原文链接：http://www.one2know.cn/keras_mnist/
import numpy
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense # 稠密层
from keras.layers import Dropout # Dropout将在训练过程中每次更新参数时按一定概率（rate）随机断开输入神经元，Dropout层用于防止过拟合。
from keras.layers import Flatten # Flatten层用来将输入“压平”，即把多维的输入一维化，常用在从卷积层到全连接层的过渡。Flatten不影响batch的大小。
from keras.layers.convolutional import Conv2D # 二维卷积层，即对图像的空域卷积。
from keras.layers.convolutional import MaxPooling2D # 空间池化（也叫亚采样或下采样）降低了每个特征映射的维度，但是保留了最重要的信息
from keras.utils import np_utils
from keras import backend as K
K.set_image_dim_ordering('th') # 设置图像的维度顺序（‘tf’或‘th’）
# 当前的维度顺序如果为'th'，则输入图片数据时的顺序为：channels,rows,cols，否则:rows,cols,channels
seed = 7
numpy.random.seed(seed)
#将数据reshape，CNN的输入是4维的张量（可看做多维的向量），第一维是样本规模，第二维是像素通道，第三维和第四维是长度和宽度。并将数值归一化和类别标签向量化。
# load data
(X_train, y_train), (X_test, y_test) = mnist.load_data()
# reshape to be [samples][pixels][width][height]
X_train = X_train.reshape(X_train.shape[0], 1, 28, 28).astype('float32')
X_test = X_test.reshape(X_test.shape[0], 1, 28, 28).astype('float32')
X_train = X_train / 255
X_test = X_test / 255
# 将标签转化成one-hot编码
y_train = np_utils.to_categorical(y_train)
y_test = np_utils.to_categorical(y_test)
num_classes = y_test.shape[1]
## 接下来构造CNN
# 第一层是卷积层。该层有32个feature map,或者叫滤波器，作为模型的输入层，接受[pixels][width][height]大小的输入数据。feature_map的大小是5*5，其输出接一个‘relu’激活函数。
# 下一层是pooling层，使用了MaxPooling，大小为2*2。
# 下一层是Dropout层，该层的作用相当于对参数进行正则化来防止模型过拟合。
# Flatten层用来将输入“压平”，即把多维的输入一维化，常用在从卷积层到全连接层的过渡。Flatten不影响batch的大小。
# 接下来是全连接层，有128个神经元，激活函数采用‘relu’。
# 最后一层是输出层，有10个神经元，每个神经元对应一个类别，输出值表示样本属于该类别的概率大小。
def baseline_model():
# create model
model = Sequential()
model.add(Conv2D(32, (5, 5), input_shape=(1, 28, 28), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dense(num_classes, activation='softmax'))
# Compile model
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
return model
# 建立模型
model = baseline_model()
# 训练模型
model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=10, batch_size=200, verbose=2)
# 模型概要打印
model.summary()
# 模型评估
scores = model.evaluate(X_test, y_test, verbose=0)
print("Baseline Error: %.2f%%" % (100-scores[1]*100))
输出：
Train on 60000 samples, validate on 10000 samples
Epoch 1/10
- 68s - loss: 0.2247 - acc: 0.9358 - val_loss: 0.0776 - val_acc: 0.9754
Epoch 2/10
- 66s - loss: 0.0709 - acc: 0.9787 - val_loss: 0.0444 - val_acc: 0.9853
Epoch 3/10
- 67s - loss: 0.0511 - acc: 0.9843 - val_loss: 0.0435 - val_acc: 0.9855
Epoch 4/10
- 66s - loss: 0.0392 - acc: 0.9880 - val_loss: 0.0391 - val_acc: 0.9873
Epoch 5/10
- 66s - loss: 0.0325 - acc: 0.9898 - val_loss: 0.0341 - val_acc: 0.9893
Epoch 6/10
- 65s - loss: 0.0266 - acc: 0.9918 - val_loss: 0.0318 - val_acc: 0.9890
Epoch 7/10
- 65s - loss: 0.0221 - acc: 0.9929 - val_loss: 0.0348 - val_acc: 0.9886
Epoch 8/10
- 65s - loss: 0.0191 - acc: 0.9941 - val_loss: 0.0308 - val_acc: 0.9890
Epoch 9/10
- 66s - loss: 0.0153 - acc: 0.9951 - val_loss: 0.0325 - val_acc: 0.9897
Epoch 10/10
- 65s - loss: 0.0143 - acc: 0.9957 - val_loss: 0.0301 - val_acc: 0.9903
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_1 (Conv2D) (None, 32, 24, 24) 832
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 32, 12, 12) 0
_________________________________________________________________
dropout_1 (Dropout) (None, 32, 12, 12) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 4608) 0
_________________________________________________________________
dense_1 (Dense) (None, 128) 589952
_________________________________________________________________
dense_2 (Dense) (None, 10) 1290
=================================================================
Total params: 592,074
Trainable params: 592,074
Non-trainable params: 0
_________________________________________________________________
Baseline Error: 0.97%