eegproc.deep_learning.unsupervised.Convolutions.CNN2D module

class eegproc.deep_learning.unsupervised.Convolutions.CNN2D.CNN2DDecoder(timesteps, n_channels, n_bands, t_down, conv_filters, temporal_pool_sizes, emb_dim=128, dropout=0.1, activation='relu', use_batch_norm=True, kernel_sizes=(3, 3), spatial_pool_sizes=None, name='decoder_2dcnn', **kwargs)[source]

Bases: Model

Spatially structured mirror decoder for CNN2DEncoder.

Parameters:

  • timesteps (int)

  • n_channels (int)

  • n_bands (int)

  • t_down (int)

  • conv_filters (tuple[int, ...])

  • temporal_pool_sizes (tuple[int, ...] | None)

  • emb_dim (int, default: 128)

  • dropout (float, default: 0.1)

  • activation (str, default: 'relu')

  • use_batch_norm (bool, default: True)

  • kernel_sizes (tuple[int, int] | tuple[tuple[int, int], ...], default: (3, 3))

  • spatial_pool_sizes (tuple[int, int] | tuple[tuple[int, int] | None, ...] | None, default: None)

  • name (str, default: 'decoder_2dcnn')

call(inputs, training=False)[source]
Parameters:

training (bool, default: False)

compute_output_shape(input_shape)[source]
fix_length(x)[source]
Parameters:

x (Tensor)

Return type:

Tensor

classmethod from_encoder(encoder, name='decoder_2dcnn')[source]
Parameters:

  • encoder (CNN2DEncoder)

  • name (str, default: 'decoder_2dcnn')

Return type:

CNN2DDecoder

get_config()[source]

Returns the config of the object.

An object config is a Python dictionary (serializable) containing the information needed to re-instantiate it.

Return type:

dict

property n_features: int
class eegproc.deep_learning.unsupervised.Convolutions.CNN2D.CNN2DEncoder(timesteps, t_down, n_channels=14, n_bands=4, conv_filters=(32, 64, 128), kernel_sizes=((3, 3), (3, 3), (2, 2)), spatial_pool_sizes=None, temporal_pool_sizes=None, emb_dim=128, dropout=0.1, activation='relu', use_batch_norm=True, name='encoder_2dcnn', **kwargs)[source]

Bases: BaseEncoder

2D CNN encoder that preserves channel-band position information.

Inputs are reshaped from (batch, time, channels * bands) to (batch, time, channels, bands, 1). Spatial Conv2D blocks operate on each timestep independently. The remaining spatial grid is flattened in a fixed order and projected to emb_dim instead of being globally averaged, so electrode and frequency-band location remain available to the latent sequence.

Parameters:

  • timesteps (int)

  • t_down (int)

  • n_channels (int, default: 14)

  • n_bands (int, default: 4)

  • conv_filters (tuple[int, ...], default: (32, 64, 128))

  • kernel_sizes (tuple[int, int] | tuple[tuple[int, int], ...], default: ((3, 3), (3, 3), (2, 2)))

  • spatial_pool_sizes (tuple[int, int] | tuple[tuple[int, int] | None, ...] | None, default: None)

  • temporal_pool_sizes (tuple[int, ...] | None, default: None)

  • emb_dim (int, default: 128)

  • dropout (float, default: 0.1)

  • activation (str, default: 'relu')

  • use_batch_norm (bool, default: True)

  • name (str, default: 'encoder_2dcnn')

call(inputs, training=False)[source]

Run the encoder forward pass.

Parameters:

training (bool, default: False)

get_config()[source]

Return serializable configuration for the base encoder.

Return type:

dict

property n_features: int

Number of input features per timestep.