cnn_graph.py

import numpy as np

from keras.datasets import cifar10
from keras.utils import to_categorical
from sklearn.model_selection import train_test_split

(train_X, train_y), (test_X, test_y) = cifar10.load_data()

train_X = train_X.astype('float32') / 255
test_X  = test_X.astype('float32') / 255
train_y = to_categorical(train_y)
test_y  = to_categorical(test_y)

def train(train_X, train_y):
    from keras.models import Model
    from keras.layers import Dense, Conv2D, MaxPooling2D, Flatten, Input, Dropout, BatchNormalization
    from keras.regularizers import l2

    # Convolution → Convolution → Pooling layers
    def ccp(x, filters, kernel_size, weight_decay, dropout):
        x = Conv2D(filters, kernel_size, padding='same', kernel_regularizer=l2(weight_decay), activation='relu')(x)
        x = BatchNormalization()(x)
        x = Conv2D(filters, kernel_size, padding='same', kernel_regularizer=l2(weight_decay), activation='relu')(x)
        x = BatchNormalization()(x)
        x = MaxPooling2D(pool_size=(2,2))(x)
        x = Dropout(dropout)(x)
        return x

    inputs = Input(shape=train_X.shape[1:])

    x = ccp(inputs, 32, (3,3), 1e-4, 0.2)
    x = Flatten()(x)
    y = Dense(10, activation='softmax')(x)

    m = Model(inputs=inputs, outputs=y)
    m.compile(
        loss='categorical_crossentropy',
        optimizer='adam',
        metrics=['accuracy']
    )
    h = m.fit(x=train_X, y=train_y, batch_size=32, epochs=20, validation_split=0.1)

    return m, h

def plot(h):
    """
    Great thanks to: http://parneetk.github.io/blog/cnn-cifar10/
    """
    import matplotlib.pyplot as plt
    fig, axs = plt.subplots(1, 2, figsize=(15,5))
    # accuracy
    axs[0].plot(range(1, len(h.history['acc']) + 1), h.history['acc'])
    axs[0].plot(range(1, len(h.history['val_acc']) + 1), h.history['val_acc'])
    axs[0].set_title('Accuracy')
    axs[0].set_ylabel('Accuracy')
    axs[0].set_xlabel('Epoch')
    axs[0].set_xticks(np.arange(1, len(h.history['acc']) + 1), len(h.history['acc'])/10)
    axs[0].legend(['trian', 'val'], loc='best')
    # loss
    axs[1].plot(range(1, len(h.history['loss']) + 1), h.history['loss'])
    axs[1].plot(range(1, len(h.history['val_loss']) + 1), h.history['val_loss'])
    axs[1].set_title('Loss')
    axs[1].set_ylabel('Loss')
    axs[1].set_xlabel('Epoch')
    axs[1].set_xticks(np.arange(1, len(h.history['loss']) + 1), len(h.history['loss'])/10)
    axs[1].legend(['trian', 'val'], loc='best')

    # plt.show()
    plt.savefig('result.png')

if __name__ == '__main__':
    model, history = train(train_X, train_y)
    plot(history)