from keras.datasets import mnist
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()
#train_images 和 train_labels 是训练集
train_images.shape#第一个数字表示图片张数，后面表示图片尺寸，和之前我在opencv上遇到的有所不同
#opencv上是前面表示图片尺寸，后面表示图片的通道数量

输出：

(60000, 28, 28)

len(train_labels)

输出：
60000

from keras import models
from keras import layers

下面开始构造神经网络：

network = models.Sequential()
network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,)))#果然shape是28*28！！！
network.add(layers.Dense(10, activation='softmax'))

预编译：

network.compile(optimizer='rmsprop',
loss='categorical_crossentropy',
metrics=['accuracy'])
train_images = train_images.reshape((60000, 28 * 28))
train_images = train_images.astype('float32') / 255
test_images = test_images.reshape((10000, 28 * 28))
test_images = test_images.astype('float32') / 255

开始训练模型：

network.fit(train_images, train_labels, epochs=5, batch_size=128)

输出：

Epoch 1/5
60000/60000 [==============================] - 7s 111us/step - loss: 0.2523 - acc: 0.9274
Epoch 2/5
60000/60000 [==============================] - 7s 111us/step - loss: 0.1029 - acc: 0.9689 5s - loss: 0.1212
Epoch 3/5
60000/60000 [==============================] - 7s 116us/step - loss: 0.0677 - acc: 0.9795
Epoch 4/5
60000/60000 [==============================] - 8s 130us/step - loss: 0.0504 - acc: 0.9848
Epoch 5/5
60000/60000 [==============================] - 7s 119us/step - loss: 0.0374 - acc: 0.9886 2s - loss: 0.0370 -
Out[12]:
<keras.callbacks.History at 0x1c6e30c1828>

因此可得识别准确度为98%

进行测试集的验证：

 test_loss, test_acc = network.evaluate(test_images, test_labels)

输出准确度：

 print('识别准确度为:', test_acc)

识别准确度为：
0.9807