بررسی شبکه عصبی کانولوشن (CNN) — بخش دوم

1X = np.array([[0, 0, 1], [1, 1, 1], [1, 0, 1], [0, 1, 1]])
2    Y = np.array([0, 1, 1, 0])

1W = np.random.rand(1, X.shape[1] + 1)

1def delta_weight(eta, true_label, predicted_label, x):
2    """
3
4    :param eta: Learning rate
5    :param true_label:
6    :param predicted_label:
7    :param x:
8    """
9    lambda_param = -1
10    delta_w = lambda_param * eta * (predicted_label - true_label) * x
11    return delta_w

1def training_perceptron(eta, X, Y, number_of_epoch=5000):
2    """
3    :param eta: learning rate of perceptron
4    :param X: the feature set for training
5    :param Y: the target value against feature set
6    """
7
8    logging.info('Training Config:\nNumber_of_epoch: {} Eta: {}'.format(number_of_epoch, eta))
9    W = np.random.rand(1, X.shape[1] + 1)
10    loss_log = []
11    X = np.insert(X, 2, values=1, axis=1)
12    for epoch in range(number_of_epoch):
13        X, Y = shuffle(X, Y)
14        loss = 0.0
15
16        for index, (feature_row, true_label) in enumerate(zip(X, Y)):
17            theta = np.dot(np.array(feature_row), W.T)
18            # predicted_output = 1 if theta > 0 else 0
19            predicted_output = float(theta)
20
21            loss += (true_label - predicted_output) ** 2
22            delta_W = [delta_weight(eta, true_label, predicted_output, x) for x in feature_row]
23            logging.debug([feature_row, true_label, np.around(W, decimals=1), predicted_output, theta, delta_W])
24
25            W = np.add(W, delta_W)
26        if epoch % 50 == 0:
27            loss_log.append([epoch, loss])
28        logging.info('Epoch Summary : Epoch: {} Loss: {}'.format(epoch, loss))
29        if loss < 0.001:
30            break
31
32        time.sleep(0.001)
33    df = pd.DataFrame(loss_log, columns=['Epoch', 'Loss'])
34    logging.info(df)
35    df.to_csv('training_log.csv')
36    return number_of_epoch

1"""
2eta: learning rate of the training
3lambda_param: to limit the maximum delta change in weights
4"""