ssl_tools.models.layers.gru

Classes

GRUEncoder

Gate Recurrent Unit (GRU) Encoder.

Module Contents

class ssl_tools.models.layers.gru.GRUEncoder(hidden_size=100, in_channels=6, encoding_size=10, num_layers=1, dropout=0.0, bidirectional=True)

Bases: torch.nn.Module

Gate Recurrent Unit (GRU) Encoder. This class is a wrapper for the GRU layer (torch.nn.GRU) followed by a linear layer, in order to obtain a fixed-size encoding of the input sequence.

The input sequence is expected to be of shape [batch_size, in_channel, seq_len]. For instance, for HAR data in MotionSense Dataset:

  • in_channel = 6 (3 for accelerometer and 3 for gyroscope); and

  • seq_len = 60 (the number of time steps).

In forward pass, the input sequence is permuted to [seq_len, batch_size, in_channel] before being fed to the GRU layer. The output of forward pass is the encoding of shape [batch_size, encoding_size].

Parameters

hidden_sizeint, optional

The number of features in the hidden state of the GRU, by default 100

in_channel: int, optional

The number of input features (e.g. 6 for HAR data in MotionSense Dataset), by default 6

encoding_sizeint, optional

Size of the encoding (output of the linear layer).

num_layersint, optional

Number of recurrent layers. E.g., setting num_layers=2 would mean stacking two GRUs together to form a stacked GRU, with the second GRU taking in outputs of the first GRU and computing the final results. By default 1

dropoutfloat, optional

If non-zero, introduces a Dropout layer on the outputs of each GRU layer except the last layer, with dropout probability equal to dropout. Default: 0

bidirectionalbool, optional

If True, becomes a bidirectional GRU, by default True

forward(x)
Parameters:

  • hidden_size (int)

  • in_channels (int)

  • encoding_size (int)

  • num_layers (int)

  • dropout (float)

  • bidirectional (bool)