Neuroevolution in 50 lines of code

sandbox.py

import random
from deap import creator, base, tools, algorithms
from scoop import futures
toolbox = base.Toolbox()
toolbox.register("map", futures.map)

creator.create("FitnessMax", base.Fitness, weights=(1.0,))
creator.create("Individual", list, fitness=creator.FitnessMax)
number_of_parameters = 100
size_of_population = 100
mutation_size=0.05
toolbox.register("attribute", random.randint,0,1e8)
toolbox.register("individual", tools.initRepeat, creator.Individual, toolbox.attribute, n=number_of_parameters)
toolbox.register("population", tools.initRepeat, list, toolbox.individual)

import tensorflow as tf
import numpy as np
tf.reset_default_graph()
x=tf.placeholder(tf.float32, shape=(None,5))
W=tf.Variable(tf.random_normal([5,5]))
y = tf.matmul(x,W)
y_true=tf.placeholder(tf.float32, shape=(None,5))
error = tf.losses.mean_squared_error(y_true,y)
sess=tf.Session()
sess.run(tf.global_variables_initializer())
trainX = np.random.randn(10,5)
trainY = np.dot(trainX,np.random.rand(5,5))

def EvaluateNet(seeds):
    vals=sess.run(tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES))
    np.random.seed(seeds[0])
    for i,val in enumerate(vals):
        vals[i] = np.random.randn(vals[i].shape[0],vals[i].shape[1])
    for seed in seeds[1:]:
        np.random.seed(seed)
        for i,val in enumerate(vals):
            vals[i] += mutation_size*np.random.randn(vals[i].shape[0],vals[i].shape[1])
    ops=[]
    for i,var in enumerate(tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES)):
        ops.append(var.assign(vals[i]))
    sess.run(ops)
    return sess.run(-error,feed_dict={x:trainX,y_true:trainY}),

toolbox.register("evaluate", EvaluateNet)
toolbox.register("select", tools.selTournament, tournsize=3)
population = toolbox.population(n=size_of_population)

def mutate(individual):
    return individual.append(random.randint(0,1e8))

if __name__ == '__main__':
    NGEN=20
    offspring = toolbox.select(population, len(population))
    for gen in range(NGEN):
        print(gen)
        offspring = list(map(toolbox.clone, offspring))
        mutants = list(map(mutate,offspring))
        fits = list(map(toolbox.evaluate, offspring))
        for fit, ind in zip(fits, offspring):
            ind.fitness.values = fit
        population = toolbox.select(list(offspring), k=len(population))
    top10 = tools.selBest(population, k=10)
    print(toolbox.evaluate(top10[1]))