Knowing, at least intuitively, the components of a deep learning model and how they interact is a must before going any further into practical details. The practical details might change with respect to which deep learning framework and API to be used; this chapter uses Keras. It will give access to Google's TensorFlow and some other frameworks.
Building cutting-edge models using Keras is easy. The workflow usually goes as follows: once you are done with data preprocessing, you design the network's architecture and choose a learning strategy, a cost function, and measures to track. The next steps are training and testing.
Four words could be used to summarize (usual) workflow: design, compile, train, and test. Although the ANN models are usually very complex, to build a very good one from scratch using Keras is very simple; it skips a bunch of human steps without losing flexibility. First things first, the environment must be properly setup.