58,
+    107,
+    70,
+    101,
+    156,
+    133,
+    106,
+    84,
+    120,
+    467,
+    520,
+    382,
+    588,
+    533,
+    207,
+    406,
+    529,
+    724,
+    252,
+    186,
+    157,
+    140,
+    155,
+    173,
+    185,
+    175,
+    156,
+    159,
+    173,
+    155,
+    203,
+    242,
+    273,
+    296,
+    257,
+    293,
+    271,
+    119,
+    124,
+    48,
+    18,
+    53,
+    53,
+    56,
+    52,
+    63,
+    44,
+    66,
+    49,
+    134,
+    144,
+    71,
+    150,
+    154,
+    156,
+    71,
+    112,
+    100,
+    137,
+    208,
+    178,
+    331,
+    211,
+    160,
+    160,
+    208,
+    180,
+    201,
+    186,
+    111,
+    281,
+    276,
+    314,
+    267,
+    826,
+    158,
+    192,
+    385,
+    245,
+    51,
+    168,
+    122,
+    96,
+    34,
+    84,
+    100,
+    142,
+    138,
+    55,
+    42,
+    94,
+    87,
+    43,
+    21,
+    88,
+    98,
+    70,
+    95,
+    34,
+    154,
+    262,
+    215,
+    142,
+    232,
+    152,
+    113,
+    290,
+    195,
+    345,
+    270,
+    477,
+    270,
+    336,
+    180,
+    256,
+    147,
+    278,
+    335,
+    220,
+    141,
+    148,
+    94,
+    217,
+    168,
+    205,
+    100,
+    231,
+    176,
+    229,
+    155,
+    184,
+    181,
+    214,
+    139,
+    236,
+    186,
+    199,
+    202,
+    272,
+    173,
+    150,
+    124,
+    191,
+    134,
+    192,
+    202,
+    194,
+    168,
+    182,
+    167,
+    133,
+    53,
+    296,
+    120,
+    85,
+    61,
+    46,
+    52,
+    79,
+    137,
+    243,
+    189,
+    194,
+    282,
+    189,
+    257,
+    318,
+    270,
+    167,
+    160,
+    222,
+    412,
+    169,
+    429,
+    597,
+    440,
+    262,
+    201,
+    266,
+    74,
+    89,
+    173,
+    9,
+    122,
+    112,
+    135,
+    42,
+    103,
+    124,
+    158,
+    149,
+    227,
+    131,
+    273,
+    177,
+    95,
+    164,
+    168,
+    87,
+    257,
+    274,
+    214,
+    192,
+    202,
+    319,
+    217,
+    169,
+    244,
+    171,
+    160,
+    135,
+    110,
+    137,
+    173,
+    126,
+    122,
+    110,
+    121,
+    138,
+    117,
+    138,
+    93,
+    149,
+    140,
+    136,
+    135,
+    98,
+    143,
+    76,
+    81,
+    77,
+    117,
+    149,
+    120,
+    120,
+    154,
+    118,
+    66,
+    92,
+    447,
+    351,
+    285,
+    456,
+    411,
+    469,
+    1173,
+    921,
+    425,
+    486,
+    126,
+    180,
+    130,
+    145,
+    120,
+    183,
+    124,
+    127,
+    145,
+    191
+  ],
+  "context_definition": "50 MERFISH PCs + 50 AGEA PCs; TRAIN-only StandardScaler; mixture weights only",
+  "context_networks": {
+    "types": {
+      "best_val_balanced_ce": 2.6379599571228027,
+      "rare_component_class_weights": [
+        0.9054701328277588,
+        0.9834902882575989,
+        1.2183433771133423,
+        2.2823283672332764,
+        1.4210728406906128,
+        0.664006769657135,
+        1.1088823080062866,
+        0.9519424438476562,
+        2.006247043609619,
+        1.284959316253662,
+        1.0000585317611694,
+        0.816394031047821,
+        1.4308336973190308,
+        1.7419129610061646,
+        0.9850553870201111,
+        0.8560022711753845,
+        1.0894670486450195,
+        1.0166257619857788,
+        1.093315839767456,
+        1.3658338785171509,
+        0.9505099058151245,
+        0.8030210137367249,
+        1.4378496408462524,
+        1.1177414655685425,
+        0.9447774291038513,
+        1.2320972681045532,
+        1.2877594232559204,
+        1.3630703687667847,
+        0.6125376224517822,
+        1.1779147386550903
+      ],
+      "history": [
+        {
+          "epoch": 1,
+          "train_balanced_ce": 3.327133729668561,
+          "val_balanced_ce": 3.1560418605804443
+        },
