def load(self, filename):
self.critic_optimizer = load_model(filename + "_critic",
self.critic_optimizer)
self.critic_optimizer = jax.device_put(self.critic_optimizer)
self.critic_target = self.critic_target.replace(
params=self.critic_optimizer.target.params)
self.actor_optimizer = load_model(filename + "_actor",
self.actor_optimizer)
self.actor_optimizer = jax.device_put(self.actor_optimizer)
self.actor_target = self.actor_target.replace(
params=self.actor_optimizer.target.params)
def report(self, x, y, title=''):
model = load_model(FER_MODEL_PATH)
y_predicted = model.predict(x)
y_predicted = (y_predicted == y_predicted.max(axis=1)[:, None]).astype(
np.int8)
result = np.zeros((len(self.labels), len(self.labels)))
for i in range(y.shape[0]):
desired_emotion_index = np.where(y[i] == 1)[0][0]
predicted_emotion_index = np.where(y_predicted[i] == 1)[0][0]
result[desired_emotion_index][predicted_emotion_index] += 1
row_format = "{:>15}" * (len(self.labels) + 1)
s = '| ' + row_format.format("", *self.labels) + ' |\n'
for emo, row in zip(self.labels, result):
row_precent = ((row * 100.) /
np.sum(row)) if row.any() else np.zeros_like(row)
final_row = [
"%d/%5.2f" % (row[i], row_precent[i]) for i in range(len(row))
]
s += '| ' + row_format.format(emo, *final_row) + ' |\n'
table_width = len(s.split('\n')[0])
header_text = " Report: %s " % title
half_header_width = (table_width - len(header_text)) // 2
header = ('-' * half_header_width) + header_text + ('-' *
half_header_width)
header += '-' * (table_width - len(header))
footer = '=' * table_width
s = header + '\n' + s + footer
return s
def on_train_end(self, logs=None):
model = load_model(FER_MODEL_PATH)
print('Best result on train set:')
self._evaluate(model, self.x_train, self.y_train)
print('Best result on dev set:')
self._evaluate(model, self.x_dev, self.y_dev)
print('Best result on test set:')
self._evaluate(model, self.x_test, self.y_test)
print('Best result on all sets:')
self._evaluate(model, self.x_all, self.y_all)
print("epoch: ", epoch)
####################### ALL POPULATION FOR 1 EPOCH #############################################
for pop_index, current_robot in enumerate(population_array):
#set title with number of epoch and robot
pygame.display.set_caption("Epoch: " + str(epoch) + " Robot: " +
str(pop_index))
collision_robot_3lvl = [
] # save collision for the single robot for all levels
score_robot_3lvl = [
] # save score for the single robot but for all levels
#initialize weights for current robot or load them
if LOAD and LOAD_EPOCH == epoch:
neuralNetwork.weights_0L, neuralNetwork.weights_1L = save.load_model(
epoch, pop_index)
else:
neuralNetwork.weights_0L = deepcopy(current_robot[0])
neuralNetwork.weights_1L = deepcopy(current_robot[1])
save.save_model_weight(epoch, pop_index, neuralNetwork.weights_0L,
neuralNetwork.weights_1L)
####################### N LEVEL FOR 1 ROBOT ##############################################
for new_map, new_position in zip(maps_list, positions_list):
#reset level position and dust
environment, robot = init_new_map(new_map, new_position)
dust = du.Dust(screen, DUST_SIZE)
collision_avoided = 0
####################### SINGLE LEVEL ################################################
for steps in range(MAP_STEPS):
##################### MANUAL DRIVE ######################################