#include <ExpensiveSamplingProblem.h>

Inheritance diagram for muq::SamplingAlgorithms::ExpensiveSamplingProblem:

Detailed Description

Definition at line 16 of file ExpensiveSamplingProblem.h.

Public Member Functions
	ExpensiveSamplingProblem (std::shared_ptr< muq::Modeling::ModPiece > const &target, std::shared_ptr< muq::Modeling::ModPiece > const &qoi, Eigen::VectorXd const &centroid, boost::property_tree::ptree pt)

	ExpensiveSamplingProblem (std::shared_ptr< muq::Modeling::ModPiece > const &target, std::shared_ptr< muq::Modeling::ModPiece > const &qoi, boost::property_tree::ptree pt)

	ExpensiveSamplingProblem (std::shared_ptr< muq::Modeling::ModPiece > const &target, boost::property_tree::ptree pt)

	ExpensiveSamplingProblem (std::shared_ptr< muq::Modeling::ModPiece > const &target, Eigen::VectorXd const &centroid, boost::property_tree::ptree pt)

	ExpensiveSamplingProblem (std::shared_ptr< muq::Modeling::ModPiece > const &target, boost::property_tree::ptree pt, std::shared_ptr< parcer::Communicator > comm)

	ExpensiveSamplingProblem (std::shared_ptr< muq::Modeling::ModPiece > const &target, Eigen::VectorXd const &centroid, boost::property_tree::ptree pt, std::shared_ptr< parcer::Communicator > comm)

virtual	~ExpensiveSamplingProblem ()=default

virtual double	LogDensity (std::shared_ptr< SamplingState > const &state) override

unsigned int	CacheSize () const

virtual void	AddOptions (boost::property_tree::ptree &pt) const override

virtual std::shared_ptr< SamplingState >	QOI () override

Public Member Functions inherited from muq::SamplingAlgorithms::SamplingProblem
	SamplingProblem (std::shared_ptr< muq::Modeling::ModPiece > const &targetIn)

	SamplingProblem (std::shared_ptr< muq::Modeling::ModPiece > const &targetIn, std::shared_ptr< muq::Modeling::ModPiece > const &qoiIn)

virtual	~SamplingProblem ()=default

virtual std::shared_ptr< AbstractSamplingProblem >	Clone () const override

virtual Eigen::VectorXd	GradLogDensity (std::shared_ptr< SamplingState > const &state, unsigned const blockWrt) override

std::shared_ptr< muq::Modeling::ModPiece >	GetDistribution ()

Public Member Functions inherited from muq::SamplingAlgorithms::AbstractSamplingProblem
	AbstractSamplingProblem (Eigen::VectorXi const &blockSizesIn, Eigen::VectorXi const &blockSizesQOIIn)

	AbstractSamplingProblem (Eigen::VectorXi const &blockSizesIn)

virtual	~AbstractSamplingProblem ()=default

Public Attributes
unsigned int	cumbeta = 0
	Cumulative beta refinements. More...

unsigned int	cumgamma = 0
	Cumulative gamma refinements. More...

unsigned int	cumkappa = 0
	Cumulative kappa refinements. More...

double	lastLyapunov = 0.0
	The Lyapunov function at the most recently accepted point. More...

double	lastThreshold = 0.0
	The error threshold at the most recently accepted point. More...

Public Attributes inherited from muq::SamplingAlgorithms::AbstractSamplingProblem
const int	numBlocks

const Eigen::VectorXi	blockSizes

const int	numBlocksQOI

const Eigen::VectorXi	blockSizesQOI

Constructor & Destructor Documentation

◆ ExpensiveSamplingProblem() [1/6]

muq::SamplingAlgorithms::ExpensiveSamplingProblem::ExpensiveSamplingProblem	(	std::shared_ptr< muq::Modeling::ModPiece > const &	target,
		std::shared_ptr< muq::Modeling::ModPiece > const &	qoi,
		Eigen::VectorXd const &	centroid,
		boost::property_tree::ptree	pt
	)

