A Neural Network library fully written in Numpy? Why would you ever want to do this in Numpy? Well, to better understand the different types of layers and operations going on in a Neural Network. For me, implementing networks based on libraries like Tensorflow and Theano was easy, especially when relying on their automatic differentiation features. However, implementing everything yourself from scratch was a challenge.
I'll try and add more kinds of layers in the future. However, progress will be quite slow due to the lack of time to work on this.