This is the training system I promised about. It takes a while to complete. The more expected answer data there is, the longer it takes to train. I did implement a feature that stops it from having a relapse. The variables 1 & 2 represent the loss. 1 is the template for the next generation.
