Using logsignatures as features or layers in a recurrent neural network

This code is for paper Learning stochastic differential equations using RNN with log signature features by Hao, Shujian, Terry, Weixin 2019

1. sde_learning: Codes for synthetic SDE data learning

2. Pen-digit_learning: Codes for Pen-digit data learning

3. gesture_learning: Codes for Chalearn 2013 gesture data learning

4. action_learning: Codes for NTURGB+D 120 learning

Python 3.7 was used. The following external packages were used, and may be installed via pip3 install -r requirements.txt.

iisignature==0.23 for calculating signatures. It is CPU-only; consequently signature-based models are currently bottlenecked here.

jupyter==1.0.0

Keras==2.2.4

esig==0.6.31

matplotlib==2.2.4

scikit-learn==0.20.3