+        {
+          "epoch": 2,
+          "train_balanced_ce": 3.0558692987554674,
+          "val_balanced_ce": 2.9454360008239746
+        },
+        {
+          "epoch": 3,
+          "train_balanced_ce": 2.8952569323986936,
+          "val_balanced_ce": 2.8263261318206787
+        },
+        {
+          "epoch": 4,
+          "train_balanced_ce": 2.8059206115424487,
+          "val_balanced_ce": 2.7610409259796143
+        },
+        {
+          "epoch": 5,
+          "train_balanced_ce": 2.753854021155978,
+          "val_balanced_ce": 2.7239744663238525
+        },
+        {
+          "epoch": 6,
+          "train_balanced_ce": 2.719195930359015,
+          "val_balanced_ce": 2.7010092735290527
+        },
+        {
+          "epoch": 7,
+          "train_balanced_ce": 2.696662917370007,
+          "val_balanced_ce": 2.6844496726989746
+        },
+        {
+          "epoch": 8,
+          "train_balanced_ce": 2.6802184304965353,
+          "val_balanced_ce": 2.6739776134490967
+        },
+        {
+          "epoch": 9,
+          "train_balanced_ce": 2.6667175963488035,
+          "val_balanced_ce": 2.6649386882781982
+        },
+        {
+          "epoch": 10,
+          "train_balanced_ce": 2.655342041759328,
+          "val_balanced_ce": 2.6595187187194824
+        },
+        {
+          "epoch": 11,
+          "train_balanced_ce": 2.646644865784778,
+          "val_balanced_ce": 2.6531670093536377
+        },
+        {
+          "epoch": 12,
+          "train_balanced_ce": 2.638922105893132,
+          "val_balanced_ce": 2.650050401687622
+        },
+        {
+          "epoch": 13,
+          "train_balanced_ce": 2.6301155032011327,
+          "val_balanced_ce": 2.6474475860595703
+        },
+        {
+          "epoch": 14,
+          "train_balanced_ce": 2.624973926805493,
+          "val_balanced_ce": 2.6443593502044678
+        },
+        {
+          "epoch": 15,
+          "train_balanced_ce": 2.6193918219616528,
+          "val_balanced_ce": 2.6420750617980957
+        },
+        {
+          "epoch": 16,
+          "train_balanced_ce": 2.6124381206435547,
+          "val_balanced_ce": 2.6418068408966064
+        },
+        {
+          "epoch": 17,
+          "train_balanced_ce": 2.6091645328416666,
+          "val_balanced_ce": 2.640631675720215
+        },
+        {
+          "epoch": 18,
+          "train_balanced_ce": 2.602932117494973,
+          "val_balanced_ce": 2.6399850845336914
+        },
+        {
+          "epoch": 19,
+          "train_balanced_ce": 2.6001689587702512,
+          "val_balanced_ce": 2.63948655128479
+        },
+        {
+          "epoch": 20,
+          "train_balanced_ce": 2.5957646330476876,
+          "val_balanced_ce": 2.6383752822875977
+        },
+        {
+          "epoch": 21,
+          "train_balanced_ce": 2.5922095317014464,
+          "val_balanced_ce": 2.6388683319091797
+        },
+        {
+          "epoch": 22,
+          "train_balanced_ce": 2.588658949846774,
+          "val_balanced_ce": 2.638301134109497
+        },
+        {
+          "epoch": 23,
+          "train_balanced_ce": 2.5849562701051028,
+          "val_balanced_ce": 2.6379599571228027
+        },
+        {
+          "epoch": 24,
+          "train_balanced_ce": 2.5813809114508808,
+          "val_balanced_ce": 2.63814115524292
+        },
+        {
+          "epoch": 25,
+          "train_balanced_ce": 2.5773516544327726,
+          "val_balanced_ce": 2.6385602951049805
+        },
+        {
+          "epoch": 26,
+          "train_balanced_ce": 2.5750698106724457,
+          "val_balanced_ce": 2.6393325328826904
+        },
+        {
+          "epoch": 27,
+          "train_balanced_ce": 2.5713846944510554,
+          "val_balanced_ce": 2.6387481689453125
