# 导入模块 import numpy as np import tensorflow as tf from tensorflow import keras import matplotlib.pyplot as plt # 读取MNIST数据 mnist = keras.datasets.mnist (x_train, y_train),(x_test, y_test) = mnist.load_data() # 重构数据至4维（样本，像素X，像素Y，通道） x_train=x_train.reshape(x_train.shape+(1,)) x_test=x_test.reshape(x_test.shape+(1,)) x_train, x_test = x_train/255.0, x_test/255.0 # 数据标签 label_train = keras.utils.to_categorical(y_train, 10) label_test = keras.utils.to_categorical(y_test, 10) # LeNet-5构筑 model = keras.Sequential([ keras.layers.Conv2D(6, kernel_size=(3, 3), strides=(1, 1), activation='tanh', padding='valid'), keras.layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2), padding='valid'), keras.layers.Conv2D(16, kernel_size=(5, 5), strides=(1, 1), activation='tanh', padding='valid'), keras.layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2), padding='valid'), keras.layers.Flatten(), keras.layers.Dense(120, activation='tanh'), keras.layers.Dense(84, activation='tanh'), keras.layers.Dense(10, activation='softmax'), ]) # 使用SGD编译模型 model.compile(loss=keras.losses.categorical_crossentropy, optimizer='SGD') # 学习30个纪元（可依据CPU计算力调整），使用20%数据交叉验证 records = model.fit(x_train, label_train, epochs=20, validation_split=0.2) # 预测 y_pred = np.argmax(model.predict(x_test), axis=1) print("prediction accuracy: {}".format(sum(y_pred==y_test)/len(y_test))) # 绘制结果 plt.plot(records.history['loss'],label='training set loss') plt.plot(records.history['val_loss'],label='validation set loss') plt.ylabel('categorical cross-entropy'); plt.xlabel('epoch') plt.legend()
