Using the Keras Functional API

The Keras Sequential API is great for developing deep learning models in most situations. However, this API has some limitations, such as a linear topology, that could be overcome with the Functional API. Note that many high-performing networks are based on a non-linear topology such as Inception, ResNet, etc.

The Functional API allows defining complex models with a non-linear topology, multiple inputs, multiple outputs, residual connections with non-sequential flows, and shared and reusable layers.

The deep learning model is usually a directed acyclic graph (DAG). The Functional API is a way to build a graph of layers and create more flexible models than the tf.keras.Sequential API.

Getting ready

This recipe will cover the main ways of creating a Functional model, using callable models, manipulating complex graph topologies, sharing layers, and finally introducing the concept of the layer "node" with the Keras Sequential...