A class for handling sampling from (transformed) Gaussian probability density distributions with bounds. More...

#include <InvLogitGaussianVectorRealizer.h>

Inheritance diagram for QUESO::InvLogitGaussianVectorRealizer< V, M >:

Public Member Functions
Constructor/Destructor methods
	InvLogitGaussianVectorRealizer (const char *prefix, const BoxSubset< V, M > &unifiedImageBoxSubset, const V &lawExpVector, const M &lowerCholLawCovMatrix)
	Constructor. More...

	InvLogitGaussianVectorRealizer (const char *prefix, const BoxSubset< V, M > &unifiedImageBoxSubset, const V &lawExpVector, const M &matU, const V &vecSsqrt, const M &matVt)
	Constructor. More...

	~InvLogitGaussianVectorRealizer ()
	Destructor. More...

Realization-related methods
const V &	unifiedLawExpVector () const
	Access to the vector of mean values of the Gaussian and private attribute: m_unifiedLawExpVector. More...

const V &	unifiedLawVarVector () const
	Access to the vector of variance values and private attribute: m_unifiedLawVarVector. More...

void	realization (V &nextValues) const
	Draws a realization. More...

void	updateLawExpVector (const V &newLawExpVector)
	Updates the mean of the Gaussian with the new value `newLawExpVector`. More...

void	updateLowerCholLawCovMatrix (const M &newLowerCholLawCovMatrix)
	Updates the lower triangular matrix from Cholesky decomposition of the covariance matrix to the new value `newLowerCholLawCovMatrix`. More...

void	updateLowerCholLawCovMatrix (const M &matU, const V &vecSsqrt, const M &matVt)
	Updates the SVD matrices from SVD decomposition of the covariance matrix to the new values: `matU`, `vecSsqrt`, and `matVt`. More...

Public Member Functions inherited from QUESO::BaseVectorRealizer< V, M >
	BaseVectorRealizer (const char *prefix, const VectorSet< V, M > &unifiedImageSet, unsigned int subPeriod)
	Default constructor. More...

virtual	~BaseVectorRealizer ()
	Virtual destructor. More...

const VectorSet< V, M > &	unifiedImageSet () const
	Image set where the realizations lie. Access to protected attribute m_unifiedImageSet. More...

unsigned int	subPeriod () const
	Sub-period of the realization. Access to protected attribute m_subPeriod. More...

Private Attributes
V *	m_unifiedLawExpVector

V *	m_unifiedLawVarVector

M *	m_lowerCholLawCovMatrix

M *	m_matU

V *	m_vecSsqrt

M *	m_matVt

const BoxSubset< V, M > &	m_unifiedImageBoxSubset

Additional Inherited Members
Protected Attributes inherited from QUESO::BaseVectorRealizer< V, M >
const BaseEnvironment &	m_env

std::string	m_prefix

const VectorSet< V, M > &	m_unifiedImageSet

unsigned int	m_subPeriod

Detailed Description

template<class V = GslVector, class M = GslMatrix>
class QUESO::InvLogitGaussianVectorRealizer< V, M >

A class for handling sampling from (transformed) Gaussian probability density distributions with bounds.

To get a realization from a transformed Gaussian (i.e. with bounds), an inverse logit transform is applied. I.e., samples are \( f(X) \) where \( X \) is drawn from a Gaussian and where

\[ f(x) = \frac{b \exp(x) + a}{1 + \exp(x)}. \]

This will produce a sample in the closed interval [a, b].

Definition at line 51 of file InvLogitGaussianVectorRealizer.h.

Constructor & Destructor Documentation

template<class V , class M >

QUESO::InvLogitGaussianVectorRealizer< V, M >::InvLogitGaussianVectorRealizer	(	const char *	prefix,
		const BoxSubset< V, M > &	unifiedImageBoxSubset,
		const V &	lawExpVector,
		const M &	lowerCholLawCovMatrix
	)

Constructor.

Constructs a new object, given a prefix and the image set of the vector realizer, a vector of mean values, lawExpVector (of the Gaussian, not the transformed Gaussian), and a lower triangular matrix resulting from Cholesky decomposition of the covariance matrix, lowerCholLawCovMatrix.

Definition at line 35 of file InvLogitGaussianVectorRealizer.C.

