class gpr::ARDGridSearch¶

Overview¶

Grid search over ARD lengthscales + noise. More…

#include <ARDGridSearch.h>
 
class ARDGridSearch {
public:
    // construction
 
    ARDGridSearch(int n_ls_per_dim = 10, int n_noise = 20, int n_refine = 1, double log_ls_min = -3.0, double log_ls_max = 1.0, double log_noise_min = -10.0, double log_noise_max = 2.0);
 
    // methods
 
    ARDSearchResult searchGP(const Eigen::VectorXd& y, KernelMatrixBuilder kernel_builder, int n_pairtypes) const;
};

Detailed Documentation¶

Grid search over ARD lengthscales + noise.

Methods¶

ARDSearchResult searchGP(const Eigen::VectorXd& y, KernelMatrixBuilder kernel_builder, int n_pairtypes) const

Search for optimal (lengthscales, noise) using GP likelihood.

Parameters:

y	Training targets
kernel_builder	Function that builds K_XX from lengthscales and noise
n_pairtypes	Number of ARD dimensions

Returns:

ARDSearchResult