+        },
+        {
+          "epoch": 28,
+          "train_balanced_ce": 2.569814251969721,
+          "val_balanced_ce": 2.6385443210601807
+        },
+        {
+          "epoch": 29,
+          "train_balanced_ce": 2.5651087804897923,
+          "val_balanced_ce": 2.6396148204803467
+        },
+        {
+          "epoch": 30,
+          "train_balanced_ce": 2.562243485110189,
+          "val_balanced_ce": 2.6402180194854736
+        },
+        {
+          "epoch": 31,
+          "train_balanced_ce": 2.560101227952157,
+          "val_balanced_ce": 2.6405320167541504
+        },
+        {
+          "epoch": 32,
+          "train_balanced_ce": 2.55661426061132,
+          "val_balanced_ce": 2.6417434215545654
+        },
+        {
+          "epoch": 33,
+          "train_balanced_ce": 2.5544163542350695,
+          "val_balanced_ce": 2.6424314975738525
+        }
+      ]
+    },
+    "subtypes": {
+      "best_val_balanced_ce": 4.74838399887085,
+      "rare_component_class_weights": [
+        0.8858819007873535,
+        1.2492200136184692,
+        0.7882211208343506,
+        0.8914970755577087,
+        0.9558954238891602,
+        0.9178048968315125,
+        0.9075556993484497,
+        0.891228437423706,
+        1.1599836349487305,
+        0.7979210615158081,
+        1.1246016025543213,
+        1.1421326398849487,
+        1.2242275476455688,
+        0.6736546754837036,
+        0.9256846904754639,
+        1.2787002325057983,
+        1.2536908388137817,
+        0.8889058828353882,
+        0.9807614088058472,
+        1.0556004047393799,
+        1.202488660812378,
+        1.1546354293823242,
+        1.376981496810913,
+        1.1242632865905762,
+        1.1243021488189697,
+        1.0378310680389404,
+        1.3716754913330078,
+        2.021135091781616,
+        1.1408162117004395,
+        1.3846776485443115,
+        2.8380837440490723,
+        2.185183048248291,
+        2.2021446228027344,
+        2.237412214279175,
+        2.4206273555755615,
+        2.520090103149414,
+        2.332657814025879,
+        2.7840943336486816,
+        2.2599968910217285,
+        1.859877347946167,
+        1.7452892065048218,
+        1.8992681503295898,
+        1.4001833200454712,
+        1.7323801517486572,
+        1.4351166486740112,
+        1.1526896953582764,
+        1.253915548324585,
+        1.3997482061386108,
+        1.5742850303649902,
+        1.3191519975662231,
+        0.6708050966262817,
+        0.636965274810791,
+        0.7368935942649841,
+        0.5934725999832153,
+        0.626522958278656,
+        1.0054035186767578,
+        0.7200466394424438,
+        0.6271424293518066,
+        0.5370163321495056,
+        0.9150728583335876,
+        1.0626749992370605,
+        1.1559722423553467,
+        1.2219047546386719,
+        1.155783772468567,
+        1.1023613214492798,
+        1.058719515800476,
+        1.100713849067688,
+        1.1606889963150024,
+        1.158084511756897,
+        1.1032922267913818,
+        1.1628111600875854,
+        1.01945161819458,
+        0.9254534840583801,
+        0.8744521141052246,
+        0.8398566842079163,
+        0.9014085531234741,
+        0.8376497626304626,
+        0.8717156052589417,
+        1.332512378692627,
+        1.299177885055542,
+        2.0868031978607178,
+        3.4089150428771973,
+        1.9909248352050781,
+        1.9970426559448242,
+        1.9287986755371094,
+        1.999740719795227,
+        1.8038841485977173,
+        2.1949074268341064,
+        1.7484748363494873,
+        2.0651419162750244,
+        1.243155837059021,
+        1.2069836854934692,
+        1.7190134525299072,
+        1.1851561069488525,
+        1.1607916355133057,
+        1.1630581617355347,
+        1.7190017700195312,
