braindecode 1.6.0.dev1024__tar.gz → 1.6.0.dev1028__tar.gz

{braindecode-1.6.0.dev1024/braindecode.egg-info → braindecode-1.6.0.dev1028}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: braindecode
-Version: 1.6.0.dev1024
+Version: 1.6.0.dev1028
 Summary: Deep learning software to decode EEG, ECG or MEG signals
 Author-email: Robin Tibor Schirrmeister <robintibor@gmail.com>, Bruno Aristimunha Pinto <b.aristimunha@gmail.com>, Alexandre Gramfort <agramfort@meta.com>
 Maintainer-email: Alexandre Gramfort <agramfort@meta.com>, Bruno Aristimunha Pinto <b.aristimunha@gmail.com>, Robin Tibor Schirrmeister <robintibor@gmail.com>

{braindecode-1.6.0.dev1024 → braindecode-1.6.0.dev1028}/braindecode/models/contrawr.py

@@ -5,6 +5,7 @@ import torch.nn as nn
 from mne.utils import warn
 
 from braindecode.models.base import EEGModuleMixin
+from braindecode.models.util import _disable_batch_norm_training_if_batch_size_one
 
 
 class ContraWR(EEGModuleMixin, nn.Module):
@@ -122,6 +123,7 @@ class ContraWR(EEGModuleMixin, nn.Module):
         self.activation_layer = activation()
         self.final_layer = nn.Linear(emb_size, self.n_outputs)
 
+    @_disable_batch_norm_training_if_batch_size_one
     def forward(self, X: torch.Tensor) -> torch.Tensor:
         """
         Forward pass.

{braindecode-1.6.0.dev1024 → braindecode-1.6.0.dev1028}/braindecode/models/deepsleepnet.py

@@ -7,6 +7,7 @@ import torch.nn as nn
 from einops.layers.torch import Rearrange
 
 from braindecode.models.base import EEGModuleMixin
+from braindecode.models.util import _disable_batch_norm_training_if_batch_size_one
 
 
 class DeepSleepNet(EEGModuleMixin, nn.Module):
@@ -289,6 +290,7 @@ class DeepSleepNet(EEGModuleMixin, nn.Module):
             else nn.Linear(fc_out_features, self.n_outputs)
         )
 
+    @_disable_batch_norm_training_if_batch_size_one
     def forward(self, x):
         if x.ndim == 3:
             x = x.unsqueeze(1)

{braindecode-1.6.0.dev1024 → braindecode-1.6.0.dev1028}/braindecode/models/eegminer.py

@@ -13,6 +13,7 @@ from torch import nn
 
 import braindecode.functional as F
 from braindecode.models.base import EEGModuleMixin
+from braindecode.models.util import _disable_batch_norm_training_if_batch_size_one
 from braindecode.modules import GeneralizedGaussianFilter
 
 _eeg_miner_methods = ["mag", "corr", "plv"]
@@ -191,6 +192,7 @@ class EEGMiner(EEGModuleMixin, nn.Module):
         self.final_layer = nn.Linear(self.n_features, self.n_outputs)
         nn.init.zeros_(self.final_layer.bias)
 
+    @_disable_batch_norm_training_if_batch_size_one
     def forward(self, x):
         """x: (batch, electrodes, time)"""
         batch = x.shape[0]

{braindecode-1.6.0.dev1024 → braindecode-1.6.0.dev1028}/braindecode/models/eegsym.py

@@ -11,6 +11,7 @@ from braindecode.datautil.channel_utils import (
     match_hemisphere_chans,
 )
 from braindecode.models.base import EEGModuleMixin
+from braindecode.models.util import _disable_batch_norm_training_if_batch_size_one
 
 
 class EEGSym(EEGModuleMixin, nn.Module):
@@ -363,6 +364,7 @@ class EEGSym(EEGModuleMixin, nn.Module):
             out_features=self.n_outputs,
         )
 
+    @_disable_batch_norm_training_if_batch_size_one
     def forward(self, x):
         """Forward pass.
 

