使用Keras/Tensorflow.keras子类API构建复杂模型

本文主要介绍了几种使用keras.Model API构建深度网络模型的基本方法。

1. 简单顺序模型

      模型由输入层、隐藏层、和输出层顺序连接构成（如图1所示）

图1. 简单顺序模型示意图

(1) 定义模型结构

def build_model(X_train):

      input_layer = keras.layers.Input(shape=X_train.shape[1:])

      hidden1_layer = keras.layers.Dense(30, activation=‘ReLU’)(input_layer)

      hidden2_layer = keras.layers.Dense(30, activation=‘linear’)(hidden1_layer)

      output_layer  = keras.layers.Dense(1)(hidden2_layer)

      model = keras.Model(inputs=input_layer, outputs=output_layer)

      return model

(2) 建模训练

model = build_model(X_train)

model.compile(loss=‘mse’, optimizer=keras.optimizers.SGD(lr=1e-3))

model.fit(X_train, Y_train)

eval = model.evaluate(X_val, Y_val)

Y_pred = model.pred(X_test)

---

 2. 短路径顺序模型

      模型由输入层、隐藏层、输出层和短路径连接构成（如图2所示）

 图2. 短路径顺序模型示意图

(1) 定义模型结构（注意红色字体部分）

def build_model(X_train):

      input_layer = keras.layers.Input(shape=X_train.shape[1:])

      hidden1_layer = keras.layers.Dense(30, activation=‘ReLU’)(input_layer)

      hidden2_layer = keras.layers.Dense(30, activation=‘linear’)(hidden1_layer)

      concat_layer = keras.layers.Concatenate()([input_layer, hidden2_layer])

      output_layer  = keras.layers.Dense(1)(concat_layer)

      model = keras.Model(inputs=input_layer, outputs=output_layer)

      return model

(2) 建模训练

model = build_model(X_train)

model.compile(loss=‘mse’, optimizer=keras.optimizers.SGD(lr=1e-3))

model.fit(X_train, Y_train)

eval = model.evaluate(X_val, Y_val)

Y_pred = model.pred(X_test)

---

 3. 多输入顺序模型

      模型由两个及以上输入层、隐藏层、输出层和短路径连接构成（如图3所示）

  图3. 多输入顺序模型示意图

(1) 定义模型结构（注意红色字体部分）

def build_model(X_train1, X_train2):

      input1_layer = keras.layers.Input(shape=X_train1.shape[1:])

      input2_layer = keras.layers.Input(shape=X_train2.shape[1:])

      hidden1_layer = keras.layers.Dense(30, activation=‘ReLU’)(input1_layer)

      hidden2_layer = keras.layers.Dense(30,            activation=‘linear’)(hidden1_layer)

      concat_layer = keras.layers.Concatenate()([input2_layer, hidden2_layer])

      output_layer  = keras.layers.Dense(1)(concat_layer)

      model = keras.Model(inputs=[input1_layer, input2_layer], outputs=output_layer)

      return model

(2) 建模训练（注意红色字体部分）

model = build_model(X_train1, X_train2)

model.compile(loss=‘mse’, optimizer=keras.optimizers.SGD(lr=1e-3))

history = model.fit((X_train1, X_train2), Y_train)

eval = model.evaluate((X_val1, X_val2), Y_val)

Y_pred = model.pred((X_test1, X_test2))

---

 4. 多输入多输出模型

      模型由两个及以上输入层、隐藏层、两个及以上输出层连接构成（如图4所示）

  图4. 多输入多输出模型示意图

(1) 定义模型结构（注意红色字体部分）

def build_model(X_train1, X_train2):

      input1_layer = keras.layers.Input(shape=X_train1.shape[1:])

      input2_layer = keras.layers.Input(shape=X_train2.shape[1:])

      hidden1_layer = keras.layers.Dense(30, activation=‘ReLU’)(input1_layer)

      hidden2_layer = keras.layers.Dense(30,            activation=‘linear’)(hidden1_layer)

      concat_layer = keras.layers.Concatenate()([input2_layer, hidden2_layer])

      output1_layer  = keras.layers.Dense(1, name=‘main_output’)(concat_layer)

      ouput2_layer = keras.layers.Dense(1, name=‘aux_output)(hidden2_layer)

      model = keras.Model(inputs=[input1_layer, input2_layer], outputs=[output1_layer, output2_layer])

      return model

(2) 建模训练（注意红色字体部分）

model = build_model(X_train1, X_train2)

model.compile(loss=[‘mse’, ‘mse’], loss_weight=[0.9, 0.1], optimizer=keras.optimizers.SGD(lr=1e-3))

history = model.fit((X_train1, X_train2), (Y_train1, Y_train2))

eval = model.evaluate((X_val1, X_val2), (Y_val1, Y_val2))

Y_pred1, Y_pred2 = model.pred((X_test1, X_test2))

---

参考文献：

[1] A GéRon. Hands-on machine learning with Scikit-Learn and TensorFlow : concepts, tools, and techniques to build intelligent systems.  2017.