+        1.360692024230957,
+        1.4396873712539673,
+        1.232232689857483,
+        0.9987742304801941,
+        1.084010124206543,
+        0.7905126214027405,
+        0.9897034764289856,
+        1.1475404500961304,
+        1.1406830549240112,
+        1.0047473907470703,
+        1.0724925994873047,
+        1.0209473371505737,
+        1.0578889846801758,
+        1.3793096542358398,
+        0.8652750253677368,
+        0.8736939430236816,
+        0.8187217116355896,
+        0.8903610706329346,
+        0.5031332969665527,
+        1.1494258642196655,
+        1.0478267669677734,
+        0.7324241995811462,
+        0.9238020777702332,
+        2.0251972675323486,
+        1.1150853633880615,
+        1.2893805503845215,
+        1.4819084405899048,
+        2.4805164337158203,
+        1.5903620719909668,
+        1.4423091411590576,
+        1.2132774591445923,
+        1.225212812423706,
+        1.9723249673843384,
+        2.2322027683258057,
+        1.4976388216018677,
+        1.5567967891693115,
+        2.207969903945923,
+        3.1560420989990234,
+        1.5449531078338623,
+        1.463883876800537,
+        1.7291325330734253,
+        1.4824727773666382,
+        2.4803500175476074,
+        1.1695085763931274,
+        0.8962638974189758,
+        0.9883393049240112,
+        1.2157013416290283,
+        0.9510790109634399,
+        1.1813758611679077,
+        1.3688195943832397,
+        0.8484463095664978,
+        1.0364807844161987,
+        0.7757946848869324,
+        0.8799888491630554,
+        0.6632192134857178,
+        0.8812764883041382,
+        0.7899048924446106,
+        1.0740567445755005,
+        0.9035863876342773,
+        1.191146731376648,
+        0.8661423325538635,
+        0.7856558561325073,
+        0.9731124639511108,
+        1.2248855829238892,
+        1.1858422756195068,
+        1.498724102973938,
+        0.9769347906112671,
+        1.114836573600769,
+        1.0007638931274414,
+        1.446586012840271,
+        0.9534030556678772,
+        1.0899989604949951,
+        0.9565661549568176,
+        1.161697268486023,
+        1.0621118545532227,
+        1.0778300762176514,
+        0.9906702041625977,
+        1.2265868186950684,
+        0.9392117857933044,
+        1.0629667043685913,
+        1.0261123180389404,
+        1.019930124282837,
+        0.8771238923072815,
+        1.1024092435836792,
+        1.1846909523010254,
+        1.310926079750061,
+        1.0492393970489502,
+        1.2498825788497925,
+        1.045751929283142,
+        1.021209955215454,
+        1.0378952026367188,
+        1.1257129907608032,
+        1.0707148313522339,
+        1.1208940744400024,
+        1.2543745040893555,
+        1.9790464639663696,
+        0.8429514765739441,
+        1.3230818510055542,
+        1.5686808824539185,
+        1.852584958076477,
+        2.1324236392974854,
+        2.007986068725586,
+        1.627545714378357,
+        1.2397407293319702,
+        0.9259937405586243,
+        1.049759864807129,
+        1.0352611541748047,
+        0.8641874194145203,
+        1.0544763803482056,
+        0.9003723859786987,
+        0.8041791915893555,
+        0.8812156319618225,
+        1.1240761280059814,
+        1.1459218263626099,
+        0.9727397561073303,
+        0.7126681208610535,
+        1.11687433719635,
+        0.6999203562736511,
+        0.5906533598899841,
+        0.6883585453033447,
+        0.8988834023475647,
+        1.020295262336731,
+        0.8898656964302063,
+        1.6926125288009644,
+        1.5300184488296509,
+        1.0966424942016602,
+        4.820933818817139,
+        1.301947832107544,
+        1.3714900016784668,
+        1.252974033355713,
+        2.23488450050354,
+        1.4274699687957764,
+        1.291433334350586,
