queso-0.55.0
InvLogitGaussianJointPdf.C
Go to the documentation of this file.
1 //-----------------------------------------------------------------------bl-
2 //--------------------------------------------------------------------------
3 //
4 // QUESO - a library to support the Quantification of Uncertainty
5 // for Estimation, Simulation and Optimization
6 //
7 // Copyright (C) 2008-2015 The PECOS Development Team
8 //
9 // This library is free software; you can redistribute it and/or
10 // modify it under the terms of the Version 2.1 GNU Lesser General
11 // Public License as published by the Free Software Foundation.
12 //
13 // This library is distributed in the hope that it will be useful,
14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 // Lesser General Public License for more details.
17 //
18 // You should have received a copy of the GNU Lesser General Public
19 // License along with this library; if not, write to the Free Software
20 // Foundation, Inc. 51 Franklin Street, Fifth Floor,
21 // Boston, MA 02110-1301 USA
22 //
23 //-----------------------------------------------------------------------el-
24 
25 #include <queso/InvLogitGaussianJointPdf.h>
26 #include <queso/GslVector.h>
27 #include <queso/GslMatrix.h>
28 
29 namespace QUESO {
30 
31 // Constructor -------------------------------------
32 template<class V,class M>
34  const char* prefix,
35  const BoxSubset<V,M>& domainBoxSubset,
36  const V& lawExpVector,
37  const V& lawVarVector)
38  :
39  BaseJointPdf<V,M>(((std::string)(prefix)+"invlogit_gau").c_str(),
40  domainBoxSubset),
41  m_lawExpVector(new V(lawExpVector)),
42  m_lawVarVector(new V(lawVarVector)),
43  m_diagonalCovMatrix(true),
44  m_lawCovMatrix(m_domainSet.vectorSpace().newDiagMatrix(lawVarVector)),
45  m_domainBoxSubset(domainBoxSubset)
46 {
47 }
48 
49 
50 template<class V,class M>
52  const char* prefix,
53  const BoxSubset<V,M>& domainBoxSubset,
54  const V& lawExpVector,
55  const M& lawCovMatrix)
56  :
57  BaseJointPdf<V,M>(((std::string)(prefix)+"invlogit_gau").c_str(),
58  domainBoxSubset),
59  m_lawExpVector(new V(lawExpVector)),
60  m_lawVarVector(domainBoxSubset.vectorSpace().newVector(INFINITY)), // FIX ME
61  m_diagonalCovMatrix(false),
62  m_lawCovMatrix(new M(lawCovMatrix)),
63  m_domainBoxSubset(domainBoxSubset)
64 {
65 }
66 
67 template<class V,class M>
69 {
70  delete m_lawCovMatrix;
71  delete m_lawVarVector;
72  delete m_lawExpVector;
73 }
74 
75 template <class V, class M>
76 const V&
78 {
79  return *m_lawExpVector;
80 }
81 
82 template <class V, class M>
83 const V&
85 {
86  return *m_lawVarVector;
87 }
88 
89 template<class V, class M>
90 double
92  const V& domainVector,
93  const V* domainDirection,
94  V* gradVector,
95  M* hessianMatrix,
96  V* hessianEffect) const
97 {
98  double returnValue;
99 
100  returnValue = std::exp(this->lnValue(domainVector,domainDirection,gradVector,hessianMatrix,hessianEffect));
101 
102  return returnValue;
103 }
104 
105 template<class V, class M>
106 double
108  const V& domainVector,
109  const V* domainDirection,
110  V* gradVector,
111  M* hessianMatrix,
112  V* hessianEffect) const
113 {
114  double returnValue;
115  double lnDeterminant = 0.0;
116  V transformedDomainVector(domainVector);
117 
118  V min_domain_bounds(this->m_domainBoxSubset.minValues());
119  V max_domain_bounds(this->m_domainBoxSubset.maxValues());
120 
121  double lnjacobian = 0.0;
122  for (unsigned int i = 0; i < domainVector.sizeLocal(); i++) {
123  double min_val = min_domain_bounds[i];
124  double max_val = max_domain_bounds[i];
125 
126  if (boost::math::isfinite(min_val) &&
127  boost::math::isfinite(max_val)) {
128 
129  if (domainVector[i] == min_val || domainVector[i] == max_val) {
130  // Exit early if we can
131  return -INFINITY;
132  }
133 
134  // Left- and right-hand sides are finite. Do full transform.
135  transformedDomainVector[i] = std::log(domainVector[i] - min_val) -
136  std::log(max_val - domainVector[i]);
137 
138  lnjacobian += std::log(max_val - min_val) -
139  std::log(domainVector[i] - min_val) -
140  std::log(max_val - domainVector[i]);
141  }
142  else if (boost::math::isfinite(min_val) &&
143  !boost::math::isfinite(max_val)) {
144 
145  if (domainVector[i] == min_val) {
146  // Exit early if we can
147  return -INFINITY;
148  }
149 
150  // Left-hand side finite, but right-hand side is not.
151  // Do only left-hand transform.
