Neural Network from scratch in Python

Following https://nnfs.io/ to create a deep learning classifier. The goal is to correctly classify the three colors into their own groups. This first implementation does not have backpropagation, and randomly adjusts weights in a direction that causes loss to decrease. 92% accuracy after about 400 iterations (epochs). The network uses one hidden layer with 3 neurons.

Using spiral dataset as it is more challenging. Basic backprop is causing pretty bad loss and accuracy, might be at a local minimum.