Parameters

[in]	target	The target distribution
[in]	qoi	The quantity of interest
[in]	centroid	The centroid of the distribution
[in]	pt	Options for the sampling problem

◆ ExpensiveSamplingProblem() [2/6]

muq::SamplingAlgorithms::ExpensiveSamplingProblem::ExpensiveSamplingProblem	(	std::shared_ptr< muq::Modeling::ModPiece > const &	target,
		std::shared_ptr< muq::Modeling::ModPiece > const &	qoi,
		boost::property_tree::ptree	pt
	)

Parameters

[in]	target	The target distribution
[in]	qoi	The quantity of interest
[in]	pt	Options for the sampling problem

◆ ExpensiveSamplingProblem() [3/6]

muq::SamplingAlgorithms::ExpensiveSamplingProblem::ExpensiveSamplingProblem	(	std::shared_ptr< muq::Modeling::ModPiece > const &	target,
		boost::property_tree::ptree	pt
	)

Parameters

[in]	target	The target distribution
[in]	pt	Options for the sampling problem

◆ ExpensiveSamplingProblem() [4/6]

muq::SamplingAlgorithms::ExpensiveSamplingProblem::ExpensiveSamplingProblem	(	std::shared_ptr< muq::Modeling::ModPiece > const &	target,
		Eigen::VectorXd const &	centroid,
		boost::property_tree::ptree	pt
	)

Parameters

[in]	target	The target distribution
[in]	centroid	The centroid of the distribution
[in]	pt	Options for the sampling problem

◆ ExpensiveSamplingProblem() [5/6]

ExpensiveSamplingProblem::ExpensiveSamplingProblem	(	std::shared_ptr< muq::Modeling::ModPiece > const &	target,
		boost::property_tree::ptree	pt,
		std::shared_ptr< parcer::Communicator >	comm
	)

Parameters

[in]	target	The target distribution
[in]	pt	Options for the sampling problem
[in]	comm	Parallel communicator for regression

Definition at line 55 of file ExpensiveSamplingProblem.cpp.

References reg, SetUp(), and muq::SamplingAlgorithms::SamplingProblem::target.

◆ ExpensiveSamplingProblem() [6/6]

ExpensiveSamplingProblem::ExpensiveSamplingProblem	(	std::shared_ptr< muq::Modeling::ModPiece > const &	target,
		Eigen::VectorXd const &	centroid,
		boost::property_tree::ptree	pt,
		std::shared_ptr< parcer::Communicator >	comm
	)

Parameters

[in]	target	The target distribution
[in]	centroid	The centroid of the distribution
[in]	pt	Options for the sampling problem
[in]	comm	Parallel communicator for regression

Definition at line 63 of file ExpensiveSamplingProblem.cpp.

References reg, SetUp(), and muq::SamplingAlgorithms::SamplingProblem::target.

◆ ~ExpensiveSamplingProblem()

virtual muq::SamplingAlgorithms::ExpensiveSamplingProblem::~ExpensiveSamplingProblem ( )

virtualdefault

Member Function Documentation

◆ AddOptions()

void ExpensiveSamplingProblem::AddOptions ( boost::property_tree::ptree & pt ) const

overridevirtual

Sometimes, there will be problem-specific options that need to be passed to the SamplingAlgorithm. This function adds any of those options to the given property_tree.

Reimplemented from muq::SamplingAlgorithms::AbstractSamplingProblem.

Definition at line 241 of file ExpensiveSamplingProblem.cpp.

◆ CacheSize()

unsigned int ExpensiveSamplingProblem::CacheSize ( ) const

Definition at line 239 of file ExpensiveSamplingProblem.cpp.

References reg.

◆ CheckNeighbors()

void ExpensiveSamplingProblem::CheckNeighbors	(	std::shared_ptr< SamplingState > const &	state,
		std::vector< Eigen::VectorXd > &	neighbors,
		std::vector< Eigen::VectorXd > &	results
	)		const

private

Check to make sure we have enough model evaluations.

