PyPI - Braindecode - Versions diffs - 1.6.1.dev1049__tar.gz → 1.6.1.dev1052__tar.gz - Mend

@@ -1,89 +1,7 @@
-from warnings import warn
 import torch
 from torch import nn
-class PatchTokenizer(nn.Module):
-    r"""Tokenize an EEG signal into non-overlapping temporal patches.
-    Transforms ``(batch, n_chans, n_times)`` into
-    ``(batch, n_chans, n_patches, patch_dim)`` by splitting the time axis into
-    non-overlapping patches of ``patch_size`` samples. This is the shared
-    patch / "tokenization" step used by transformer EEG foundation models
-    (e.g. LaBraM, CBraMod, EEG-DINO).
-    As in the filter-bank models (:class:`~braindecode.models.FBCNet`,
-    :class:`~braindecode.models.FBMSNet`), when ``n_times`` is not a multiple of
-    ``patch_size`` the input is right zero-padded (a warning is emitted at
-    construction); it is never an error.
-    Two modes:
-    - **non-learnable** (``learnable=False``, default): a pure reshape, so
-      ``patch_dim == patch_size`` and the raw samples of each patch are kept
-      (the patch embedding, if any, lives in the model).
-    - **learnable** (``learnable=True``): a strided ``Conv1d`` (kernel and
-      stride equal to ``patch_size``, applied per channel) maps each patch to
-      ``emb_dim`` features, so ``patch_dim == emb_dim``.
-    Parameters
-    ----------
-    patch_size : int
-        Number of time samples per patch.
-    n_times : int
-        Number of time samples of the input, used to set up the right-padding
-        when ``n_times`` is not a multiple of ``patch_size``.
-    emb_dim : int, optional
-        Output features per patch in learnable mode. Defaults to ``patch_size``.
-        Ignored when ``learnable=False``.
-    learnable : bool, default=False
-        Whether the tokenizer is a learned convolution or a fixed reshape.
-    Examples
-    --------
-    >>> import torch
-    >>> from braindecode.modules import PatchTokenizer
-    >>> tokenizer = PatchTokenizer(patch_size=200, n_times=1000)
-    >>> tokenizer(torch.randn(2, 19, 1000)).shape
-    torch.Size([2, 19, 5, 200])
-    """
-    def __init__(self, patch_size, n_times, emb_dim=None, learnable=False):
-        super().__init__()
-        self.patch_size = patch_size
-        self.learnable = learnable
-        self.emb_dim = (emb_dim or patch_size) if learnable else patch_size
-        if n_times % patch_size:
-            warn(
-                f"Time dimension ({n_times}) is not divisible by patch_size "
-                f"({patch_size}). Input will be padded.",
-                UserWarning,
-            )
-            self.padding_layer = nn.ConstantPad1d(
-                (0, patch_size - n_times % patch_size), 0.0
-            )
-        else:
-            self.padding_layer = nn.Identity()
-        # Defined unconditionally (Identity when not learnable) so the attribute
-        # always exists for torch.jit.script, which type-checks both branches.
-        if learnable:
-            self.patcher = nn.Conv1d(1, self.emb_dim, patch_size, stride=patch_size)
-        else:
-            self.patcher = nn.Identity()
-    def forward(self, x):
-        x = self.padding_layer(x)
-        batch_size, n_chans, _ = x.shape
-        if self.learnable:
-            x = x.flatten(0, 1).unsqueeze(1)  # (batch * chans, 1, time)
-            x = self.patcher(x)  # (batch * chans, emb, patches)
-            return x.reshape(batch_size, n_chans, x.shape[-2], x.shape[-1]).permute(
-                0, 1, 3, 2
-            )
-        return x.reshape(batch_size, n_chans, -1, self.patch_size)
 class InceptionBlock(nn.Module):
     """
     Inception block module.

