class gpr::NoiseGridSearch¶

Overview¶

Exponential grid search over noise hyperparameter. More…

#include <NoiseGridSearch.h>
 
class NoiseGridSearch {
public:
    // construction
 
    NoiseGridSearch(int n_grid = 80, int n_refine = 2, double log_noise_min = -10.0, double log_noise_max = 2.0);
 
    // methods
 
    NoiseSearchResult searchGP(const FactorizedLikelihood& likelihood) const;
    NoiseSearchResult searchTP(const FactorizedLikelihood& likelihood, double a, double b) const;
};

Detailed Documentation¶

Exponential grid search over noise hyperparameter.

Construction¶

NoiseGridSearch(int n_grid = 80, int n_refine = 2, double log_noise_min = -10.0, double log_noise_max = 2.0)

Configure the grid search.

Parameters:

n_grid	Number of grid points per refinement level
n_refine	Number of refinement iterations
log_noise_min	Minimum log10(noise) to search
log_noise_max	Maximum log10(noise) to search

Methods¶

NoiseSearchResult searchGP(const FactorizedLikelihood& likelihood) const

Search for optimal noise using GP factorized likelihood.

NoiseSearchResult searchTP(const FactorizedLikelihood& likelihood, double a, double b) const

Search for optimal noise using TP factorized likelihood.

Parameters:

a	inverse-gamma shape parameter
b	inverse-gamma scale parameter