https://github.com/adelsz/oxigrad

Toy autodifferentiation engine in Rust
https://github.com/adelsz/oxigrad

Last synced: 2 days ago
JSON representation

Toy autodifferentiation engine in Rust

• Host: GitHub
• URL: https://github.com/adelsz/oxigrad
• Owner: adelsz
• License: mit
• Created: 2024-02-01T22:22:50.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2024-02-01T22:25:55.000Z (10 months ago)
• Last Synced: 2024-10-29T12:16:14.589Z (14 days ago)
• Language: Rust
• Size: 7.42 MB
• Stars: 1
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Oxigrad - toy automatic differentiation engine in Rust

This is a self-contained toy automatic differentiation engine written in Rust.  

Built for learning purposes and partially inspired by [micrograd](https://github.com/karpathy/micrograd).  

API is focused on ease of use and educational clarity, performance was not a priority.

## Usage

```rust

    #[test]

    fn backprop_test() {

        let a = Value::new(2.0);

        let b = Value::new(1.0);

    

        // z = (a * b + a) * (a * b + b) = a^2 * b^2 + a^2 * b + a * b^2 + a * b

        let z = (&a * &b + &a) * (&a * &b + &b);

    

        backprop(&z);

    

        // dz/da = 2ab^2 + 2ab + b^2 + b = 2*2*1 + 2*2 + 1 + 1 = 10

        assert_eq!(a.grad().get(), 10.0);

    }

```

## Basic neural network training

There is a basic neural network training example in `src/neuron.rs` and plots the training progress in a GIF file.