Made the backwards for dense layers and ReLU, but I think I soon need to refactor some code to make it all steamlined.
Fixed the loss functions for categories and is ready for accuracy calculation on page 129