Noeda / rcmaes

Rust bindings to libcmaes

Rust libcmaes bindings

This is a simple library that lets you optimize some model. If you can implement a trait that turns a struct into a Vec<f64> and back (i.e. there is an isomorphism between them), then you can pass it to CMA-ES library.

The C++ CMA-ES project is located here: https://github.com/beniz/libcmaes

I'm using this mostly for some personal projects so it's not all that comprehensive. The bindings implement just enough that you can adjust most important parameters and run heavy optimization concurrently and with reasonable efficiency.

Example

extern crate rcmaes;

use rcmaes::{optimize, CMAESParameters, Vectorizable};

// This example "trains" a very simple model to go to point (2, 8)
#[derive(Clone)]
struct TwoPoly {
    x: f64,
    y: f64,
}

// Vectorizable is essential to specify how to turn something into a Vec<f64>
// and back.
impl Vectorizable for TwoPoly {
    type Context = ();

    fn to_vec(&self) -> (Vec<f64>, Self::Context) {
        (vec![self.x, self.y], ())
    }

    fn from_vec(vec: &[f64], _: &Self::Context) -> Self {
        TwoPoly {
            x: vec[0],
            y: vec[1],
        }
    }
}

fn train_model()
{
    let optimized = optimize(
        &TwoPoly { x: 5.0, y: 6.0 },
        &CMAESParameters::default(),
        |twopoly| (twopoly.x - 2.0).abs() + (twopoly.y - 8.0).abs(),
    ).unwrap();

    let model = optimized.0;
    assert!((model.x - 2.0).abs() < 0.00001);
    assert!((model.y - 8.0).abs() < 0.00001);
}