Publish 2026_W39_Cosmos_comfortable-cerulean-serval

README.md

@@ -9,12 +9,12 @@ tags:
   - international-brain-lab
 ---
 
-# Ephys Atlas region classifier (2026_W26, Cosmos)
+# Ephys Atlas region classifier (2026_W39, Cosmos)
 
 Predicts the **brain region of each Neuropixels recording channel** from electrophysiological
 features alone -- no histology required. Trained by the
 [International Brain Laboratory](https://www.internationalbrainlab.com/) on the Ephys Atlas
-feature release `2026_W26`.
+feature release `2026_W39`.
 
 > **What you can and cannot do without IBL access.** The *model* runs for anyone. Computing
 > the *input features* from raw Neuropixels data needs `ibllib` / `ibl-neuropixel`, and the
@@ -27,7 +27,7 @@ feature release `2026_W26`.
 import pandas as pd
 from ephysatlas import load_pretrained
 
-model = load_pretrained("int-brain-lab/ea-decoder-channel-xgboost", revision="2026_W26")
+model = load_pretrained("int-brain-lab/ea-decoder-channel-xgboost", revision="2026_W39")
 df = pd.read_parquet("example/features_sample.parquet")   # or your own features
 out = model.predict(df)
 print(out[["predicted_acronym", "prediction_probability", "fold_agreement"]].head())
@@ -54,10 +54,10 @@ two modes disagree on a small fraction of channels, so pick one per analysis.
 
 ## Inputs
 
-- **51 features**, listed in `ephysatlas_model.json` under `inputs.features`. Every
+- **48 features**, listed in `ephysatlas_model.json` under `inputs.features`. Every
   one must be present; `predict` raises and names anything missing.
 - Indexed by `(pid, channel)`, one row per recording channel.
-- **Must be the denoised aggregated features of vintage `2026_W26`** -- that is, the
+- **Must be the denoised aggregated features of vintage `2026_W39`** -- that is, the
   `raw_ephys_features_denoised.pqt` table produced by the Ephys Atlas aggregation pipeline,
   as loaded by `ephysatlas.data.read_features_from_disk`. Units are baked into that table by
   the pipeline (RMS features in dB, `spike_count` in log2), so feeding raw features, or
@@ -66,8 +66,8 @@ two modes disagree on a small fraction of channels, so pick one per analysis.
 
 ## Performance
 
-Pooled out-of-fold accuracy: **0.5959** over 13 Cosmos regions,
-761 insertions. Splits are by insertion (`pid`), so no channel from a test
+Pooled out-of-fold accuracy: **0.5946** over 13 Cosmos regions,
+811 insertions. Splits are by insertion (`pid`), so no channel from a test
 insertion appears in training. See `confusion_matrix.png`.
 
 Note this figure scores each channel with the one fold that held it out, whereas the default
@@ -85,7 +85,7 @@ typically the marginally better estimator, so treat the number as slightly conse
 
 ## Reproducibility
 
-**Pin the revision.** `revision="2026_W26"` is an immutable tag. Omitting `revision` resolves
+**Pin the revision.** `revision="2026_W39"` is an immutable tag. Omitting `revision` resolves
 to `main`, which tracks whichever model is currently recommended and *will* change when a new
 feature vintage is published — fine for a first look, not for anything you publish or re-run.
 
@@ -101,5 +101,5 @@ which the feature transformer relies on).
 
 ## Citation
 
-Please cite the International Brain Laboratory Ephys Atlas. Model id `2026_W26_Cosmos_agitated-latte-sandpiper`,
-feature vintage `2026_W26`.
+Please cite the International Brain Laboratory Ephys Atlas. Model id `2026_W39_Cosmos_comfortable-cerulean-serval`,
+feature vintage `2026_W39`.

