Modular neural network from scratch in numpy. (+ activation, loss functions etc.)
This implements a dense neural network, as well as activation, loss functions etc., from scratch in numpy.
As such, only numpy (and, if you want learning curves, also matplotlib) is needed.
It is designed to have a general, hierarchical and modular architecture.
The repo’s core is more concisely detailed in this jupyter notebook.
Feel free to reach out for suggestions :)