This is a quick port of Phased LSTM to Pytorch. Original is here: https://github.com/dannyneil/public_plstm
In general, if you are using ~1000 timesteps or more in your input sequence, or you have asynchronously sampled data, you can benefit from PLSTM.
To run, simply execute:
python main.py
This will train an LSTM and Phased LSTM on the asynchronous sin wave classification task.
Requires Pytorch. Currently tested on torch==1.8.0.