References QUESO::InvLogitGaussianVectorRealizer< V, M >::m_unifiedLawExpVector.

   : BaseVectorRealizer<V, M>(((std::string)(prefix)+"invlogit_gau").c_str(),
       unifiedImageBoxSubset, std::numeric_limits<unsigned int>::max()),
     m_unifiedLawExpVector(new V(lawExpVector)),
     m_unifiedLawVarVector(
         unifiedImageBoxSubset.vectorSpace().newVector(INFINITY)),  // FIX ME
     m_lowerCholLawCovMatrix(new M(lowerCholLawCovMatrix)),
     m_matU(NULL),
     m_vecSsqrt(NULL),
     m_matVt(NULL),
     m_unifiedImageBoxSubset(unifiedImageBoxSubset)
 {
   *m_unifiedLawExpVector = lawExpVector;
 }

template<class V , class M >

QUESO::InvLogitGaussianVectorRealizer< V, M >::InvLogitGaussianVectorRealizer	(	const char *	prefix,
		const BoxSubset< V, M > &	unifiedImageBoxSubset,
		const V &	lawExpVector,
		const M &	matU,
		const V &	vecSsqrt,
		const M &	matVt
	)

Constructor.

Constructs a new object, given a prefix and the image set of the vector realizer, a vector of mean values, lawExpVector (of the Gaussian, not the transformed Gaussian), and a set of two matrices and one vector resulting from the Single Value Decomposition of the covariance matrix, matU, vecSsqrt and matVt.

Definition at line 55 of file InvLogitGaussianVectorRealizer.C.

References QUESO::InvLogitGaussianVectorRealizer< V, M >::m_unifiedLawExpVector.

   : BaseVectorRealizer<V, M>(((std::string)(prefix)+"invlogit_gau").c_str(),
       unifiedImageBoxSubset, std::numeric_limits<unsigned int>::max()),
     m_unifiedLawExpVector(new V(lawExpVector)),
     m_unifiedLawVarVector(
         unifiedImageBoxSubset.vectorSpace().newVector( INFINITY)), // FIX ME
     m_lowerCholLawCovMatrix(NULL),
     m_matU(new M(matU)),
     m_vecSsqrt(new V(vecSsqrt)),
     m_matVt(new M(matVt)),
     m_unifiedImageBoxSubset(unifiedImageBoxSubset)
 {
   *m_unifiedLawExpVector = lawExpVector; // ????
 }

template<class V , class M >

QUESO::InvLogitGaussianVectorRealizer< V, M >::~InvLogitGaussianVectorRealizer ( )

Destructor.

Definition at line 77 of file InvLogitGaussianVectorRealizer.C.

 {
   delete m_matVt;
   delete m_vecSsqrt;
   delete m_matU;
   delete m_lowerCholLawCovMatrix;
   delete m_unifiedLawVarVector;
   delete m_unifiedLawExpVector;
 }

Member Function Documentation

template<class V , class M >

void QUESO::InvLogitGaussianVectorRealizer< V, M >::realization ( V & nextValues ) const

virtual

Draws a realization.

This function draws a realization of a (transformed) Gaussian distribution with mean m_unifiedLawExpVector and variance m_unifiedLawVarVector and saves it in nextValues.

Implements QUESO::BaseVectorRealizer< V, M >.

Definition at line 103 of file InvLogitGaussianVectorRealizer.C.

References QUESO::queso_isfinite().

 {
   V iidGaussianVector(m_unifiedImageSet.vectorSpace().zeroVector());
 
   iidGaussianVector.cwSetGaussian(0.0, 1.0);
 
   if (m_lowerCholLawCovMatrix) {
     nextValues = (*m_unifiedLawExpVector) +
       (*m_lowerCholLawCovMatrix) * iidGaussianVector;
   }
   else if (m_matU && m_vecSsqrt && m_matVt) {
     nextValues = (*m_unifiedLawExpVector) +
       (*m_matU) * ((*m_vecSsqrt) * ((*m_matVt) * iidGaussianVector));
   }
   else {
     queso_error_msg("GaussianVectorRealizer<V,M>::realization() inconsistent internal state");
   }
 
