torch.conv_tbc

torch.conv_tbc()

Applies a 1-dimensional sequence convolution over an input sequence. Input and output dimensions are (Time, Batch, Channels) - hence TBC.

Parameters:

• input – input tensor of shape $(\text{sequence length} \times batch \times \text{in\_channels})$

• weight – filter of shape ($\text{kernel width} \times \text{in\_channels} \times \text{out\_channels}$)

• bias – bias of shape ($\text{out\_channels}$)

• pad – number of timesteps to pad. Default: 0