+        1.1498477458953857,
+        1.1872669458389282,
+        0.9627104997634888,
+        1.2622126340866089,
+        0.8774643540382385,
+        1.0874232053756714,
+        1.4879547357559204,
+        1.1244398355484009,
+        1.1150504350662231,
+        1.5549542903900146,
+        0.8934657573699951,
+        0.8667140603065491,
+        0.9911537170410156,
+        1.0433052778244019,
+        1.0105829238891602,
+        0.8080525398254395,
+        0.9776186943054199,
+        1.1107953786849976,
+        0.9223783016204834,
+        1.1027956008911133,
+        1.1353992223739624,
+        1.2483879327774048,
+        1.3906948566436768,
+        1.230483889579773,
+        1.0961272716522217,
+        1.2923738956451416,
+        1.311789870262146,
+        1.3655481338500977,
+        1.3092725276947021,
+        1.2180919647216797,
+        1.3389089107513428,
+        1.2350339889526367,
+        1.499922275543213,
+        1.185589075088501,
+        1.2140130996704102,
+        1.2390824556350708,
+        1.243676781654358,
+        1.4656331539154053,
+        1.2080615758895874,
+        1.660159945487976,
+        1.6063001155853271,
+        1.6516400575637817,
+        1.3429843187332153,
+        1.1819355487823486,
+        1.3197319507598877,
+        1.3205816745758057,
+        1.1656744480133057,
+        1.3268711566925049,
+        1.782961368560791,
+        1.5089362859725952,
+        0.6819408535957336,
+        0.770066499710083,
+        0.8574084043502808,
+        0.6776633262634277,
+        0.7155359983444214,
+        0.6677842140197754,
+        0.4237954914569855,
+        0.47637006640434265,
+        0.6997332572937012,
+        0.6579715609550476,
+        1.2903233766555786,
+        1.0791043043136597,
+        1.275104284286499,
+        1.2030340433120728,
+        1.3166049718856812,
+        1.0697232484817505,
+        1.2875763177871704,
+        1.2980549335479736,
+        1.202796220779419,
+        1.0546985864639282
+      ],
+      "history": [
+        {
+          "epoch": 1,
+          "train_balanced_ce": 5.666822061502066,
+          "val_balanced_ce": 5.566389083862305
+        },
+        {
+          "epoch": 2,
+          "train_balanced_ce": 5.472308804158065,
+          "val_balanced_ce": 5.3895487785339355
+        },
+        {
+          "epoch": 3,
+          "train_balanced_ce": 5.301024833573665,
+          "val_balanced_ce": 5.229682445526123
+        },
+        {
+          "epoch": 4,
+          "train_balanced_ce": 5.152229858647048,
+          "val_balanced_ce": 5.097817420959473
+        },
+        {
+          "epoch": 5,
+          "train_balanced_ce": 5.0383959429022305,
+          "val_balanced_ce": 5.002829551696777
+        },
+        {
+          "epoch": 6,
+          "train_balanced_ce": 4.954631677755819,
+          "val_balanced_ce": 4.93775749206543
+        },
+        {
+          "epoch": 7,
+          "train_balanced_ce": 4.894083097782557,
+          "val_balanced_ce": 4.892715930938721
+        },
+        {
+          "epoch": 8,
+          "train_balanced_ce": 4.8506515156027,
+          "val_balanced_ce": 4.8612542152404785
+        },
+        {
+          "epoch": 9,
+          "train_balanced_ce": 4.818030038377822,
+          "val_balanced_ce": 4.837561130523682
+        },
+        {
+          "epoch": 10,
+          "train_balanced_ce": 4.78920033082841,
+          "val_balanced_ce": 4.819528579711914
+        },
+        {
+          "epoch": 11,
+          "train_balanced_ce": 4.7647132249478314,
+          "val_balanced_ce": 4.8058037757873535
+        },
+        {
+          "epoch": 12,
+          "train_balanced_ce": 4.743304482978629,
+          "val_balanced_ce": 4.794022083282471
+        },
+        {
