神经网络手写数字识别

#!/usr/bin/env python3
from keras import layers
from keras import models
from keras.datasets import mnist
from keras.utils import to_categorical
​
def hello_world():
    # 下载手写数字数据集, https://s3.amazonaws.com/img-datasets/mnist.npz
    (train_images, train_labels), (test_images, test_labels) = mnist.load_data()
​
    print('训练集：', train_images.shape)   # (60000, 28, 28)
    print('训练集：', len(train_labels))    # 60000
    print('训练集：', train_labels)         # [5 0 4 ... 5 6 8]
​
    print()
    print('测试集：', test_images.shape)    # (10000, 28, 28)
    print('测试集：', len(test_labels))     # 10000
    print('测试集：', test_labels)          # [7 2 1 ... 4 5 6]
​
    # 模型，可以理解为就是前文说的黑盒子
    network = models.Sequential()
    # 神经网络的核心就是 layer，是真正用来加工处理数据的，Dense 是定义网络层的基础方法
    # 512 定义神经元个数，activation 定义激活函数，input_shape 定义输入尺寸
    network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,)))
    network.add(layers.Dense(10, activation='softmax'))
​
    # 配置训练模型，optimizer 优化器名，loss 损失函数，metrics 模型评估标准
    network.compile(optimizer='rmsprop',
                    loss='categorical_crossentropy',
                    metrics=['accuracy'])
​
    # 不改变数组内容，改变数组格式
    train_images = train_images.reshape((60000, 28 * 28))
    # 数据类型转换
    train_images = train_images.astype('float32') / 255
​
    test_images = test_images.reshape((10000, 28 * 28))
    test_images = test_images.astype('float32') / 255
​
    # 转换成二进制
    train_labels = to_categorical(train_labels)
    test_labels = to_categorical(test_labels)
​
    # 在数据集上进行迭代
    network.fit(train_images, train_labels, epochs=5, batch_size=128, callbacks=[TensorBoard(log_dir='./log')])
    # 测试模式下损失值和指标值
    test_loss, test_acc = network.evaluate(test_images, test_labels)
​
    print('test_loss', test_loss, ', test_acc:', test_acc)
​
if __name__ == "__main__":
    hello_world()