Did you forget yet? Let's pull up the diagram on GANs so that we can discuss the different parts of the structure we will be producing classes for in this chapter:

This basic structure is what we will be converting to code. The key to this particular recipe is understanding what pieces we need to convert and what pieces are simply going to be wrapped up into a single class. For example, the latent space will be sampled from a Gaussian distribution that's available in the NumPy library. Since we are just sampling from this Gaussian distribution, it is necessary to know the size of the latent space at each step, as you will see as we make progress on these GAN tools.

There are three core classes that we need to develop to build the structure of a GAN for adversarial training. First, we need a discriminator. This is the piece of the architecture that is focused on deciding whether an input belongs to a particular class or not. In this chapter, we will simply limit the discriminator model to that of a binary classification (true or false).