A from-scratch Pytorch-like implementation of neural networks using only functional programming

Done using numpy for matrix operations, mostly as a review exercise for my own sake.

Easy to extend and modify to support any new layers or initialization schemes.

See example_usage.ipynb for example usage