This is a neural network made in collaboration with Niklas Kniesche (github.com/bigol-tauri) for our senior project at the Wheeler School. It was made following along with Daniel Shiffman's Neural Network Coding Challenge
The main method can be made to either train from the MNIST data set or read from a file of weights. It can create the file itself but will need to be manually set up.
The best percentage we were able to achieve was 97.5% accuracy across the entire data set