   V min_domain_bounds(this->m_unifiedImageBoxSubset.minValues());
   V max_domain_bounds(this->m_unifiedImageBoxSubset.maxValues());
 
   for (unsigned int i = 0; i < nextValues.sizeLocal(); i++) {
     double temp = std::exp(nextValues[i]);
     double min_val = min_domain_bounds[i];
     double max_val = max_domain_bounds[i];
 
     if (queso_isfinite(min_val) &&
         queso_isfinite(max_val)) {
         // Left- and right-hand sides are finite.  Do full transform.
         nextValues[i] = (max_val * temp + min_val) / (1.0 + temp);
       }
     else if (queso_isfinite(min_val) &&
              !queso_isfinite(max_val)) {
       // Left-hand side finite, but right-hand side is not.
       // Do only left-hand transform.
       nextValues[i] = temp + min_val;
     }
     else if (!queso_isfinite(min_val) &&
              queso_isfinite(max_val)) {
       // Right-hand side is finite, but left-hand side is not.
       // Do only right-hand transform.
       nextValues[i] = (max_val * temp - 1.0) / temp;
     }
   }
 }

template<class V , class M >

const V & QUESO::InvLogitGaussianVectorRealizer< V, M >::unifiedLawExpVector ( ) const

Access to the vector of mean values of the Gaussian and private attribute: m_unifiedLawExpVector.

Definition at line 89 of file InvLogitGaussianVectorRealizer.C.

 {
   return *m_unifiedLawExpVector;
 }

template<class V , class M >

const V & QUESO::InvLogitGaussianVectorRealizer< V, M >::unifiedLawVarVector ( ) const

Access to the vector of variance values and private attribute: m_unifiedLawVarVector.

Definition at line 96 of file InvLogitGaussianVectorRealizer.C.

 {
   return *m_unifiedLawVarVector;
 }

template<class V , class M >

void QUESO::InvLogitGaussianVectorRealizer< V, M >::updateLawExpVector ( const V & newLawExpVector )

Updates the mean of the Gaussian with the new value newLawExpVector.

Definition at line 151 of file InvLogitGaussianVectorRealizer.C.

 {
   // delete old expected values (allocated at construction or last call to this function)
   delete m_unifiedLawExpVector;
 
   m_unifiedLawExpVector = new V(newLawExpVector);
 }

template<class V , class M >

void QUESO::InvLogitGaussianVectorRealizer< V, M >::updateLowerCholLawCovMatrix ( const M & newLowerCholLawCovMatrix )

Updates the lower triangular matrix from Cholesky decomposition of the covariance matrix to the new value newLowerCholLawCovMatrix.

The lower triangular matrix results resulting from a Cholesky decomposition of the covariance matrix. This routine deletes old expected values: m_lowerCholLawCovMatrix; m_matU, m_vecSsqrt, m_matVt.

Definition at line 162 of file InvLogitGaussianVectorRealizer.C.

 {
   // delete old expected values (allocated at construction or last call to this function)
   delete m_lowerCholLawCovMatrix;
   delete m_matU;
   delete m_vecSsqrt;
   delete m_matVt;
 
   m_lowerCholLawCovMatrix = new M(newLowerCholLawCovMatrix);
   m_matU                  = NULL;
   m_vecSsqrt              = NULL;
   m_matVt                 = NULL;
 }

template<class V , class M >

void QUESO::InvLogitGaussianVectorRealizer< V, M >::updateLowerCholLawCovMatrix	(	const M &	matU,
		const V &	vecSsqrt,
		const M &	matVt
	)

Updates the SVD matrices from SVD decomposition of the covariance matrix to the new values: matU, vecSsqrt, and matVt.

The lower triangular matrix results resulting from a Cholesky decomposition of the covariance matrix. This routine deletes old expected values: m_lowerCholLawCovMatrix; m_matU, m_vecSsqrt, m_matVt.

Definition at line 179 of file InvLogitGaussianVectorRealizer.C.

 {
   // delete old expected values (allocated at construction or last call to this function)
   delete m_lowerCholLawCovMatrix;
   delete m_matU;
   delete m_vecSsqrt;
   delete m_matVt;
 
   m_lowerCholLawCovMatrix = NULL;
   m_matU                  = new M(matU);
   m_vecSsqrt              = new V(vecSsqrt);
   m_matVt                 = new M(matVt);
 }