from __future__ import print_function
import numpy as np
np.random.seed(1337)

from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Convolution2D, MaxPooling2D
from keras.utils import np_utils
from keras import backend as K

batch_size = 128
nb_classes = 10
nb_epoch = 12

# 输入图像的维度，此处是mnist图像，因此是28*28
img_rows, img_cols = 28, 28
# 卷积层中使用的卷积核的个数
nb_filters = 32
# 池化层操作的范围
pool_size = (2,2)
# 卷积核的大小
kernel_size = (3,3)
# keras中的mnist数据集已经被划分成了60,000个训练集，10,000个测试集的形式，按以下格式调用即可
(X_train, y_train), (X_test, y_test) = mnist.load_data()

# 后端使用tensorflow时，即tf模式下，
# 会将100张RGB三通道的16*32彩色图表示为(100,16,32,3)，
# 第一个维度是样本维，表示样本的数目，
# 第二和第三个维度是高和宽，
# 最后一个维度是通道维，表示颜色通道数
X_train = X_train.reshape(X_train.shape[0], img_rows, img_cols, 1)
X_test = X_test.reshape(X_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)

# 将X_train, X_test的数据格式转为float32
X_train = X_train.astype('float32')
X_test = X_test.astype('float32')
# 归一化
X_train /= 255
X_test /= 255
# 打印出相关信息
print('X_train shape:', X_train.shape)
print(X_train.shape[0], 'train samples')
print(X_test.shape[0], 'test samples')

X_train shape: (60000, 28, 28, 1)
60000 train samples
10000 test samples

# 将类别向量(从0到nb_classes的整数向量)映射为二值类别矩阵，
# 相当于将向量用one-hot重新编码
Y_train = np_utils.to_categorical(y_train, nb_classes)
Y_test = np_utils.to_categorical(y_test, nb_classes)

# 建立序贯模型
model = Sequential()

# 卷积层，对二维输入进行滑动窗卷积
# 当使用该层为第一层时，应提供input_shape参数，在tf模式中，通道维位于第三个位置
# border_mode：边界模式，为"valid","same"或"full"，即图像外的边缘点是补0
# 还是补成相同像素，或者是补1
model.add(Convolution2D(nb_filters, kernel_size[0] ,kernel_size[1],
                        border_mode='valid',
                        input_shape=input_shape))
model.add(Activation('relu'))

# 卷积层，激活函数是ReLu
model.add(Convolution2D(nb_filters, kernel_size[0], kernel_size[1]))
model.add(Activation('relu'))

# 池化层，选用Maxpooling，给定pool_size，dropout比例为0.25
model.add(MaxPooling2D(pool_size=pool_size))
model.add(Dropout(0.25))

# Flatten层，把多维输入进行一维化，常用在卷积层到全连接层的过渡
model.add(Flatten())

# 包含128个神经元的全连接层，激活函数为ReLu，dropout比例为0.5
model.add(Dense(128))
model.add(Activation('relu'))
model.add(Dropout(0.5))

# 包含10个神经元的输出层，激活函数为Softmax
model.add(Dense(nb_classes))
model.add(Activation('softmax'))

# 输出模型的参数信息
model.summary()
# 配置模型的学习过程
model.compile(loss='categorical_crossentropy',
              optimizer='adadelta',
              metrics=['accuracy'])

____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
convolution2d_3 (Convolution2D)  (None, 26, 26, 32)    320         convolution2d_input_2[0][0]      
____________________________________________________________________________________________________
activation_5 (Activation)        (None, 26, 26, 32)    0           convolution2d_3[0][0]            
____________________________________________________________________________________________________
convolution2d_4 (Convolution2D)  (None, 24, 24, 32)    9248        activation_5[0][0]               
____________________________________________________________________________________________________
activation_6 (Activation)        (None, 24, 24, 32)    0           convolution2d_4[0][0]            
____________________________________________________________________________________________________
maxpooling2d_2 (MaxPooling2D)    (None, 12, 12, 32)    0           activation_6[0][0]               
____________________________________________________________________________________________________
dropout_3 (Dropout)              (None, 12, 12, 32)    0           maxpooling2d_2[0][0]             
____________________________________________________________________________________________________
flatten_2 (Flatten)              (None, 4608)          0           dropout_3[0][0]                  
____________________________________________________________________________________________________
dense_3 (Dense)                  (None, 128)           589952      flatten_2[0][0]                  
____________________________________________________________________________________________________
activation_7 (Activation)        (None, 128)           0           dense_3[0][0]                    
____________________________________________________________________________________________________
dropout_4 (Dropout)              (None, 128)           0           activation_7[0][0]               
____________________________________________________________________________________________________
dense_4 (Dense)                  (None, 10)            1290        dropout_4[0][0]                  
____________________________________________________________________________________________________
activation_8 (Activation)        (None, 10)            0           dense_4[0][0]                    
====================================================================================================
Total params: 600,810
Trainable params: 600,810
Non-trainable params: 0
____________________________________________________________________________________________________

# 训练模型
model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,
          verbose=1, validation_data=(X_test, Y_test))

# 按batch计算在某些输入数据上模型的误差
score = model.evaluate(X_test, Y_test, verbose=0)

Train on 60000 samples, validate on 10000 samples
Epoch 1/12
60000/60000 [==============================] - 18s - loss: 0.3675 - acc: 0.8886 - val_loss: 0.0877 - val_acc: 0.9722
Epoch 2/12
60000/60000 [==============================] - 13s - loss: 0.1346 - acc: 0.9598 - val_loss: 0.0623 - val_acc: 0.9802
Epoch 3/12
60000/60000 [==============================] - 13s - loss: 0.1039 - acc: 0.9691 - val_loss: 0.0527 - val_acc: 0.9837
Epoch 4/12
60000/60000 [==============================] - 13s - loss: 0.0887 - acc: 0.9736 - val_loss: 0.0462 - val_acc: 0.9849
Epoch 5/12
60000/60000 [==============================] - 13s - loss: 0.0778 - acc: 0.9763 - val_loss: 0.0420 - val_acc: 0.9860
Epoch 6/12
60000/60000 [==============================] - 13s - loss: 0.0698 - acc: 0.9794 - val_loss: 0.0383 - val_acc: 0.9871
Epoch 7/12
60000/60000 [==============================] - 14s - loss: 0.0659 - acc: 0.9802 - val_loss: 0.0374 - val_acc: 0.9868
Epoch 8/12
60000/60000 [==============================] - 14s - loss: 0.0616 - acc: 0.9818 - val_loss: 0.0385 - val_acc: 0.9877
Epoch 9/12
60000/60000 [==============================] - 14s - loss: 0.0563 - acc: 0.9829 - val_loss: 0.0338 - val_acc: 0.9881
Epoch 10/12
60000/60000 [==============================] - 14s - loss: 0.0531 - acc: 0.9845 - val_loss: 0.0320 - val_acc: 0.9889
Epoch 11/12
60000/60000 [==============================] - 13s - loss: 0.0498 - acc: 0.9855 - val_loss: 0.0323 - val_acc: 0.9890
Epoch 12/12
60000/60000 [==============================] - 14s - loss: 0.0479 - acc: 0.9852 - val_loss: 0.0329 - val_acc: 0.9892

# 输出训练好的模型在测试集上的表现
print('Test score:', score[0])
print('Test accuracy:', score[1])

Test score: 0.032927570413
Test accuracy: 0.9892