The project is a rust implementation of guessing hand-written digits using a deep learning algorithm, with the help of the MNIST database. It has been heavily inspired by 3Blue1Brown's video introduction to deep learning.
- Install rust
- Clone this repository
git clone https://github.com/jookzie/digit-guesser-rs.git --depth=1
cd digit-guesser-rs
- Run the project
cargo run --release
- Parse database files
- Train model
- Test model
- Command line interface
- Benchmarks
- Tests
- Serialize/deseralize model
- Multi-threading
So far, the trained models manage to guess the digits 36% of the time. Most likely the backprop is subpar and needs optimizations.
P.S. I have not had any AI courses, this is purely an attempt to learn rust and deep learning.