spt-neural-net

Native Rust implementation of the Neural Network defined in Tariq Rashid's excellent book "Make Your Own Neural Network".

Pre-requisites

The application takes a single (optional) boolean argument that results in each of the images used for training and testing the NeuralNetwork being displayed - this is useful when debugging the application. The display of the images uses the matplotlib crate, which simply wraps the Python library of the same name. In order therefore, for the image display to work, the following Python library pre-requisites must be installed.

$ pip install --user matplotlib
$ pip install --user PyQt5

Building and running the application

The standard cargo commands can be used for building, testing and running the application.

$ cargo clippy
$ cargo test
$ cargo build --release
$ cargo run --release

Example output when running the application.

$  cargo run --release
    Finished `release` profile [optimized] target(s) in 0.13s
     Running `target/release/spt-neural-net`
2024-12-15T17:13:51.391128 Training the network...
2024-12-15T17:14:13.786799 Network trained with 10000 records...
2024-12-15T17:14:35.126296 Network trained with 20000 records...
2024-12-15T17:14:55.012176 Network trained with 30000 records...
2024-12-15T17:15:15.502282 Network trained with 40000 records...
2024-12-15T17:15:35.603023 Network trained with 50000 records...
2024-12-15T17:15:55.829038 Network trained with 60000 records...
2024-12-15T17:15:55.829095 Testing the network...
2024-12-15T17:15:57.872578 Network performance: 0.97

NOTE In order to get results in the region of 97% as shown in the example above, it will be necessary to train the neural network with a larger data set than mnist_train_100.csv; the full dataset is available here.