General response surface. More...

#include <ResponseSurfaceVector.hpp>

Inheritance diagram for sgpp::optimization::ResponseSurfaceVector:

Public Member Functions
sgpp::base::DataVector	averageL2Error (std::shared_ptr< sgpp::base::VectorFunction > objectiveFunc, sgpp::base::DataVector &componentwiseErrors, size_t numMCPoints=1000)
	Calculates the average l2 error between interpolant and objective function.

sgpp::base::DataVector	averageNRMSE (std::shared_ptr< sgpp::base::VectorFunction > objectiveFunc, sgpp::base::DataMatrix &componentwiseErrorData, size_t numMCPoints=1000)
	Calculates the normalized root mean square error between interpolant and objective function at random points.

virtual sgpp::base::DataVector	eval (sgpp::base::DataVector v)
	evaluates the response surface

virtual sgpp::base::DataVector	evalJacobian (sgpp::base::DataVector v, sgpp::base::DataMatrix &jacobian)=0
	evaluates the response surface and its gradients

size_t	getNumDim ()
	Calculates the normalized root mean square error between interpolant and objective function from a precalculated set of test points and according values (this set is usually generated by ResponseSurfaceVector::precalculateErrorTestData)

size_t	getNumRes ()

	ResponseSurfaceVector (size_t numDim, size_t numRes)
	Constructor.

virtual	~ResponseSurfaceVector ()
	Destructor.

Protected Member Functions
double	domainVolume ()
	calculates the volume of the tensor product domain given by lb and ub

void	transformPoint (sgpp::base::DataVector &v, sgpp::base::DataVector lBounds, sgpp::base::DataVector uBounds, sgpp::base::DataVector newlBounds, sgpp::base::DataVector newuBounds)
	transforms a point in hyper-rectangle [lBounds,rBounds] to the hyper-rectangle [newlBounds,newuBounds]

Protected Attributes
std::shared_ptr< sgpp::base::VectorFunctionGradient >	interpolantGradients

std::shared_ptr< sgpp::base::VectorFunction >	interpolants

sgpp::base::DataVector	lb

size_t	numDim

size_t	numRes

sgpp::base::DataVector	ub

Detailed Description

General response surface.

Represents an approximation of some function with vector valued output. Usually the approximation is created via interpolation. (But could also be regression for example)

Constructor & Destructor Documentation

◆ ResponseSurfaceVector()

sgpp::optimization::ResponseSurfaceVector::ResponseSurfaceVector	(	size_t	numDim,
		size_t	numRes
	)

inlineexplicit

Constructor.

◆ ~ResponseSurfaceVector()

virtual sgpp::optimization::ResponseSurfaceVector::~ResponseSurfaceVector ( )

inlinevirtual

Destructor.

Member Function Documentation

◆ averageL2Error()

sgpp::base::DataVector sgpp::optimization::ResponseSurfaceVector::averageL2Error	(	std::shared_ptr< sgpp::base::VectorFunction >	objectiveFunc,
		sgpp::base::DataVector &	componentwiseErrors,
		size_t	numMCPoints = `1000`
	)

Calculates the average l2 error between interpolant and objective function.

Parameters

objectiveFunc	the objectiveFunction
componentwiseErrors	reference to return the l2 error in each individual of the numRes components
numMCPoints	number of Monte Carlo Points

Returns: vector with [average l2 error, min l2 error, max l2 error]

References sgpp::base::DataVector::abs(), sgpp::base::DataVector::add(), sgpp::base::DataVector::componentwise_mult(), eval(), sgpp::base::RandomNumberGenerator::getInstance(), sgpp::base::RandomNumberGenerator::getUniformRN(), lb, sgpp::base::DataVector::max(), sgpp::base::DataVector::min(), numDim, numRes, sgpp::base::DataVector::resizeZero(), sgpp::base::DataVector::sub(), and ub.

◆ averageNRMSE()

sgpp::base::DataVector sgpp::optimization::ResponseSurfaceVector::averageNRMSE	(	std::shared_ptr< sgpp::base::VectorFunction >	objectiveFunc,
		sgpp::base::DataMatrix &	componentwiseErrorData,
		size_t	numMCPoints = `1000`
	)

Calculates the normalized root mean square error between interpolant and objective function at random points.

Parameters

objectiveFunc	the objectiveFunction
componentwiseErrorData	reference to return the NRMSE, l2, min and max for each individual of the numRes components
numMCPoints	number of Monte Carlo Points

Returns: vector [average NRMSE, average l2 error, min NRMSE, max NRMSE]

References sgpp::base::DataVector::abs(), sgpp::base::DataVector::add(), averageNRMSE(), sgpp::base::DataVector::componentwise_mult(), eval(), sgpp::base::RandomNumberGenerator::getInstance(), sgpp::base::RandomNumberGenerator::getUniformRN(), lb, sgpp::base::DataVector::max(), sgpp::base::DataVector::min(), numDim, numRes, sgpp::base::DataMatrix::resize(), sgpp::base::DataMatrix::setRow(), sgpp::base::DataVector::sub(), and ub.

Referenced by averageNRMSE().

◆ domainVolume()

double sgpp::optimization::ResponseSurfaceVector::domainVolume ( )

protected

calculates the volume of the tensor product domain given by lb and ub

Returns: volume

References lb, numDim, and ub.

Referenced by sgpp::optimization::SplineResponseSurfaceVector::getIntegrals().

◆ eval()

sgpp::base::DataVector sgpp::optimization::ResponseSurfaceVector::eval ( sgpp::base::DataVector v )

virtual

evaluates the response surface

Parameters

v	point in which the response surface shall be evaulated

Returns: evaluation

Reimplemented in sgpp::optimization::SplineResponseSurfaceVector.

References interpolants, and numRes.

Referenced by averageL2Error(), averageNRMSE(), and python.uq.analysis.asgc.ASGCAnalysis.ASGCAnalysis::estimateDensity().

◆ evalJacobian()

virtual sgpp::base::DataVector sgpp::optimization::ResponseSurfaceVector::evalJacobian	(	sgpp::base::DataVector	v,
		sgpp::base::DataMatrix &	jacobian
	)

pure virtual

evaluates the response surface and its gradients

Parameters

v	point in which the response surface shall be evaluated
jacobian	reference to return the jacobian evaluated in v. Each row is one gradient

Returns: evaluation

Implemented in sgpp::optimization::SplineResponseSurfaceVector.

◆ getNumDim()

size_t sgpp::optimization::ResponseSurfaceVector::getNumDim ( )

inline

Calculates the normalized root mean square error between interpolant and objective function from a precalculated set of test points and according values (this set is usually generated by ResponseSurfaceVector::precalculateErrorTestData)

Parameters

fileName	path to the stored data set
numMCPoints	the first numMCPoints of the data set are used
numDim	number of dimensions

Returns: vector [NRMSE, l2 error, min, max] Evaluates the objective function in random points and stores these points with the according evaluation value The stored set of test data can then later be used to calculate and compare the approximation error

Parameters

objectiveFunc	the objective function
numMCPoints	number of Monte carlo Points
path	path specifying where to save the generated data

References numDim.

◆ getNumRes()

size_t sgpp::optimization::ResponseSurfaceVector::getNumRes ( )

inline

References numRes.

◆ transformPoint()

void sgpp::optimization::ResponseSurfaceVector::transformPoint	(	sgpp::base::DataVector &	v,
		sgpp::base::DataVector	lBounds,
		sgpp::base::DataVector	uBounds,
		sgpp::base::DataVector	newlBounds,
		sgpp::base::DataVector	newuBounds
	)