def evolution(gen_size, num_of_gens, num_of_parents, num_of_bests, epochs, to_print):
start_time = time.time()
cur_gen = []
prev_gen = []
n_gen = 0
best_org_fit = 0
best_org_weights = []
# initializing generation 0
for i in range(0, gen_size):
cur_gen.append(Organism())
# start evolution
while n_gen < num_of_gens:
n_gen += 1
prev_gen.clear()
# compute fitness for each organism on current generation
for org in cur_gen:
org.forward_propagation(X_train, Y_train)
prev_gen.append(org)
cur_gen.clear()
prev_gen = sorted(prev_gen, key=lambda x: x.fitness)
prev_gen.reverse()
# save the fitness of the best organism of current generation
cur_best_fit = prev_gen[0].fitness
# save the fitness and weights of the best organism until now
if cur_best_fit > best_org_fit:
best_org_fit = cur_best_fit
best_org_weights.clear()
for layer in prev_gen[0].layers:
best_org_weights.append(layer.get_weights()[0])
if to_print:
print('Generation: %1.f' % n_gen, '\tBest Fitness: %.4f' % cur_best_fit)
# crossover current generation
cur_gen = crossover(gen_size, prev_gen, num_of_bests, num_of_parents)
best_organism = Organism(best_org_weights)
best_organism.compile_train(epochs, X_train, Y_train, to_print)
# test
Y_hat = best_organism.predict(X_test)
Y_hat = Y_hat.argmax(axis=1)
end_time = time.time()
return accuracy_score(Y_test, Y_hat), end_time - start_time