{braindecode-1.6.0.dev1024 → braindecode-1.6.0.dev1028}/braindecode/models/usleep.py

@@ -9,6 +9,7 @@ import torch
 from torch import nn
 
 from braindecode.models.base import EEGModuleMixin
+from braindecode.models.util import _disable_batch_norm_training_if_batch_size_one
 
 
 class USleep(EEGModuleMixin, nn.Module):
@@ -295,6 +296,7 @@ class USleep(EEGModuleMixin, nn.Module):
             # output is (B, n_classes, S)
         )
 
+    @_disable_batch_norm_training_if_batch_size_one
     def forward(self, x: torch.Tensor) -> torch.Tensor:
         """If input x has shape (B, S, C, T), return y_pred of shape (B, n_classes, S).
         If input x has shape (B, C, T), return y_pred of shape (B, n_classes).

{braindecode-1.6.0.dev1024 → braindecode-1.6.0.dev1028}/braindecode/models/util.py

@@ -12,6 +12,8 @@ from typing import Any, Dict, Literal, Optional, Sequence
 import numpy as np
 import pandas as pd
 import pydantic
+import torch  # noqa: F401  # exposed so TorchScript can resolve ``torch`` through the BatchNorm-guard wrapper's __globals__
+from torch import nn
 
 models_dict = {}
 
@@ -23,6 +25,38 @@ _EEG_PARAMS = frozenset(
 )
 
 _JSON_SAFE = (int, float, str, bool, type(None))
+_BATCH_NORM_MODULES = (
+    nn.BatchNorm1d,
+    nn.BatchNorm2d,
+    nn.BatchNorm3d,
+    nn.SyncBatchNorm,
+)
+
+
+def _disable_batch_norm_training_if_batch_size_one(forward):
+    """Temporarily set BatchNorm layers to train(False) for batch size one."""
+    forward_signature = inspect.signature(forward)
+
+    @wraps(forward)
+    def wrapped(self, *args, **kwargs):
+        bound = forward_signature.bind(self, *args, **kwargs)
+        x = next(value for name, value in bound.arguments.items() if name != "self")
+        batch_norms = []
+        if self.training and x.shape[0] == 1:
+            batch_norms = [
+                layer
+                for layer in self.modules()
+                if isinstance(layer, _BATCH_NORM_MODULES) and layer.training
+            ]
+            for batch_norm in batch_norms:
+                batch_norm.train(False)
+        try:
+            return forward(self, *args, **kwargs)
+        finally:
+            for batch_norm in batch_norms:
+                batch_norm.train(True)
+
+    return wrapped
 
 
 def _is_jsonable(val):

braindecode-1.6.0.dev1028/braindecode/version.py

@@ -0,0 +1 @@
+__version__ = "1.6.0.dev1028"

{braindecode-1.6.0.dev1024 → braindecode-1.6.0.dev1028/braindecode.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: braindecode
-Version: 1.6.0.dev1024
+Version: 1.6.0.dev1028
 Summary: Deep learning software to decode EEG, ECG or MEG signals
 Author-email: Robin Tibor Schirrmeister <robintibor@gmail.com>, Bruno Aristimunha Pinto <b.aristimunha@gmail.com>, Alexandre Gramfort <agramfort@meta.com>
 Maintainer-email: Alexandre Gramfort <agramfort@meta.com>, Bruno Aristimunha Pinto <b.aristimunha@gmail.com>, Robin Tibor Schirrmeister <robintibor@gmail.com>

{braindecode-1.6.0.dev1024 → braindecode-1.6.0.dev1028}/docs/whats_new.rst

@@ -41,7 +41,14 @@ Requirements
 Bug fixes
 ==========
 
-- None yet
+- Fix :class:`~braindecode.models.ContraWR`,
+  :class:`~braindecode.models.DeepSleepNet`, and
+  :class:`~braindecode.models.EEGMiner` raising a confusing
+  :class:`ValueError` on ``batch_size=1`` forward passes in train mode due to
+  :class:`~torch.nn.BatchNorm1d` (and its 2D/3D variants) requiring more than
+  one sample per channel during training.  The affected BatchNorm layers now
+  use running statistics when ``batch_size == 1`` in train mode.
+  By `Bruno Aristimunha`_.
 
 Code health
 ============

braindecode-1.6.0.dev1024/braindecode/version.py

@@ -1 +0,0 @@
-__version__ = "1.6.0.dev1024"