Hellow World

需要安装 tensorflow tensorflow-datasets matplotlib

TensorFlow Datasets 文档

import keras  
import tensorflow_datasets as tfds  
import matplotlib.pyplot as plt  
  
  
def mnist_load_1():  
    # 基于 tensorflow_datasets 加载  
    train_ds = tfds.load('mnist', split='train', shuffle_files=True)  
    test_ds = tfds.load('mnist', split='train', shuffle_files=True)  
    # 打印数据集描述信息  
    print(train_ds.element_spec)  
    # 分离 image 和 label    train_image = train_ds.map(lambda example: example['image'])  
    train_label = test_ds.map(lambda example: example['label'])  
    test_image = test_ds.map(lambda example: example['image'])  
    test_label = test_ds.map(lambda example: example['label'])  
    return (train_image, train_label), (test_image, test_label)  
  
  
# 查看数据集内容  
def show_example_1(images_ds, labels_ds):  
    for image in images_ds.take(1):  
        print('Image shape:', image.shape)  
        plt.imshow(image, cmap='gray')  
        plt.show()  
    for label in labels_ds.take(1):  
        print('Label:', label)  
  
  
def mnist_load_2():  
    # 基于 keras 加载  
    return keras.api.datasets.mnist.load_data()  
  
  
# 查看数据集内容  
def show_example_2(images, labels):  
    print('Image shape:', images.shape)  
    print('Label length::', len(labels))  
    plt.imshow(images[4], cmap='gray')  
    plt.show()  
  
  
if __name__ == '__main__':  
    # 1. 加载 mnist 数据集  
  
    # 加载方法1： 通过 tensorflow_datasets    
    # (x_train, y_train), (x_test, y_test) = mnist_load_1()    
    # show_example_1(x_train, y_train)  
    # 加载方法2： 通过 keras    
    (x_train, y_train), (x_test, y_test) = mnist_load_2()  
    # show_example_2(x_train, y_train)  
  
    # 2. 神经网络  
    # 设置模型层  
    model = keras.Sequential([  
        keras.layers.Dense(512, activation='relu'),  
        keras.layers.Dense(10, activation='softmax'),  
    ])  
    # 设置编译步骤： 优化器、损失函数、指标  
    model.compile(optimizer='rmsprop',  
                  loss='sparse_categorical_crossentropy',  
                  metrics=['accuracy'])  
  
    # 3. 准备数据  
    train_images = x_train.reshape((60000, 28 * 28)).astype('float32') / 255  
    test_images = x_test.reshape((10000, 28 * 28)).astype('float32') / 255  
  
    # 4. 训练数据  
    model.fit(train_images, y_train, epochs=5, batch_size=128)  
  
    #5. 利用模型进行预测  
    # 取10条测试数据进行预测  
    predict = model.predict(test_images[0:10])  
    # 打印第一条预测结果  
    print_prefix = "== > prediction "  
    print(print_prefix, ": ")  
    print(print_prefix, "info: ", predict[0])  
    print(print_prefix, "number: ",predict[0].argmax())  
    print(print_prefix, "actual number:",y_test[0])  
    plt.imshow(x_test[0], cmap='gray')  
    plt.show()  
  
  
    #6. 使用测试数据评估新模型  
    test_loss, test_acc = model.evaluate(test_images, y_test)  
    print(f'Test accuracy:{test_acc * 100:.2f}%')