tension 0.2

A Python package for training chaotic Echo State and Spiking RNNs

tension

The tension package is a Python package for building and FORCE training chaotic recurrent neural networks. tension is built to integrate seamlessly with TensorFlow/Keras, a widely used software package for training artificial neural networks.

Code and documentation

Installation guide, contributing guide, and API reference are available on the documentation website

Example notebooks for reproducing the experiments in the associated software report can also be found on the documentation website and in the examples folder. These notebooks can be run either in Google Colab or in a local jupyter notebook.

Documentation for TensorFlow/Keras, on which tension is based and interoperable with, can be found at the TensorFlow guide.

Dependencies

tension depends on Python >= 3.7, tensorflow>=2.5, numpy and matplotlib.

Installation

Local installation

We recommend installing tension into a conda environment

conda create -n tension python=3.7

followed by

conda activate tension

From GitHub

Clone this repo using

git clone https://github.com/zhenruiliao/tension.git

Change into the tension directory and install using pip

cd tension/
pip install -e .

From PyPI

pip install tension

Google Colab

To quickly get started with tension, the package can also be installed in Google Colab using the following commands

!git clone https://github.com/zhenruiliao/tension.git tension
!pip install -e tension

The runtime must be restarted for the package to become importable

Contributing

Bug reports, feature requests, and pull requests are welcome and encouraged! Use the Issues and Pull requests tabs to open new issues or pull requests. Always be kind and respectful.

License

tension is provided under the MIT License. TensorFlow is provided under the Apache 2.0 license.

Acknowledgments

Based on work by David Sussillo and Larry Abbott https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2756108/

With thanks to James Priestley for the package name.