quantlab.model.predefined.tcn#

Qlib’s TCN benchmark model: a temporal convolutional network.

TCNNet reproduces Qlib’s TCNModel (qlib/contrib/model/pytorch_tcn_ts.py over qlib/contrib/model/tcn.py, itself the reference implementation of Bai et al., “An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling”, 2018, by CMU Locus Lab, MIT License). TCNRegressor trains it with Qlib’s time-series regimen (QlibSequenceHead) and its Alpha158 benchmark settings.

Classes

Chomp1d

Drops the last chomp_size steps, the right padding of a causal convolution.

TCNNet

Qlib's TCNModel with an output of any width.

TCNRegressor

Qlib's TCN benchmark model on (bar, symbol) samples.

TemporalBlock

Two weight-normalised dilated causal convolutions with ReLU, dropout and a residual.