深度神经网络——神经网络案例

手写数字mnist

60000个训练样本和10000个测试样本，图像是固定大小（28x28像素），其值为0-255

import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense,Dropout,Activation,BatchNormalization
from tensorflow.keras import utils
#正则化
from tensorflow.keras import regularizers
#数据集
from tensorflow.keras.datasets import mnist

数据加载

(x_train,y_train),(x_test,y_test)=mnist.load_data()
plt.figure()
plt.imshow(x_train[1],cmap='gray')

数据处理

x_train=x_train.reshape(60000,784)
x_test=x_test.reshape(10000,784)
x_train=x_train.astype('float32')
x_test=x_test.astype('float32')
x_train=x_train/255
x_test=x_test/255
y_train=utils.to_categorical(y_train,10)
y_test=utils.to_categorical(y_test,10)

模型构建

model=Sequential()
#全连接层：2个隐层，1个输出层
#第一个：512个神经元，先激活后bn，随机失活
model.add(Dense(512,activation='relu',input_shape=(784,)))
model.add(BatchNormalization())
model.add(Dropout(0.2))
#第二个：512个神经元，先bn后激活，随机失活
model.add(Dense(512,kernel_regularizer=regularizers.l2(0.01)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dropout(0.2))

model.add(Dense(10,activation='softmax'))
model.summary()

注意：dropout放最后，bn和激活顺序没有要求

模型编译

model.compile(optimizer=tf.keras.optimizers.Adam(),loss=tf.keras.losses.CategoricalCrossentropy,metrics=tf.keras.metrics.Accuracy())

模型训练

history=model.fit(x_train,y_train,epochs=4,batch_size=128,validation_data=(x_test,y_test),verbose=1)

plt.figure()
plt.plot(history.history['loss'],label='train')
plt.plot(history.history['val_loss'],label='val')
plt.legend()
plt.grid()
plt.show()

plt.figure()
plt.plot(history.history['accuracy'],label='train')
plt.plot(history.history['val_accuracy'],label='val')
plt.legend()
plt.grid()
plt.show()

tensorboard=tf.keras.callbacks.tensorboard(log_dir='./graph')
history=model.fit(x_train,y_train,epochs=4,batch_size=128,validation_data=(x_test,y_test),verbose=1,callbacks=[tensorboard])

#指定存在文件的目录，打开下面命令
tensorboard --logdir='./'

在浏览器中打开指定网址，可查看损失函数和准确率的变化，图结构等

模型测试

score=model.evaluate(x_test,y_test)

模型保存

model.save('model.h5')
loadmodel=tf.keras.models.load_model('model.h5')