pyneat

This is a python implementation of the original NEAT algorithm, formally known as Neuro-Evolution of Augmenting Topologies. In summary, NEAT is an evolutionary strategy for optimizing neural networks with non-static structures.

Although the library was created for use in experimentation with the game of snake, https://github.com/Veemon/deep-snake, it is a fully functional - user focused library.

To get an idea of what I mean by user focused, refer to the example code:
examples/input_detector.py

Install

To install simply run install.sh .

Basic Usage

The basic workflow is as follows. Import the library.

from pyneat import pyneat

Create a fitness function for the agent, within file scope.

def fitness_function_foo(self):
  print("I am a genome, here is my representation.", self)
  self.fitness = 1.0
  return True, self 

Create a Gene Pool, this will be the primary interface to the algorithm. Upon calling evolve, the evolutionary loop will run depending on the parameters provided.

gene_pool = pyneat.GenePool(...)
gene_pool.init(...)
gene_pool.evolve(...)

There are a lot of parameters, so if you are new to the algorithm I do recommend reading the original paper here: http://nn.cs.utexas.edu/downloads/papers/stanley.ec02.pdf