P = 40 # population of neural network pool
N = 10 # neurons number in a neural netowrk
S = 1 # stimuli pool size
G = 30 # generations of evolution
I = 10 * 10 ** 3 # iterations of stimuli on neural network
strengthen_rate = 0.001
#
print('#population: %s #neuron: %s #propagation: %s #generation: %s #iteration: %s lr: %s' % (
P, N, S, G, I, strengthen_rate))
np.random.seed()
stimu_pool = StimuliPool(N, S)
print((' stimulation ').center(100, '-'))
stimu_pool.info()
env = Environment()
# nn_pool = NeuralNetwork.creations(P, N)
gene_pool = Gene.creations(P, N)
for g in range(G):
# One Generation
print((' generation %s ' % g).center(100, '-'))
# continiously stimulate NNs with randomly picked stimuli for I iterations
stats_l = env.evaluate_fitness(gene_pool, stimu_pool, NeuralNetwork, I, strengthen_rate)
# env.output_stats(nn_pool)
# env.store_stats(nn_pool)
# print(stats_l[0]['accuracy'])
new_gene_pool = env.build_new_gene_pool(P, gene_pool, stats_l, strength_threshhold=.15)
if not new_gene_pool:
break
# print(mating_pool)
gene_pool = new_gene_pool
# for nn in nn_pool: print nn.i
gene_pool = []
for _ in range(P):
new_gene = Gene.crossover(random.sample(mating_pool, 2))
gene_pool.append(new_gene)
# mutation
return gene_pool
def output_stats(self, nn_pool):
pass
def store_stats(self, nn_pool):
pass
if __name__ == "__main__":
env = Environment()
gene_pool = Gene.creations(1, 7)
stats_l = []
for gene in gene_pool:
tmp_m = np.random.rand(7, 7)
connection_strength_m = np.where(gene.connections == 1,tmp_m, -1)
stat = {}
stat['strength_matrix'] = connection_strength_m
stat['accuracy'] = np.random.rand()
stats_l.append(stat)
mating_pool = env.select_mating_pool(gene_pool, stats_l, mating_pool_size=10, strength_threshhold=.1)
print(mating_pool)