Check to see if the state has the nearest neighbors already stored. If yes, return them, if not, find and return them.

Parameters

[in]	state	Find the nearest neighbors closest to this point
[out]	neighbors	The nearest neighbors
[out]	results	The log-target at the nearest neighbors

Definition at line 142 of file ExpensiveSamplingProblem.cpp.

References reg.

Referenced by RefineSurrogate().

◆ CheckNumNeighbors()

void ExpensiveSamplingProblem::CheckNumNeighbors ( std::shared_ptr< SamplingState > const & state )

private

Check to make sure we have enough model evaluations.

Parameters

[in] state If we do not have enough points, sample from a standard Gaussian centered at this point

Definition at line 132 of file ExpensiveSamplingProblem.cpp.

References cumkappa, gamma, reg, and regQoI.

Referenced by RefineSurrogate().

◆ ErrorThreshold()

double ExpensiveSamplingProblem::ErrorThreshold ( Eigen::VectorXd const & x ) const

private

Compute the error threshold.

Parameters

[in] x The state of the MCMC chain

Returns: The error threshild

Definition at line 150 of file ExpensiveSamplingProblem.cpp.

References gamma, level, and LogLyapunovFunction().

Referenced by RefineSurrogate().

◆ LogDensity()

double ExpensiveSamplingProblem::LogDensity ( std::shared_ptr< SamplingState > const & state )

overridevirtual

Reimplemented from muq::SamplingAlgorithms::SamplingProblem.

Definition at line 91 of file ExpensiveSamplingProblem.cpp.

References beta, cumbeta, cumgamma, eta, lastLyapunov, muq::SamplingAlgorithms::SamplingProblem::lastState, lastThreshold, LogLyapunovFunction(), RefineSurrogate(), reg, and step.

◆ LogLyapunovFunction()

double ExpensiveSamplingProblem::LogLyapunovFunction ( Eigen::VectorXd const & x ) const

private

Estimate the Lyapunov function.

Defaults to \(V(x) = \exp{(\|x-\bar{x}\|_2)}\).

Parameters

[in] x the input point \(x\)

Returns: \(V(x)\)

Definition at line 128 of file ExpensiveSamplingProblem.cpp.

References centroid, and lyapunovPara.

Referenced by ErrorThreshold(), and LogDensity().

◆ QOI()

std::shared_ptr< SamplingState > ExpensiveSamplingProblem::QOI ( )

overridevirtual

Reimplemented from muq::SamplingAlgorithms::SamplingProblem.

Definition at line 245 of file ExpensiveSamplingProblem.cpp.

References muq::SamplingAlgorithms::SamplingProblem::lastState, muq::SamplingAlgorithms::SamplingProblem::qoi, and regQoI.

◆ RefineSurrogate() [1/3]

void ExpensiveSamplingProblem::RefineSurrogate	(	Eigen::VectorXd const &	point,
		unsigned int const	index,
		std::vector< Eigen::VectorXd > &	neighbors,
		std::vector< Eigen::VectorXd > &	results
	)

private

Replace the point that is farthest from the new point

Parameters

[in]	state	The point where we are evalauting the log target
[in]	index	The index of the point that we are replacing
[out]	neighbors	The nearest neighbors
[out]	results	The log-target at the nearest neighbors

Definition at line 225 of file ExpensiveSamplingProblem.cpp.

References reg, and regQoI.

◆ RefineSurrogate() [2/3]

double ExpensiveSamplingProblem::RefineSurrogate	(	std::shared_ptr< SamplingState > const &	state,
		double const	radius,
		std::vector< Eigen::VectorXd > &	neighbors,
		std::vector< Eigen::VectorXd > &	results
	)

private

Parameters

[in]	state	The state where we are evalauting the log target
[in]	radius	The radus of the nearest neighbor ball
[out]	neighbors	The nearest neighbors
[out]	results	The log-target at the nearest neighbors

Returns: The poisedness constant \(\Lambda\)

Definition at line 154 of file ExpensiveSamplingProblem.cpp.

