neuraltrain.models.simplerconv.SimplerConvModel¶

class neuraltrain.models.simplerconv.SimplerConvModel(in_channels: int, out_channels: int | None, config: SimplerConv | None = None)[source][source]¶

Convolutional encoder inspired from BENDR/wav2vec2.0, with channel attention.

forward(x: Tensor, subject_ids: Tensor | None = None, channel_positions: Tensor | None = None) → Tensor[source][source]¶

Run the convolutional encoder.

Parameters:

x (Tensor) – Input of shape (B, C, T).
subject_ids (Tensor or None) – Per-example subject indices, shape (B,).
channel_positions (Tensor or None) – Normalised electrode coordinates, shape (B, C, D).

property receptive_field: int[source]¶: Compute the receptive field of the encoder. See https://distill.pub/2019/computing-receptive-fields/, Eq. 2