Struct ActiveSampler 

pub struct ActiveSampler { /* private fields */ }

Active sampling strategy for minimising interpolation error.

Example

use scirs2_interpolate::active_learning::{
    ActiveSampler, ActiveSamplerConfig, ActiveAcquisitionFunction,
};

let config = ActiveSamplerConfig {
    acquisition: ActiveAcquisitionFunction::MaximumVariance,
    n_candidates: 20,
    domain: vec![[0.0, 1.0], [0.0, 1.0]],
    seed: 7,
};
let mut sampler = ActiveSampler::new(config);

// Seed with one observation
sampler.observe(vec![0.5, 0.5], 1.0);

let next = sampler.suggest_next();
assert_eq!(next.len(), 2);

Implementations

impl ActiveSampler

pub fn new(config: ActiveSamplerConfig) -> Self

Create a new sampler. The number of dimensions is inferred from config.domain.len().

pub fn suggest_next(&self) -> Vec<f64>

Select the next query point by evaluating the acquisition function at config.n_candidates randomly sampled candidate points.

If no candidates score above zero, returns the first candidate (random).

pub fn observe(&mut self, point: Vec<f64>, value: f64)

Register a new observation.

pub fn acquisition_value(&self, point: &[f64]) -> f64

Compute the acquisition value for a single candidate point.

Returns 0.0 when there are no observations.

pub fn loo_error(&self) -> f64

Leave-one-out cross-validation error estimate.

For each observed point, fits a simple GP on the remaining n-1 points and measures the squared prediction error. Returns the RMS LOO error.

Returns 0.0 when fewer than 2 observations are available.

pub fn n_observed(&self) -> usize

Number of observations recorded so far.

pub fn observed_points(&self) -> &[Vec<f64>]

Slice of all observed points.

pub fn n_dims(&self) -> usize

Dimensionality of the domain.

Trait Implementations

impl Debug for ActiveSampler

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.