EchoStateNetwork

A python framework for Echo State Network applications.

Documentation

The documentation can be found here.

Installation

Download EchoStateNetwork.py and place it, where you want to import it from.

Current capabilities

• Supervised Learning

  • excite, reg_fit and validate can be used in combination to complete supervised learning tasks.
  • Another easier way to execute a supervised learning job is to call session method, where one passes all training and validation data at once as arguments.

• Reinforcement Learning

  • Examples of PPO and DQN algorithms will be provided soon.

• Learning with Batches

  Use ESNX, ESNS or ESNN to enable (mini)batching.

• Learning in Parallel

  Use ESNS or ESNN to learn with an ensemble of Echo State Networks and (mini)batching. This is useful when using vectorized environments in reinforcement learning applications.

• Gradient Based Optimization

  • Use ESNN to make the reservoir computer compatible with PyTorch.
  • It is possible to use PyTorch with the rest of the reservoir objects (ESN, ESNX, ESNS, ESNN) as well without the usage of gradients but regression.
  • ESNN and ESNS work with PyTorch only whereas the other reservoir objects also work with Numpy.