1.导入依赖包

将tensorflow和numpy导入

from tensorflow import keras

from tensorflow.keras.layers import Flatten, Dense

import numpy as np

2.加载数据集

使用tensorflow内置的mnist数据集，返回训练集图片、训练集标签、测试集图片、测试集标签

(train_images, train_labels), (test_images, test_labels) = keras.datasets.mnist.load_data()

3.对图片进行归一化

图片每个像素的数值都是在[0, 255]之间，所以归一化要除以255，数据要是浮点数，所以要添加一个小数点

train_images, test_images = train_images / 255.0, test_images / 255.0

4.定义模型

搭建一个顺序模型，第一层先将数据展平，原始图片是28x28的灰度图，所以输入尺寸是（28，28），第二层节点数可以自己选择一个合适值，这里用128个节点，激活函数用relu，第三层有多少个种类就写多少，[0, 9]一共有10个数字,所以必须写10，激活函数用softmax

model = keras.Sequential([

    Flatten(input_shape=(28, 28)),

    Dense(128, activation='relu'),

    Dense(10, activation='softmax')

])

5.指定优化器、损失函数、评价指标

优化器使用adam，损失函数使用交叉熵损失函数，评价指标用准确率

model.compile(optimizer='adam',

              loss='sparse_categorical_crossentropy',

              metrics=['acc'])

6.训练模型

将训练集输入模型进行训练，一共训练10次

model.fit(train_images, train_labels, epochs=10)

Epoch 1/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.2575 - acc: 0.9261
Epoch 2/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.1135 - acc: 0.9661
Epoch 3/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0776 - acc: 0.9764
Epoch 4/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0589 - acc: 0.9818
Epoch 5/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0452 - acc: 0.9861
Epoch 6/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0371 - acc: 0.9890
Epoch 7/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0279 - acc: 0.9912
Epoch 8/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0226 - acc: 0.9932
Epoch 9/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0203 - acc: 0.9935
Epoch 10/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0153 - acc: 0.9954

<tensorflow.python.keras.callbacks.History at 0x24ecdc1c460>

7.用测试集验证模型效果

用测试集去验证训练好的模型，日志等级设置为2

test_loss, test_acc = model.evaluate(test_images, test_labels, verbose=2)

print('Test acc:', test_acc)

313/313 - 0s - loss: 0.0845 - acc: 0.9791
Test acc: 0.9790999889373779

8.将图片输入模型，返回预测结果

将测试集中的第一张图片输入模型，看是哪个数字的概率最大，并输出真实值

predictions = model.predict(test_images)

print('预测值:', np.argmax(predictions[0]))

print('真实值:', test_labels[0])

预测值: 7
真实值: 7