#computation #graph #gradient #value #backpropagate

bin+lib regrad

A simple library to backpropagate gradients through a computation graph

2 unstable releases

0.2.0 Oct 5, 2024
0.1.0 Oct 5, 2024

#213 in Machine learning

Download history 264/week @ 2024-10-02 44/week @ 2024-10-09 3/week @ 2024-10-16

311 downloads per month

MIT license

11KB
279 lines

regrad

backtrack in rust

Usage

Add to Cargo.toml

cargo add regrad

Example

use regrad::Value;

fn main() {
    let v1 = Value::from(1.2);
    let v2 = Value::from(3.4);
    let v3 = &(&v1 * &v1) * &v2;

    dbg!(v3.data());
    assert_eq!(v3.data(), 4.896);

    v3.backward();
    dbg!(v1.gradient());
    dbg!(v2.gradient());
    dbg!(v3.gradient());

    assert_eq!(v1.gradient(), 8.16);
    assert_eq!(v2.gradient(), 1.44);
    assert_eq!(v3.gradient(), 1.0);
}

TODO

  • Add more operations
  • Add Tensor support
  • Add more tests
  • Add more examples
  • Add more documentation
  • Add GPU support
  • Crates.io publish

No runtime deps