152  transformedDomainVector[i] = std::log(domainVector[i] - min_val);
153 
154  lnjacobian += -std::log(domainVector[i] - min_val);
155  }
156  else if (!boost::math::isfinite(min_val) &&
157  boost::math::isfinite(max_val)) {
158 
159  if (domainVector[i] == max_val) {
160  // Exit early if we can
161  return -INFINITY;
162  }
163 
164  // Right-hand side is finite, but left-hand side is not.
165  // Do only right-hand transform.
166  transformedDomainVector[i] = -std::log(max_val - domainVector[i]);
167 
168  lnjacobian += -std::log(max_val - domainVector[i]);
169  }
170  else {
171  // No transform.
172  transformedDomainVector[i] = domainVector[i];
173  }
174  }
175 
176  V diffVec(transformedDomainVector - this->lawExpVector());
177  if (m_diagonalCovMatrix) {
178  returnValue = ((diffVec * diffVec) /
179  this->lawVarVector()).sumOfComponents();
180  if (m_normalizationStyle == 0) {
181  unsigned int iMax = this->lawVarVector().sizeLocal();
182  for (unsigned int i = 0; i < iMax; ++i) {
183  lnDeterminant += log(this->lawVarVector()[i]);
184  }
185  }
186  }
187  else {
188  V tmpVec = this->m_lawCovMatrix->invertMultiply(diffVec);
189  returnValue = (diffVec * tmpVec).sumOfComponents();
190  if (m_normalizationStyle == 0) {
191  lnDeterminant = this->m_lawCovMatrix->lnDeterminant();
192  }
193  }
194  if (m_normalizationStyle == 0) {
195  returnValue += ((double) this->lawVarVector().sizeLocal()) * log(2 * M_PI);
196  returnValue += lnDeterminant;
197  }
198  returnValue *= -0.5;
199  returnValue += m_logOfNormalizationFactor;
200  returnValue += lnjacobian;
201 
202  return returnValue;
203 }
204 
205 template<class V, class M>
206 double
207 InvLogitGaussianJointPdf<V,M>::computeLogOfNormalizationFactor(unsigned int numSamples, bool updateFactorInternally) const
208 {
209  double value = 0.;
210 
211  if ((m_env.subDisplayFile()) && (m_env.displayVerbosity() >= 2)) {
212  *m_env.subDisplayFile() << "Entering GaussianJointPdf<V,M>::computeLogOfNormalizationFactor()"
213  << std::endl;
214  }
215  value = BaseJointPdf<V,M>::commonComputeLogOfNormalizationFactor(numSamples, updateFactorInternally);
216  if ((m_env.subDisplayFile()) && (m_env.displayVerbosity() >= 2)) {
217  *m_env.subDisplayFile() << "Leaving GaussianJointPdf<V,M>::computeLogOfNormalizationFactor()"
218  << ", m_logOfNormalizationFactor = " << m_logOfNormalizationFactor
219  << std::endl;
220  }
221 
222  return value;
223 }
224 
225 template<class V, class M>
226 void
228 {
229  // delete old expected values (allocated at construction or last call to this function)
230  delete m_lawExpVector;
231  m_lawExpVector = new V(newLawExpVector);
232 }
233 
234 template<class V, class M>
235 void
237 {
238  // delete old expected values (allocated at construction or last call to this function)
239  delete m_lawCovMatrix;
240  m_lawCovMatrix = new M(newLawCovMatrix);
241 }
242 
243 template<class V, class M>
244 const M&
246 {
247  return *m_lawCovMatrix;
248 }
249 
250 } // End namespace QUESO
251 
void updateLawExpVector(const V &newLawExpVector)
Updates the mean of the Gaussian (not transformed) with the new value newLawExpVector.
double lnValue(const V &domainVector, const V *domainDirection, V *gradVector, M *hessianMatrix, V *hessianEffect) const
Logarithm of the value of the (transformed) Gaussian PDF (scalar function).
A templated (base) class for handling joint PDFs.
Definition: JointPdf.h:56
const V & lawExpVector() const
Access to the vector of mean values of the Gaussian (not transformed) and private attribute: m_lawExp...
double commonComputeLogOfNormalizationFactor(unsigned int numSamples, bool updateFactorInternally) const
Common method (to the derived classes) to compute the logarithm of the normalization factor...
Definition: JointPdf.C:77
void updateLawCovMatrix(const M &newLawCovMatrix)
Updates the lower triangular matrix from Cholesky decomposition of the covariance matrix to the new v...
const V & lawVarVector() const
Access to the vector of variance values and private attribute: m_lawVarVector.
double actualValue(const V &domainVector, const V *domainDirection, V *gradVector, M *hessianMatrix, V *hessianEffect) const
Actual value of the (transformed) Gaussian PDF.
const M & lawCovMatrix() const
Returns the covariance matrix; access to protected attribute m_lawCovMatrix.
double computeLogOfNormalizationFactor(unsigned int numSamples, bool updateFactorInternally) const
Computes the logarithm of the normalization factor.
Class representing a subset of a vector space shaped like a hypercube.
Definition: BoxSubset.h:44
InvLogitGaussianJointPdf(const char *prefix, const BoxSubset< V, M > &domainBoxSubset, const V &lawExpVector, const V &lawVarVector)
Constructor.
A class for handling hybrid (transformed) Gaussians with bounds.

Generated on Fri Jun 17 2016 14:17:41 for queso-0.55.0 by  doxygen 1.8.5