keras实现多层感知机进行手写字识别

微信公众号:数据挖掘与分析学习

1.数据准备

(x_train,y_train),(x_test,y_test)=mnist.load_data() #加载数据   #数据预处理 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') #归一化 x_train/=255 x_test/=255   #将类标签转换为one-hot向量 y_train=keras.utils.to_categorical(y_train,num_classes=num_classes) y_test=keras.utils.to_categorical(y_test,num_classes=num_classes)

2.参数设置

batch_size=128 #批处理大小 num_classes=10 #类标签数量 epochs=20

3.模型构建

model=Sequential() #定义一个序贯模型 #input_shape:模型的第一层需要指明输入数据的大小,后面各层自动推断 model.add(Dense(512,activation='relu',input_shape=(784,)))#全连接层，然后激活层 model.add(Dropout(0.2))#Dropout model.add(Dense(512,activation='relu')) #全连接层 model.add(Dropout(0.2)) model.add(Dense(num_classes,activation='softmax')) #全连接，然后使用Softmax激活求概率

4.模型编译、训练和测试

model.compile(loss='categorical_crossentropy',optimizer=RMSprop(),metrics=['accuracy']) history=model.fit(x_train,y_train,batch_size=batch_size,epochs=epochs,verbose=1,validation_data=(x_test,y_test)) score=model.evaluate(x_test,y_test,verbose=0) print('Test loss:',score[0]) #损失 print('Test accuracy:',score[1]) #精度

5.打印网络结构信息

model.summary()

输出如下:

Layer (type)                 Output Shape              Param #   ================================================================= dense_4 (Dense)              (None, 512)               401920    _________________________________________________________________ dropout_3 (Dropout)          (None, 512)               0         _________________________________________________________________ dense_5 (Dense)              (None, 512)               262656    _________________________________________________________________ dropout_4 (Dropout)          (None, 512)               0         _________________________________________________________________ dense_6 (Dense)              (None, 10)                5130      ================================================================= Total params: 669,706 Trainable params: 669,706 Non-trainable params: 0