Struct AutoRegressiveGp 

pub struct AutoRegressiveGp { /* private fields */ }

AR(1) coregionalization multi-output GP.

Maintains one independent GP per fidelity level; level l is fitted to the residual y_l - rho_l * mu_{l-1}(x_l).

Implementations

impl AutoRegressiveGp

pub fn new(levels: &[FidelityLevel]) -> Self

Construct an unfitted AR-GP for the given fidelity levels.

pub fn fit( &mut self, x_list: &[Array2<f64>], y_list: &[Array1<f64>], ) -> OptimizeResult<()>

Fit the AR-GP to multi-fidelity data.

x_list[l] and y_list[l] are the observed inputs/outputs at level l. Levels must be ordered from lowest (0) to highest fidelity.

pub fn update( &mut self, fidelity: usize, x: Array1<f64>, y: f64, ) -> OptimizeResult<()>

Add a single new observation at a given fidelity level and refit.

pub fn predict( &self, x: &Array2<f64>, fidelity: usize, ) -> OptimizeResult<(Array1<f64>, Array1<f64>)>

Predict mean and variance at query points for a given fidelity level.

The prediction is obtained by propagating the AR(1) recursion upward from level 0 to the requested level.