evonet

Evolutionary Neural Network library

NOTE: Evonet is still in the initial planning & implementation stages, I will be developing this library over the course of the next few months (November 2017-January 2018).

What?

Evonet is a library that makes the creation of a fully-connected neural network (AKA "multilayer perceptron") super simple, and provides an easy way to evolve this network over time using a built-in genetic algorithm.

Aspects of both the neural network and the genetic elgorithm can be tweaked and modified, and custom fitness functions and evolution thresholds can be specified.

Why?

As part of my undergraduate studies, I built an 'artifical intelligence' that learned how to play different computer games from basic sensor inputs. Although many great existing libraries exist for both neural networks and genetic algorithms, I decided to make my own which combined them both into a single-purpose library.

However, Evonet does also support standard reinforcement learning through backpropogation, as I also had a system whereby the NN would learn by previous gameplay from a human.

How?

Evonet is available via npm.

npm install --save evonet