checksums.json

@@ -1,90 +1,55 @@
 {
   "algo": "sha1",
   "files": [
-    {
-      "path": "confusion_matrix.png",
-      "hash": "490a7208f56840764c382070196a46f46e56fe9b",
-      "bytes": 210725
-    },
     {
       "path": "ephysatlas_model.json",
-      "hash": "65932d7dd76206647e2abbeef575771778dfaad2",
-      "bytes": 2875
+      "hash": "bf98daff2073309d99ffba9b87af3adaa59d619e",
+      "bytes": 2830
     },
     {
       "path": "example/expected_predictions.parquet",
-      "hash": "b76e75acb1087fc01f7b9ec368f2b177b981c740",
-      "bytes": 85727
+      "hash": "9e492b6eff2aeec7cf7221959bb80e6fc82d9f50",
+      "bytes": 85371
     },
     {
       "path": "example/features_sample.parquet",
-      "hash": "0e0386a5f8e9a0112d2eed7d1e9b4dffeaefa319",
-      "bytes": 274157
-    },
-    {
-      "path": "folds/FOLD00/meta.yaml",
-      "hash": "da8e2925cbae2c4b5ac582fc1188a833e11241f9",
-      "bytes": 1260
+      "hash": "ae608b7c51acd4fd8e4aaa30adb326f8b796d7cd",
+      "bytes": 258470
     },
     {
       "path": "folds/FOLD00/model.ubj",
-      "hash": "61c2349c734e115b753cab9f519cc0cba001ddcc",
-      "bytes": 4667085
-    },
-    {
-      "path": "folds/FOLD01/meta.yaml",
-      "hash": "e45952ed7121248ed436305bbd7cddc3582b2be7",
-      "bytes": 1260
+      "hash": "c8a5b78ae465c044941894586d50952b03bca006",
+      "bytes": 5011871
     },
     {
       "path": "folds/FOLD01/model.ubj",
-      "hash": "ac129dc6fb50cc61f9e61dc7b2bc672ff2a9c0e2",
-      "bytes": 4695254
-    },
-    {
-      "path": "folds/FOLD02/meta.yaml",
-      "hash": "3bbb6c16f2bbcd7535f9c0fb1da25e8b90eeaced",
-      "bytes": 1260
+      "hash": "664fde50893806d25af17199fde1bc43dcfa212a",
+      "bytes": 4974940
     },
     {
       "path": "folds/FOLD02/model.ubj",
-      "hash": "f66e41919925ec0506242d14735778af81ef7c7b",
-      "bytes": 4709750
-    },
-    {
-      "path": "folds/FOLD03/meta.yaml",
-      "hash": "3180f02a6359ac1a1599d3f71b6ce98c20ea7349",
-      "bytes": 1260
+      "hash": "fb21b5290f8cc07aa1283710ef81aa1ce88d05c4",
+      "bytes": 4997608
     },
     {
       "path": "folds/FOLD03/model.ubj",
-      "hash": "6b51a4c2c226ac363cddae7d7ab2d462b9f306ce",
-      "bytes": 4736616
-    },
-    {
-      "path": "folds/FOLD04/meta.yaml",
-      "hash": "54fab968b9f03bde8b8b585fe41fce53e46ba9f9",
-      "bytes": 1260
+      "hash": "afb097d38ae7a7995bfd562823a72d5d878ff0e2",
+      "bytes": 4982063
     },
     {
       "path": "folds/FOLD04/model.ubj",
-      "hash": "635e8a7c72f9dd7a4c54550e5e4baae71fac8e5d",
-      "bytes": 4705436
-    },
-    {
-      "path": "meta.yaml",
-      "hash": "70fcaea1a9a89b3021a5e7b9b0dcca0fbf25474d",
-      "bytes": 1198
+      "hash": "00c6e2559e7713211cb680ae24582662d080eded",
+      "bytes": 4968939
     },
     {
       "path": "model.ubj",
-      "hash": "35ebf36032b127a6a7e15bae2aad14c76d844d14",
-      "bytes": 4883969
+      "hash": "33e4a0c9baf46f77ce8ec0283b4bdb2b24c23582",
+      "bytes": 5074753
     },
     {
       "path": "split.json",
-      "hash": "bb9b993f5b18f57f1936c2ce2bfc0c4bb47633c5",
-      "bytes": 217742
+      "hash": "69640ef12e8c08ecf466f3d37ad7c28cd1af5cd0",
+      "bytes": 231942
     }
   ]
 }

