21  Running the GAN for the first time

Now it is time to run the GAN and try to really simulate our synthetic data.

In the previous chapter “Fixing what found in GitHub” I already discussed what I changed before the first run.
In addition I needed the changed the output of the Generator to be 512 units, in line with what is the input of the Decoder.

I started running the model and these were my observation:

1. At the very beginning there a moderate decrease in Generator while the Discriminator stayed pretty much the same
2. The Generator loss continued to decrease and the Discriminator one started to increase
3. This trend was not linear in both networks
4. After some time the Discriminator loss started to decrease and the Generator one to increase
5. The Discriminator loss converged to a very small number while the Generator loss converged to a value higher than the starting point
6. I then tried to generate some synthetic data using the trained Generator but I discovered how it only generated observation with the event DEATH which is the most common one
7. The GAN was mode collapsing (see point 6)

The next section will explain mode collapsing.