+          "epoch": 13,
+          "train_balanced_ce": 4.726508649431089,
+          "val_balanced_ce": 4.784908294677734
+        },
+        {
+          "epoch": 14,
+          "train_balanced_ce": 4.711932225111474,
+          "val_balanced_ce": 4.777336120605469
+        },
+        {
+          "epoch": 15,
+          "train_balanced_ce": 4.696698321248227,
+          "val_balanced_ce": 4.771921634674072
+        },
+        {
+          "epoch": 16,
+          "train_balanced_ce": 4.684071786750078,
+          "val_balanced_ce": 4.766389846801758
+        },
+        {
+          "epoch": 17,
+          "train_balanced_ce": 4.671723687720725,
+          "val_balanced_ce": 4.763198375701904
+        },
+        {
+          "epoch": 18,
+          "train_balanced_ce": 4.660583729241521,
+          "val_balanced_ce": 4.759703159332275
+        },
+        {
+          "epoch": 19,
+          "train_balanced_ce": 4.6499694841217,
+          "val_balanced_ce": 4.756467342376709
+        },
+        {
+          "epoch": 20,
+          "train_balanced_ce": 4.639899801435261,
+          "val_balanced_ce": 4.75478982925415
+        },
+        {
+          "epoch": 21,
+          "train_balanced_ce": 4.631779058153828,
+          "val_balanced_ce": 4.752882480621338
+        },
+        {
+          "epoch": 22,
+          "train_balanced_ce": 4.624053954917764,
+          "val_balanced_ce": 4.751321315765381
+        },
+        {
+          "epoch": 23,
+          "train_balanced_ce": 4.61459136138771,
+          "val_balanced_ce": 4.749642848968506
+        },
+        {
+          "epoch": 24,
+          "train_balanced_ce": 4.608131133123974,
+          "val_balanced_ce": 4.749549388885498
+        },
+        {
+          "epoch": 25,
+          "train_balanced_ce": 4.600742366370187,
+          "val_balanced_ce": 4.749795436859131
+        },
+        {
+          "epoch": 26,
+          "train_balanced_ce": 4.593222081375318,
+          "val_balanced_ce": 4.74838399887085
+        },
+        {
+          "epoch": 27,
+          "train_balanced_ce": 4.585634634182795,
+          "val_balanced_ce": 4.748930931091309
+        },
+        {
+          "epoch": 28,
+          "train_balanced_ce": 4.579226414735358,
+          "val_balanced_ce": 4.748929023742676
+        },
+        {
+          "epoch": 29,
+          "train_balanced_ce": 4.573005742110314,
+          "val_balanced_ce": 4.749650955200195
+        },
+        {
+          "epoch": 30,
+          "train_balanced_ce": 4.565381892519081,
+          "val_balanced_ce": 4.749942302703857
+        },
+        {
+          "epoch": 31,
+          "train_balanced_ce": 4.560266610708739,
+          "val_balanced_ce": 4.750671863555908
+        },
+        {
+          "epoch": 32,
+          "train_balanced_ce": 4.554183052563381,
+          "val_balanced_ce": 4.751022815704346
+        },
+        {
+          "epoch": 33,
+          "train_balanced_ce": 4.548510943728996,
+          "val_balanced_ce": 4.75252103805542
+        },
+        {
+          "epoch": 34,
+          "train_balanced_ce": 4.54363978846566,
+          "val_balanced_ce": 4.754114151000977
+        },
+        {
+          "epoch": 35,
+          "train_balanced_ce": 4.537244271014157,
+          "val_balanced_ce": 4.755335330963135
+        },
+        {
+          "epoch": 36,
+          "train_balanced_ce": 4.53145268155969,
+          "val_balanced_ce": 4.756485462188721
+        }
+      ]
+    }
+  },
+  "readout": {
+    "method": "members",
+    "features": [
+      "depolarisation_slope",
+      "recovery_slope",
+      "repolarisation_slope",
+      "spatial_spread_um",
+      "tip_val",
+      "spike_width_secs",
+      "predepolarisation_width_secs",
+      "spike_amplitude",
+      "peak_to_trough_ratio_log",