References RefineSurrogate(), and reg.

◆ RefineSurrogate() [3/3]

double ExpensiveSamplingProblem::RefineSurrogate	(	unsigned int const	step,
		std::shared_ptr< SamplingState > const &	state,
		std::vector< Eigen::VectorXd > &	neighbors,
		std::vector< Eigen::VectorXd > &	results
	)

private

Parameters

[in]	step	The current MCMC step
[in]	state	The state where we are evalauting the log target
[out]	neighbors	The nearest neighbors
[out]	results	The log-target at the nearest neighbors

Returns: The error threshold

Definition at line 188 of file ExpensiveSamplingProblem.cpp.

References beta, CheckNeighbors(), CheckNumNeighbors(), cumbeta, cumgamma, ErrorThreshold(), gamma, level, reg, step, and tau0.

Referenced by LogDensity(), and RefineSurrogate().

◆ SetUp()

void ExpensiveSamplingProblem::SetUp ( boost::property_tree::ptree & pt )

private

Set up the sampling problem.

Parameters

[in] pt Options for the sampling problem

Definition at line 72 of file ExpensiveSamplingProblem.cpp.

References beta, eta, gamma, lyapunovPara, reg, muq::SamplingAlgorithms::SamplingProblem::target, and tau0.

Referenced by ExpensiveSamplingProblem().

Member Data Documentation

◆ beta

std::pair<double, double> muq::SamplingAlgorithms::ExpensiveSamplingProblem::beta

private

Parameters for random refinement.

Refine with probability \(\beta = \beta_0 t^{-\beta_1}\). \(\beta_0\) is "BetaScale" and it defaults to \(0\). \(\beta_1\) is "BetaExponent" and it defaults to \(RAND_MAX\).

Definition at line 165 of file ExpensiveSamplingProblem.h.

Referenced by LogDensity(), RefineSurrogate(), and SetUp().

◆ centroid

const Eigen::VectorXd muq::SamplingAlgorithms::ExpensiveSamplingProblem::centroid

private

The centroid of the distribution (input parameter)

Definition at line 183 of file ExpensiveSamplingProblem.h.

Referenced by LogLyapunovFunction().

◆ cumbeta

unsigned int muq::SamplingAlgorithms::ExpensiveSamplingProblem::cumbeta = 0

Cumulative beta refinements.

Definition at line 195 of file ExpensiveSamplingProblem.h.

Referenced by LogDensity(), and RefineSurrogate().

◆ cumgamma

unsigned int muq::SamplingAlgorithms::ExpensiveSamplingProblem::cumgamma = 0

Cumulative gamma refinements.

Definition at line 198 of file ExpensiveSamplingProblem.h.

Referenced by LogDensity(), and RefineSurrogate().

◆ cumkappa

unsigned int muq::SamplingAlgorithms::ExpensiveSamplingProblem::cumkappa = 0

Cumulative kappa refinements.

Definition at line 201 of file ExpensiveSamplingProblem.h.

Referenced by CheckNumNeighbors().

◆ eta

double muq::SamplingAlgorithms::ExpensiveSamplingProblem::eta

private

Tail correction parameter.

Definition at line 186 of file ExpensiveSamplingProblem.h.

Referenced by LogDensity(), and SetUp().

◆ gamma

std::pair<double, double> muq::SamplingAlgorithms::ExpensiveSamplingProblem::gamma

private

Parameters for structural refinement.

Refine if the error threshold exceeds \(\gamma = \gamma_0 l^{-\gamma_1}\), where \(l\) is the current error threshold level (ExpensiveSamplingProblem::level). \(\gamma_0\) is "GammaScale" and it defaults to \(1\). \(\gamma_1\) is "GammaExponent" and it defaults to \(1.0\).

Definition at line 177 of file ExpensiveSamplingProblem.h.

Referenced by CheckNumNeighbors(), ErrorThreshold(), RefineSurrogate(), and SetUp().