def __init__(self, hidden_neurons, evolving_iterations, game_iterations, population_size):
table = [yamb.Column(i) for i in range(6)]
self.best_game = yamb.Yamb(table)
self.game_iterations = game_iterations
self.population_size = population_size
self.iterations = evolving_iterations
self.hidden_neurons = hidden_neurons
self.population = [NeuralNetwork(self.hidden_neurons, game_iterations) for _ in range(population_size)]
def __init__(self, populationSize):
self.__population_size = populationSize
self.__networks = []
self.__parameter_list = (3, 8, 3)
for i in range(self.__population_size):
self.__networks.append(NeuralNetwork(self.__parameter_list))
self.__best_network = self.__networks[0]
self.__generation = 1
def breed(individual1, individual2):
child = Individual(NeuralNetwork.get_new_model(board))
for layer in range(len(individual1.model.layers)):
# Breed the weights
w1 = individual1.model.layers[layer].get_weights()[0]
w_mask = np.random.randint(0,
int(random.random() * w1.size) + 1,
size=w1.shape).astype(np.bool)
w = np.copy(w1)
w[w_mask] = individual2.model.layers[layer].get_weights()[0][w_mask]
# Breed the biases
b1 = individual1.model.layers[layer].get_weights()[1]
b_mask = np.random.randint(0,
int(random.random() * b1.size) + 1,
size=b1.shape).astype(np.bool)
b = np.copy(b1)
b[b_mask] = individual2.model.layers[layer].get_weights()[1][b_mask]
child.model.layers[layer].set_weights(np.asarray([w, b]))
return child
def open():
nn = NeN.NNCreater(
keras.models.load_model("AI/NeuralNetwork/SavedModel.h5"))
return nn
def populate(individual_class, size=20):
models = []
for ignored in range(size):
models.append(
individual_class(NeuralNetwork.get_new_model(board, random=True)))
return models
generations = 20
population_size = 10
max_moves = 300
retain = 0.3
retain_chance = 0.1
mutate_chance = 0.2
mutation_severity = 0.1
if args.load is not None:
print("Loading population from file")
population = []
try:
with open(args.load, "rb") as file:
loaded = pickle.load(file)
for weights in loaded:
model = NeuralNetwork.get_new_model(board)
model.set_weights(weights)
population.append(Individual(model))
except IOError:
print("Error reading file")
raise
else:
population = populate(Individual, population_size)
best = []
for i in range(generations):
start = time.time()
# Get the new parents
parents, evaluation_stat = select(population)
# Short-circuit if not enough parents
def fill_population(self):
while len(self.population) < self.population_size:
self.population.append(NeuralNetwork(self.hidden_neurons, self.game_iterations))