ephysatlas_model.json

@@ -1,23 +1,23 @@
 {
-  "model_id": "2026_W26_Cosmos_agitated-latte-sandpiper",
+  "model_id": "2026_W39_Cosmos_comfortable-cerulean-serval",
   "task": "region-classification",
   "model_class": "xgboost.sklearn.XGBClassifier",
-  "vintage": "2026_W26",
+  "vintage": "2026_W39",
   "training": {
     "random_seed": 12345,
-    "training_size": 761,
+    "training_size": 811,
     "testing_size": 0,
-    "hash_training": "2b1c68c524d9a6bf1bc0c5766feba0b5fe070789ec0cb523036517aa57e1c100",
+    "hash_training": "05a75d4702030cce0e44c86d9d3fecc91f67a9702e1c1493716c7d841344b38b",
     "hash_testing": null
   },
   "environment": {
     "python": "3.11.14",
     "xgboost": "3.2.0",
     "scikit_learn": "1.8.0",
-    "numpy": "2.4.6",
+    "numpy": "2.3.5",
     "pandas": "3.0.5",
     "torch": "2.14.0",
-    "ephysatlas": "0.7.0"
+    "ephysatlas": "0.9.0"
   },
   "method": "xgboost",
   "granularity": "channel",
@@ -57,7 +57,6 @@
       "decay_fit_error",
       "decay_fit_r_squared",
       "decay_n_peaks",
-      "rms_lf_no_car",
       "rms_lf_csd",
       "psd_delta_csd",
       "psd_theta_csd",
@@ -78,20 +77,18 @@
       "alpha_mean",
       "alpha_std",
       "depolarisation_slope",
-      "peak_time_secs",
-      "peak_val",
       "polarity",
       "recovery_slope",
-      "recovery_time_secs",
       "repolarisation_slope",
+      "slowness_s_per_m",
+      "slowness_s_per_m_std",
+      "spatial_spread_um",
+      "spatial_spread_um_std",
       "spike_count",
-      "tip_time_secs",
       "tip_val",
-      "trough_time_secs",
-      "trough_val",
       "outside"
     ],
-    "feature_order_sha256": "7e37a356402fb16faa605aab5ba4ec129d9886a7c1ace493351c8802c1d1c78a"
+    "feature_order_sha256": "b1ff4d2e78a689518c73eb480649b09789d3d6d613ae2b4585850b813e24094f"
   },
   "outputs": {
     "kind": "categorical",
@@ -134,6 +131,6 @@
       "void_fluid"
     ],
     "region_map": "Cosmos",
-    "accuracy": 0.5958579719667104
+    "accuracy": 0.5945880856966708
   }
 }

example/expected_predictions.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c937f8e3996a574b1ad497bd67cce14f2d449391dcef0c826042989d5e0325d
-size 85727
+oid sha256:75bf524095dd94ef5a970cbdc16798344f3dbc74a925353fbd3e5e878f03de2b
+size 85371

example/features_sample.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f32bada42429cf891c08d2f443f4db46f2ff6a2692b843a29897896484f7664b
-size 274157
+oid sha256:020655a5f746c30862c2a27be2e9be2b910b149ef3b97fb17064538487aa481d
+size 258470

folds/FOLD00/model.ubj

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e1714dadd326f59335c1c4c23e34fb480f272676a7d927f606632999353245a
-size 4667085
+oid sha256:999fc52438951497bebbe0d7a335f77db9881d94488dd06580a90ecc297959ae
+size 5011871

folds/FOLD01/model.ubj

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0be7566640aa775322d5093f1c6f90cb6f47b86bdc84b1f635affacdd5f6e27a
-size 4695254
+oid sha256:fcc90566edc818b6510ca160d6a69c4f418ccdacf1a68225cf5fc3418070104e
+size 4974940

folds/FOLD02/model.ubj

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bdd37a338ce35c5dbb8a0c3d54b1c169d412ad1482ffde705c0ab6a20d8655b
-size 4709750
+oid sha256:19f5a7ef02346c67705576ed2c47725831971d5c3cba1b67ba959088da243635
+size 4997608

folds/FOLD03/model.ubj

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a741524ebb9acb63c96b8448a1f3d7ef7e829dea4350ac33e2d1cf18352d72dd
-size 4736616
+oid sha256:53b55f8415feb1dca256a19a51670b08d8264ff448ad02cf903a73c236031183
+size 4982063

folds/FOLD04/model.ubj

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aef4a89b6f92839be59d3705619a8525374f9bc354de4a00d57d01dda430b22d
-size 4705436
+oid sha256:04fad23202941f688e8920e56d691c3f41ce43ec9791a061be54d5821ced4130
+size 4968939

model.ubj

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a1cf2ad112654a1a4ab96744402711b023f230d839691a3977905c773f636838
-size 4883969
+oid sha256:aa30724d94a82799b65fd13801f75d424704209ab34c2329d091647a21c774f0
+size 5074753