+      "polarity"
+    ],
+    "stored_members_per_subtype": [
+      570,
+      287,
+      1019,
+      678,
+      634,
+      609,
+      641,
+      608,
+      412,
+      671,
+      396,
+      317,
+      380,
+      807,
+      615,
+      302,
+      354,
+      781,
+      530,
+      497,
+      327,
+      285,
+      158,
+      342,
+      320,
+      391,
+      218,
+      53,
+      356,
+      146,
+      26,
+      78,
+      45,
+      74,
+      38,
+      33,
+      48,
+      27,
+      42,
+      196,
+      80,
+      58,
+      186,
+      95,
+      210,
+      503,
+      190,
+      251,
+      144,
+      164,
+      1400,
+      1011,
+      1047,
+      1859,
+      1015,
+      895,
+      1140,
+      1273,
+      1407,
+      597,
+      429,
+      456,
+      357,
+      459,
+      494,
+      494,
+      347,
+      360,
+      518,
+      422,
+      544,
+      664,
+      695,
+      926,
+      1085,
+      1306,
+      888,
+      749,
+      272,
+      293,
+      58,
+      18,
+      74,
+      80,
+      89,
+      73,
+      118,
+      45,
+      160,
+      101,
+      502,
+      296,
+      117,
+      312,
+      349,
+      404,
+      107,
+      376,
+      245,
+      397,
+      496,
+      329,
+      722,
+      715,
+      357,
+      446,
+      426,
+      427,
+      472,
+      428,
+      190,
+      789,
+      1140,
+      1004,
+      738,
+      1908,
+      422,
+      688,
+      1361,
+      458,
+      51,
+      200,
+      487,
+      274,
+      34,
+      125,
+      327,
+      352,
+      324,
+      61,
+      42,
+      155,
+      122,
+      43,
+      21,
+      119,
+      239,
+      105,
+      229,
+      34,
+      369,
+      488,
+      514,
+      330,
+      555,
+      418,
+      255,
+      534,
+      497,
+      912,
+      968,
+      1083,
+      985,
+      979,
+      577,
+      1068,
+      483,
+      1246,
+      1666,
+      1135,
+      245,
+      268,
+      118,
+      384,
+      296,
+      511,
+      111,
+      501,
+      280,
+      417,
+      249,
+      487,
+      304,
+      542,
+      237,
+      456,
+      486,
+      318,
+      558,
+      511,
+      339,
+      262,
+      259,
+      389,
+      174,
+      497,
+      474,
+      540,
+      459,
+      267,
+      247,
+      181,
+      59,
+      429,
+      171,
+      88,
+      62,
+      47,
+      52,
+      79,
+      213,
+      412,
+      424,
+      533,
+      711,
+      600,
+      835,
+      1120,
+      688,
+      463,
+      386,
+      337,
+      622,
+      305,
+      719,
+      1266,
+      844,
+      587,
+      363,
+      960,
+      79,
+      124,
+      390,
+      9,
+      239,
+      245,
+      334,
+      42,
+      132,
+      196,
+      405,
+      254,
+      444,
+      168,
+      520,
+      245,
+      141,
+      532,
+      426,
+      127,
+      1178,
+      997,
+      546,
+      691,
+      839,
+      1286,
+      1006,
+      764,
+      992,
+      652,
+      576,
+      370,
+      268,
+      453,
+      429,
+      360,
+      200,
+      283,
+      292,
+      601,
+      189,
+      181,
+      98,
+      269,
+      295,
+      233,
+      238,
+      149,
+      330,
+      113,
+      86,
+      84,
+      242,
+      252,
+      211,
+      133,
+      274,
+      201,
+      69,
+      106,
+      1697,
+      1305,
+      1160,
+      990,
+      997,
+      1389,
+      2132,
+      2419,
+      1212,
+      1154,
+      351,
+      747,
+      399,
+      427,
+      249,
+      850,
+      519,
+      357,
+      373,
+      574
+    ]
+  }
+}

types_gmm.joblib ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:9ecb10a764de5607004108310818bedbc85368da41c876bf064a210285bfbbbf
+size 2608443