eliavaux / machine-learning Goto Github PK

A very simple machine learning project using backpropagation built from the ground up.

The neural network is trained on the MNIST database and can recognize handwritten digits fairly well after about two minutes of training. Trained neural networks can be saved to a JSON file inside the saves/ directory with Network::save() and loaded in with Network::load().

Inside the data/ folder lie the training data of the MNIST database with their headers removed for simplicity. The parser::create_img() function can convert the training data into an PNG of size 28x28. The first 100 images are already inside the images/ folder.

The Matrix struct is written as the amount of columns and a one-dimensional vector of the data with column-major order.

This is not meant to be practical code for a neural network, but more of a learning project on the basic concepts of machine learning and my implementation of them into Rust

3Blue1Brown for his informative and straightforward Video Series on the basics of neural networks and backpropagation.

MathleteDev's Matrices and Neural Networks Tutorial which has been a great guide throughout the project.