Keras functional API introduction

参考文献：Géron, Aurélien. Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems. O'Reilly Media, 2019.

Keras' Sequential sequence model can quickly build a simple neural network, and Keras also provides a functional API (Functional API) for customizing various types of grid structures.

Concatenate

When building the Wide & Deep neural grid, layer concatenation is required. The fusion layer in the figure adds the Input layer and the last layer of the hidden layer together.

input_ = keras.layers.Input(shape=X_train.shape[1:])
hidden1 = keras.layers.Dense(30, activation="relu")(input_)
hidden2 = keras.layers.Dense(30, activation="relu")(hidden1)
concat = keras.layers.concatenate([input_, hidden2])
output = keras.layers.Dense(1)(concat)
model = keras.models.Model(inputs=[input_], outputs=[output])

Multi-inputs

The input features can be divided into multiple groups (there can be overlapping parts), and let them pass through different paths in the neural network.

input_A = keras.layers.Input(shape=[5], name="wide_input")
input_B = keras.layers.Input(shape=[6], name="deep_input")
hidden1 = keras.layers.Dense(30, activation="relu")(input_B)
hidden2 = keras.layers.Dense(30, activation="relu")(hidden1)
concat = keras.layers.concatenate([input_A, hidden2])
output = keras.layers.Dense(1, name="output")(concat)
model = keras.models.Model(inputs=[input_A, input_B], outputs=[output])

Multi-outputs

input_A = keras.layers.Input(shape=[5], name="wide_input")
input_B = keras.layers.Input(shape=[6], name="deep_input")
hidden1 = keras.layers.Dense(30, activation="relu")(input_B)
hidden2 = keras.layers.Dense(30, activation="relu")(hidden1)
concat = keras.layers.concatenate([input_A, hidden2])
output = keras.layers.Dense(1, name="main_output")(concat)
aux_output = keras.layers.Dense(1, name="aux_output")(hidden2)
model = keras.models.Model(inputs=[input_A, input_B],
                           outputs=[output, aux_output])

Each output can set the loss function separately

model.compile(loss=[“mse”,”mse”], loss_weights=[0.9, 0.1], optimizer=“sgd”)

If not set, Keras uses the same loss function by default. During training, Keras will calculate two loss functions separately and add them together to get the final loss value.