Struct AutoKernelGp 

pub struct AutoKernelGp { /* private fields */ }

Gaussian Process interpolator with automatic kernel structure discovery.

Example

use scirs2_interpolate::auto_kernel_gp::{AutoKernelGp, AutoKernelGpConfig};

let x: Vec<f64> = (0..20).map(|i| i as f64 * std::f64::consts::PI / 10.0).collect();
let y: Vec<f64> = x.iter().map(|&xi| xi.sin()).collect();

let config = AutoKernelGpConfig {
    max_depth: 1,
    cv_folds: 3,
    ..Default::default()
};
let mut gp = AutoKernelGp::new(config);
gp.fit(&x, &y).expect("fit ok");

let x_new = vec![0.1, 0.5, 1.0];
let preds = gp.predict(&x_new).expect("predict ok");
assert_eq!(preds.len(), 3);

Implementations

impl AutoKernelGp

pub fn new(config: AutoKernelGpConfig) -> Self

Create a new (unfitted) AutoKernelGp.

pub fn fit(&mut self, x: &[f64], y: &[f64]) -> InterpolateResult<()>

Search the kernel grammar and fit the GP on x, y.

x must be strictly sorted (ascending) for well-defined kernel matrices, though this is not enforced. Duplicate x values will cause a near-singular kernel matrix which is handled by jitter.

pub fn predict(&self, x_new: &[f64]) -> InterpolateResult<Vec<f64>>

Predict at new input locations x_new.

pub fn selected_kernel_description(&self) -> String

Return a human-readable description of the selected kernel structure.

pub fn best_cv_score(&self) -> f64

Return the best cross-validation score (lower is better).

pub fn kernel_search_results(&self) -> &[(String, f64)]

Return the full ranked list of (kernel_description, cv_mse_score) pairs.

Sorted by CV-MSE ascending (best first).

pub fn kernel(&self) -> &KernelExpr

Return the selected kernel expression.