GP_Objective Class Reference

ROL objective function for the Gaussian Process (GP) surrogate. More...

Inherits Objective< double >.

Public Member Functions
	GP_Objective (GaussianProcess &gp_model)
	Constructor for GP_Objective. More...
double	value (const ROL::Vector< double > &p, double &tol) override
	Get the value of the objective function at a point. More...
void	gradient (ROL::Vector< double > &g, const ROL::Vector< double > &p, double &tol) override
	Get the gradient of the objective function at a point. More...

Private Member Functions
bool	pdiff (const VectorXd &pnew)
	Compute the l2 norm of the difference between new and old parameter vectors. More...

Private Attributes
GaussianProcess &	gp
	Reference to the GaussianProcess surrogate.
int	nopt
	Number of optimization variables.
double	Jold
	Previously computed value of the objective function.
VectorXd	grad_old
	Previously computed gradient of the objective function.
VectorXd	pold
	Previous value of the parameter vector.

Detailed Description

ROL objective function for the Gaussian Process (GP) surrogate.

Constructor & Destructor Documentation

GP_Objective()

GP_Objective

(

GaussianProcess &

gp_model

)

Constructor for GP_Objective.

Parameters

[in]

gp_model

Reference to the GaussianProcess surrogate.

References GaussianProcess::get_num_opt_variables(), GP_Objective::gp, GP_Objective::grad_old, GP_Objective::Jold, GP_Objective::nopt, and GP_Objective::pold.

Member Function Documentation

value()

double value ( const ROL::Vector< double > &  p, double &  tol  )

override

Get the value of the objective function at a point.

Parameters

[in]	p	ROL vector of parameters.
[in]	tol	Tolerance for inexact evaluation (not used here).

References GP_Objective::gp, GaussianProcess::negative_marginal_log_likelihood(), GP_Objective::nopt, GP_Objective::pdiff(), GaussianProcess::set_opt_params(), and dakota::silence_unused_args().

gradient()

void gradient ( ROL::Vector< double > &  g, const ROL::Vector< double > &  p, double &  tol  )

override

Get the gradient of the objective function at a point.

Parameters

[out]	g	Gradient of the objective function.
[in]	p	ROL vector of parameters.
[in]	tol	Tolerance for inexact evaluation (not used here).

References GP_Objective::gp, GaussianProcess::negative_marginal_log_likelihood(), GP_Objective::pdiff(), GaussianProcess::set_opt_params(), and dakota::silence_unused_args().

pdiff()

bool pdiff ( const VectorXd &  pnew )

private

Compute the l2 norm of the difference between new and old parameter vectors.

Parameters

[in]

pnew

New value of the parameter vector.

References dakota::near_zero, and GP_Objective::pold.

Referenced by GP_Objective::gradient(), and GP_Objective::value().

---

The documentation for this class was generated from the following files:

• SurrogatesGPObjective.hpp
• SurrogatesGPObjective.cpp