Applied Deep Learning with Python

Model Components

As we have seen in Chapter 4, Introduction to Neural Networks and Deep Learning, LSTM networks also have input, hidden, and output layers. Each hidden layer has an activation function that evaluates that layer's associated weights and biases. As expected, the network moves data sequentially from one layer to another and evaluates the results by the output at every iteration (that is, an epoch).

Keras provides intuitive classes that represent each one of those components:

Component	Keras Class
High-level abstraction of a complete sequential neural network.	keras.models.Sequential()
Dense, fully-connected layer.	keras.layers.core.Dense()
Activation function.	keras.layers.core.Activation()
LSTM recurrent neural network. ...