A PyTorch implementation of Clockwork RNN (CW-RNN) proposed by Koutnik et al. in 2014.
data.py contains a data object class that preprocesses audio data for a sequence generation task. See original paper for more details on the task.
main.py trains a model of choice and reports the losses.
model.py implements CW-RNN as well as RNN/LSTM/GRU baselines.
Loss curve and output predictions after training CW-RNN for 2000 epochs with 200 hidden units.
