Implementation of a simple Neural Network to solve MNIST in Rust

A feedforward network trained with stochastic gradient descent and backpropagation, written from scratch with only ndarray for the linear algebra. Inspired from Michael Nielsen's tutorial repo.

Building

cargo build --release

Preparing the data

The MNIST files in data/ are gzipped, they have to be decompressed first:

gunzip -k data/*.gz

Running

Program arguments are:

./target/release/neural_networks [data_path] [epoch] [mini_batch_size] [learning_rate]

Example:

./target/release/neural_networks data 30 10 3.0