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: braindecode
-Version: 1.6.1.dev1049
+Version: 1.6.1.dev1052
 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>

@@ -15,7 +15,6 @@ from .deep4 import Deep4Net
 from .deepsleepnet import DeepSleepNet
 from .dgcnn import DGCNN
 from .eegconformer import EEGConformer
-from .eegdino import EEGDINO
 from .eeginception_erp import EEGInceptionERP
 from .eeginception_mi import EEGInceptionMI
 from .eegitnet import EEGITNet
@@ -85,7 +84,6 @@ __all__ = [
     "DeepSleepNet",
     "BrainModule",
     "EEGConformer",
-    "EEGDINO",
     "EEGPT",
     "InterpolatedEEGPT",
     "EEGInceptionERP",

@@ -56,4 +56,3 @@ InterpolatedBIOT,"Sleep Staging, Epilepsy",Classification,200,"chs_info, n_outpu
 InterpolatedEEGPT,General,Classification,256,"chs_info, n_outputs, n_times",25323137,"InterpolatedEEGPT(chs_info=<user>, n_outputs=4, n_times=1024)","Foundation Model,Attention/Transformer,Channel"
 InterpolatedLaBraM,General,"Classification, Embedding",200,"chs_info, n_outputs, n_times",5866196,"InterpolatedLaBraM(chs_info=<user>, n_outputs=4, n_times=1000)","Convolution,Foundation Model,Channel"
 InterpolatedSignalJEPA,"Motor Imagery, ERP, SSVEP",Embedding,128,"chs_info",3456882,"InterpolatedSignalJEPA(chs_info=<user>)","Convolution,Channel,Foundation Model"
-EEGDINO,General,Classification,200,"n_chans, n_outputs, n_times",4539698,"EEGDINO(n_chans=19, n_outputs=4, n_times=1000)","Foundation Model,Attention/Transformer"

@@ -432,7 +432,6 @@ models_mandatory_parameters: list[
         {"n_chans": 19, "n_times": 6000},
     ),
     ("DGCNN", ["n_chans", "n_outputs", "n_times", "chs_info"], None),
-    ("EEGDINO", ["n_chans", "n_outputs", "n_times"], None),
 ]
 ################################################################

@@ -14,7 +14,7 @@ from .attention import (
     MultiHeadAttention,
     SqueezeAndExcitation,
 )
-from .blocks import MLP, FeedForwardBlock, InceptionBlock, PatchTokenizer
+from .blocks import MLP, FeedForwardBlock, InceptionBlock
 from .convolution import (
     AvgPool2dWithConv,
     CausalConv1d,
@@ -68,7 +68,6 @@ __all__ = [
     "MLP",
     "FeedForwardBlock",
     "InceptionBlock",
-    "PatchTokenizer",
     "AvgPool2dWithConv",
     "CausalConv1d",
     "CombinedConv",

@@ -67,7 +67,6 @@ braindecode/models/deep4.py
 braindecode/models/deepsleepnet.py
 braindecode/models/dgcnn.py
 braindecode/models/eegconformer.py
-braindecode/models/eegdino.py
 braindecode/models/eeginception_erp.py
 braindecode/models/eeginception_mi.py
 braindecode/models/eegitnet.py

@@ -36,8 +36,6 @@ Enhancements
   variant of :class:`braindecode.models.EEGPT` built with
   :func:`~braindecode.models.interpolated.InterpolatedModel`.
   By `Pierre Guetschel`_.
-- Add :class:`braindecode.models.EEGDINO`, the EEG-DINO self-distillation
-  foundation model (Small/Medium/Large) with pretrained S/M weights. By `Bruno Aristimunha`_.
 API and behavior changes
 ========================

	@@ -0,0 +1 @@
1	+ __version__ = "1.6.1.dev1052"

Braindecode 1.6.1.dev1049__tar.gz → 1.6.1.dev1052__tar.gz

Braindecode 1.6.1.dev1049tar.gz → 1.6.1.dev1052tar.gz