Crate egobox_moe

source
Expand description

This library implements Mixture of Experts method using GP models.

MoE method aims at increasing the accuracy of a function approximation by replacing a single global model by a weighted sum of local gp regression models (experts). It is based on a partition of the problem domain into several subdomains via clustering algorithms followed by a local expert training on each subdomain.

The recombination between the GP models can be either:

  • hard: one GP model is being responsible to provide the predicted value at the given point. GP selection is done by taking the largest probability of the given point being part of the cluster corresponding to the expert GP. In hard mode, transition between models leads to discontinuity.
  • smooth: all GPs models are taken and their predicted values at a given point are weighted regarding their responsability (probability of the given point being part of the cluster corresponding to the expert GP). In this case the MoE model is continuous. The smoothness is automatically adjusted using a factor, the heaviside factor, which can also be set manually.

§Implementation

  • Clusters are defined by clustering the training data with linfa-clustering gaussian mixture model.
  • This library is a port of the SMT MoE method using egobox GP models as experts.
  • It leverages on the egobox GP PLS reduction feature to handle high dimensional problems.
  • MoE trained model can be save to disk and reloaded. See

§Features

§serializable

The serializable feature enables serialization based on serde crate.

§persistent

The persistent feature enables save()/load() methods for a MoE model to/from a json file using the serde and serde_json crates.

§Example

use ndarray::{Array2, Array1, Zip, Axis};
use egobox_moe::{GpMixture, Recombination};
use ndarray_rand::{RandomExt, rand::SeedableRng, rand_distr::Uniform};
use rand_xoshiro::Xoshiro256Plus;
use linfa::{traits::Fit, ParamGuard, Dataset};

// One-dimensional test function with 3 modes
fn f3modes(x: &Array2<f64>) -> Array2<f64> {
    let mut y = Array2::zeros(x.dim());
    Zip::from(&mut y).and(x).for_each(|yi, &xi| {
        if xi < 0.4 {
            *yi = xi * xi;
        } else if (0.4..0.8).contains(&xi) {
            *yi = 3. * xi + 1.;
        } else {
            *yi = f64::sin(10. * xi);
        }
    });
    y
}

// Training data
let mut rng = Xoshiro256Plus::from_entropy();
let xt = Array2::random_using((50, 1), Uniform::new(0., 1.), &mut rng);
let yt = f3modes(&xt);
let ds = Dataset::new(xt, yt);

// Predictions
let observations = Array1::linspace(0., 1., 100).insert_axis(Axis(1));
let predictions = GpMixture::params()
                    .n_clusters(3)
                    .recombination(Recombination::Hard)
                    .fit(&ds)
                    .expect("MoE model training")
                    .predict(&observations)
                    .expect("MoE predictions");

§Reference

Bettebghor, Dimitri, et al. Surrogate modeling approximation using a mixture of experts based on EM joint estimation Structural and multidisciplinary optimization 43.2 (2011): 243-259.

Structs§

Enums§

Traits§

Functions§

Type Aliases§

  • A result type for Moe algorithm