Road 2 AI-NN简易神经网络的建立流程

神经网络建立简易流程

import tensorflow as tf

#创建连接层
model = tf.keras.Sequential()
#kernel_initializer，对hidden层初始化操作
initializer = tf.initializers.RandomNormal(stddev=0.01)

model.add(tf.keras.layers.Dense(1024,activation = 'relu',kernel_initializer = initializer ,input_dim = len(feature)))
model.add(tf.keras.layers.Dense(512,activation = 'relu.',kernel_initializer = initializer ))
model.add(tf.keras.layers.Dense(256,activation = 'relu.',kernel_initializer = initializer ))
model.add(tf.keras.layers.Dense(128,activation = 'relu.',kernel_initializer = initializer ))
model.add(tf.keras.layers.Dense(64,activation = 'relu.',kernel_initializer = initializer ))
model.add(tf.keras.layers.Dense(1,activation = 'sigmoid'))#如果是分类任务，就需要调用sigmoid激活函数，如果是回归，则无需写

#设置优化器
adam = keras.optimizers.Adam(learning_rate = 0.015)
model.compile(loss = 'binary_crossentropy',optimizer = adam,metric = ['accuracy'])#loss是根据你的任务而定，此处是分类任务

#定义回调函数，
callbacks = [
    #设置早停指标
    tf.keras.callbacks.EarlyStopping(monitor='loss',patience=20,verbose=10),
    #保存最优模型
    tf.keras.callbacks.ModelCheckpoint('./heckpoint/multi_layer_best_model.h5',monitor='val_loss',save_best_only=True,verbose=0),
    #实现动态学习率,factor学习率变化百分比，patience表示如果val_loss在5轮没有提升则改变学习率，min_lr设定最小学习率为0.001
    tf.keras.callbacks.ReduceLROnPlateau(
    monitor='val_loss', factor=0.2, patience=5, verbose=0,min_lr=0.001
)
    ]

#模型训练
model.fit(X_train,y_train,batch_size=batch,callbacks=[callbacks],validation_data=(X_val,y_val),epochs = epochs)

#模型评估
model_evaluate = model.evaluate(x,y) #查看准确率

#模型预测
y_pred = model.predict(x_test)