ammar-raneez / liquid-time-stochasticity-networks

Code repository for Liquid Time-stochasticity networks (LTSs)

Liquid Time-stochasticity Networks (LTSs)

This is the official repository for Liquid TIme-stochasticity networks described in paper: https://doi.org/10.1109/CCWC57344.2023.10099071

This implementation utilizes the Euler Maruyama solver to perform forward propagation and relies on the conventional backpropagation through-time (BPTT) to train the models.

Prerequisites

The architecture was built using Keras and TensorFlow 2.0+ and Python 3+ on the Windows 11 machine.

Experiments

experiments/experiments.ipynb demonstrates a couple of experiments attempting to model bitcoin prices.

liquid-time-stochasticity-networks's Issues

BTC

Nice work Ammar!

Did you try to use the code with a demo account? I would like to know your opinion and results :-D

If yes I will adopt it for yfinance or oanda.

Exploding Gradient

Current implementation of the algorithm can often run into the Exploding Gradient issue.

Following mitigation must be taken:

• Gradient Clipping
  • https://cnvrg.io/gradient-clipping/
  • https://machinelearningmastery.com/how-to-avoid-exploding-gradients-in-neural-networks-with-gradient-clipping/
  • Clip Adam optimizer / LTC cell
• Regularization
• LSTM