NEAT.py

from Population import *
from Species import *


def fitness(network):
    network.clear_state()
    err = 0
    network.set_input([0, 0])
    network.eval_asynch()
    err += np.abs(network.get_current_output()[0])
    network.clear_state()
    network.set_input([0, 1])
    network.eval_asynch()
    err += np.abs(network.get_current_output()[0] - 1)
    network.clear_state()
    network.set_input([1, 0])
    network.eval_asynch()
    err += np.abs(network.get_current_output()[0] - 1)
    network.clear_state()
    network.set_input([1, 1])
    network.eval_asynch()
    err += np.abs(network.get_current_output()[0])
    return (4-err)**2


def NEAT(fitness_fn, n_inputs, n_outputs, n_generations):
    population = Population(n_inputs, n_outputs)
    population.create()
    for generation in range(n_generations):
        population.evaluate(fitness_fn)
        population.speciate()
        print (generation, population.getBestChromosome().fitness)
        population.select()
        population.reproduce()
        population.evaluate(fitness_fn)
        print (generation, max(population.chromosomes).fitness)
        GeneE.clear_history()
        population.incrementGeneration()


NEAT(fitness, 2, 1, 2000)