import sys import time import math import random def sigmoid(x): return 2 / (1 + math.exp(-4.9)) class Neuron(): weights = [] def __init__(self, connections, read=[]): if read == []: self.weights = [random.randint(-127, 127) for i in range(connections)] else: self.weights = [binToInt(read[i]) for i in range(connections)] def activate(self, connections): temp = 0 for i in range(len(self.weights)): temp += (self.weights[i] / 100.0) * connections[i] return int(temp > 0) # return sigmoid(sum([(self.weights[i] / 100.0) * connections[i] for i in range(len(self.weights))])) def encode(self): return ''.join([intToBin(self.weights[i]) for i in range(len(self.weights))]) class Layer(): neurons = [] def __init__(self, nodes, inputs, read=""): if read == "": self.neurons = [Neuron(inputs) for i in range(nodes)] else: temp = [read[i:i + 8] for i in range(0, len(read), 8)] self.neurons = [ Neuron(inputs, temp[inputs * i:inputs * (i + 1)]) for i in range(nodes)] def activate(self, connections): return [self.neurons[i].activate(connections) for i in range(len(self.neurons))] def encode(self): return ''.join([self.neurons[i].encode() for i in range(len(self.neurons))]) class NeuralNet(): layers = [] def __init__(self, size, inputs, read=""): if read == "": self.layers = [Layer(size[i], inputs if i == 0 else size[ i - 1]) for i in range(len(size))] else: temp = read.split(',') self.layers = [Layer(size[i], inputs if i == 0 else size[ i - 1], temp[i]) for i in range(len(size))] def activate(self, inputs): connections = self.layers[0].activate(inputs) for i in range(1, len(self.layers)): connections = self.layers[i].activate(connections) return connections def encode(self): return ','.join([self.layers[i].encode() for i in range(len(self.layers))]) def breed(a, b): return ''.join([a[i] if random.random() < .5 else b[i] for i in range(len(a))]) def mutate(g, rate): return ''.join([g[i] if random.random() > rate or g[i] == ',' else str(int(not(int(g[i])))) for i in range(len(g))]) def binToInt(b): return int(b[1:], 2) * (int(b[0], 2) * -2 + 1) def intToBin(i): return ('1' + '0' * (7 - len(bin(i)[3:])) + bin(i)[3:] if i < 0 else '0' * (8 - len(bin(i)[2:])) + bin(i)[2:]) def main(): networkSize = [2, 4, 4, 1] genSize = 100 mutationRate = .001 gen = 0 genIterator = 0 genome = [(0, "") for i in range(genSize)] dna = '' network = NeuralNet(networkSize, 3, dna) for q in range(10000): error = 0.0 inputs = [[1.0,1.0,1.0],[1.0,0.0,1.0],[1.0,1.0,0.0],[1.0,0.0,0.0]] outputs = [0.0,1.0,1.0,0.0] for i in range(4): n = network.activate(inputs[i]) error+=abs(n[0]-outputs[i]) genome[genIterator] = (-error+5, network.encode()) genIterator += 1 # print "Organism #" + str(genIterator) + " Fitness: " + # str(genome[genIterator-1][0]) if genIterator == genSize: gen += 1 genome.sort() print "Generation: " + str(gen) + " Fitness: " + str(genome[genSize - 1]) genome = [(0, mutate(breed(genome[int(random.triangular(genSize * .75, genSize, genSize))][1], genome[ int(random.triangular(genSize * .75, genSize, genSize))][1]), mutationRate)) for i in range(genSize)] genIterator = 0 network = NeuralNet(networkSize, 3, genome[genIterator][1]) if __name__ == "__main__": main()