DilatedTCN

class daart.models.tcn.DilatedTCN(hparams, type='encoder', in_size=None, hid_size=None, out_size=None)

Bases: BaseModel

Temporal Convolutional Model with dilated convolutions and no temporal downsampling.

Code adapted from: https://www.kaggle.com/ceshine/pytorch-temporal-convolutional-networks

Methods Summary

build_decoder(in_size, hid_size, out_size)	Construct the decoder using hparams.
build_encoder(in_size, hid_size, out_size)	Construct encoder model using hparams.
forward(x, **kwargs)	Process input data.

Methods Documentation

build_decoder(in_size, hid_size, out_size)

Construct the decoder using hparams.

build_encoder(in_size, hid_size, out_size)

Construct encoder model using hparams.

forward(x, **kwargs)

Process input data.

Parameters:

x (torch.Tensor object) – input data of shape (n_sequences, sequence_length, n_markers)

Returns:

shape (n_sequences, sequence_length, n) where n is the embedding dimension if an encoder, or n_markers if a decoder/predictor

Return type:

torch.Tensor