pclib 2.0.0b2

A torch-like package for building Predictive Coding Neural Networks.

PCLib

PCLib is a python package with a torch-like API for building and training Predictive Coding Networks.

Documentation can be found here.

The package includes a fully-connected layer implementation, as well as a convolutional one. Both are customisable and can be used together or separately for building neural networks.

The package also includes a helper class for constructing fully-connected PCNs. This class has been designed to be extremely customiseable such that the network it builds can be used in a wide range of tasks: supervised/unsupervised, classic/inverted, etc. There is also a CNN class, however it is not customisable in shape. For more detailed explanations, please see the documentation.

Installation

pip install pclib

Example usage

In the examples folder you will find two different classification tasks which demonstrate the usage of this package.