def train_3(weights, moderate_constant, max_error):
board = experiment_generator()
game_trace = get_game_trace_with_old_vesion(board, weights, weights)
training_examples = get_training_examples(game_trace, weights)
weights = gen(training_examples, weights, moderate_constant)
error = squared_error(training_examples, weights)
return weights, error, error <= max_error
def test_squared_errors2(self):
logging.basicConfig(filename='./logs/test_with_old_version_mu02.log', level=logging.INFO)
logging.info('------------------------------------------------------------------------------------------------')
logging.info('Started')
moderate_constant = 0.2
iterations = 50
weights = [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
old_weights = weights
errors = []
won = 0
lost = 0
for i in range(iterations):
board = experiment_generator()
logging.info('Obteniendo traza del juego...')
game_trace = get_game_trace_with_old_vesion(board, weights, old_weights)
logging.info(f'Se obtuvieron {game_trace.__len__()} tuplas')
training_examples = get_training_examples(game_trace, weights)
board_features = training_examples[training_examples.__len__() - 1][0]
logging.info(training_examples[training_examples.__len__() - 1])
if board_features[6] >= 1 :
won = won + 1
elif board_features[13] >= 1 :
lost = lost + 1
logging.info('Ajustando pesos...')
old_weights = weights
weights = gen(training_examples, weights, moderate_constant)
errors.append(squared_error(training_examples, weights))
logging.info(weights)
# if errors[i -1] < errors[i]:
# moderate_constant = max(0.1, moderate_constant - 0.1)
# logging.info(moderate_constant)
for i in range(iterations):
logging.info('Error {}: {}'.format(i, errors[i]))
# for i in range(iterations-1):
# self.assertGreaterEqual(errors[i], errors[i+1],
# f'El error {i} no es mayor o igual que el error {i+1}, los errores deben decrecer')
logging.info(f'Ganados: {won}')
logging.info(f'Perdidos: {lost}')
logging.info('Finished')
logging.info('------------------------------------------------------------------------------------------------')
return weights
