• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2012 Desire Nuentsa <desire.nuentsa_wakam@inria.fr>
5 // Copyright (C) 2012 Gael Guennebaud <gael.guennebaud@inria.fr>
6 //
7 // This Source Code Form is subject to the terms of the Mozilla
8 // Public License v. 2.0. If a copy of the MPL was not distributed
9 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
10 
11 #include <iostream>
12 #include <fstream>
13 #include <iomanip>
14 
15 #include "main.h"
16 #include <Eigen/LevenbergMarquardt>
17 using namespace std;
18 using namespace Eigen;
19 
20 template<typename Scalar>
21 struct DenseLM : DenseFunctor<Scalar>
22 {
23   typedef DenseFunctor<Scalar> Base;
24   typedef typename Base::JacobianType JacobianType;
25   typedef Matrix<Scalar,Dynamic,1> VectorType;
26 
DenseLMDenseLM27   DenseLM(int n, int m) : DenseFunctor<Scalar>(n,m)
28   { }
29 
modelDenseLM30   VectorType model(const VectorType& uv, VectorType& x)
31   {
32     VectorType y; // Should change to use expression template
33     int m = Base::values();
34     int n = Base::inputs();
35     eigen_assert(uv.size()%2 == 0);
36     eigen_assert(uv.size() == n);
37     eigen_assert(x.size() == m);
38     y.setZero(m);
39     int half = n/2;
40     VectorBlock<const VectorType> u(uv, 0, half);
41     VectorBlock<const VectorType> v(uv, half, half);
42     for (int j = 0; j < m; j++)
43     {
44       for (int i = 0; i < half; i++)
45         y(j) += u(i)*std::exp(-(x(j)-i)*(x(j)-i)/(v(i)*v(i)));
46     }
47     return y;
48 
49   }
initPointsDenseLM50   void initPoints(VectorType& uv_ref, VectorType& x)
51   {
52     m_x = x;
53     m_y = this->model(uv_ref, x);
54   }
55 
operator ()DenseLM56   int operator()(const VectorType& uv, VectorType& fvec)
57   {
58 
59     int m = Base::values();
60     int n = Base::inputs();
61     eigen_assert(uv.size()%2 == 0);
62     eigen_assert(uv.size() == n);
63     eigen_assert(fvec.size() == m);
64     int half = n/2;
65     VectorBlock<const VectorType> u(uv, 0, half);
66     VectorBlock<const VectorType> v(uv, half, half);
67     for (int j = 0; j < m; j++)
68     {
69       fvec(j) = m_y(j);
70       for (int i = 0; i < half; i++)
71       {
72         fvec(j) -= u(i) *std::exp(-(m_x(j)-i)*(m_x(j)-i)/(v(i)*v(i)));
73       }
74     }
75 
76     return 0;
77   }
dfDenseLM78   int df(const VectorType& uv, JacobianType& fjac)
79   {
80     int m = Base::values();
81     int n = Base::inputs();
82     eigen_assert(n == uv.size());
83     eigen_assert(fjac.rows() == m);
84     eigen_assert(fjac.cols() == n);
85     int half = n/2;
86     VectorBlock<const VectorType> u(uv, 0, half);
87     VectorBlock<const VectorType> v(uv, half, half);
88     for (int j = 0; j < m; j++)
89     {
90       for (int i = 0; i < half; i++)
91       {
92         fjac.coeffRef(j,i) = -std::exp(-(m_x(j)-i)*(m_x(j)-i)/(v(i)*v(i)));
93         fjac.coeffRef(j,i+half) = -2.*u(i)*(m_x(j)-i)*(m_x(j)-i)/(std::pow(v(i),3)) * std::exp(-(m_x(j)-i)*(m_x(j)-i)/(v(i)*v(i)));
94       }
95     }
96     return 0;
97   }
98   VectorType m_x, m_y; //Data Points
99 };
100 
101 template<typename FunctorType, typename VectorType>
test_minimizeLM(FunctorType & functor,VectorType & uv)102 int test_minimizeLM(FunctorType& functor, VectorType& uv)
103 {
104   LevenbergMarquardt<FunctorType> lm(functor);
105   LevenbergMarquardtSpace::Status info;
106 
107   info = lm.minimize(uv);
108 
109   VERIFY_IS_EQUAL(info, 1);
110   //FIXME Check other parameters
111   return info;
112 }
113 
114 template<typename FunctorType, typename VectorType>
test_lmder(FunctorType & functor,VectorType & uv)115 int test_lmder(FunctorType& functor, VectorType& uv)
116 {
117   typedef typename VectorType::Scalar Scalar;
118   LevenbergMarquardtSpace::Status info;
119   LevenbergMarquardt<FunctorType> lm(functor);
120   info = lm.lmder1(uv);
121 
122   VERIFY_IS_EQUAL(info, 1);
123   //FIXME Check other parameters
124   return info;
125 }
126 
127 template<typename FunctorType, typename VectorType>
test_minimizeSteps(FunctorType & functor,VectorType & uv)128 int test_minimizeSteps(FunctorType& functor, VectorType& uv)
129 {
130   LevenbergMarquardtSpace::Status info;
131   LevenbergMarquardt<FunctorType> lm(functor);
132   info = lm.minimizeInit(uv);
133   if (info==LevenbergMarquardtSpace::ImproperInputParameters)
134       return info;
135   do
136   {
137     info = lm.minimizeOneStep(uv);
138   } while (info==LevenbergMarquardtSpace::Running);
139 
140   VERIFY_IS_EQUAL(info, 1);
141   //FIXME Check other parameters
142   return info;
143 }
144 
145 template<typename T>
test_denseLM_T()146 void test_denseLM_T()
147 {
148   typedef Matrix<T,Dynamic,1> VectorType;
149 
150   int inputs = 10;
151   int values = 1000;
152   DenseLM<T> dense_gaussian(inputs, values);
153   VectorType uv(inputs),uv_ref(inputs);
154   VectorType x(values);
155 
156   // Generate the reference solution
157   uv_ref << -2, 1, 4 ,8, 6, 1.8, 1.2, 1.1, 1.9 , 3;
158 
159   //Generate the reference data points
160   x.setRandom();
161   x = 10*x;
162   x.array() += 10;
163   dense_gaussian.initPoints(uv_ref, x);
164 
165   // Generate the initial parameters
166   VectorBlock<VectorType> u(uv, 0, inputs/2);
167   VectorBlock<VectorType> v(uv, inputs/2, inputs/2);
168 
169   // Solve the optimization problem
170 
171   //Solve in one go
172   u.setOnes(); v.setOnes();
173   test_minimizeLM(dense_gaussian, uv);
174 
175   //Solve until the machine precision
176   u.setOnes(); v.setOnes();
177   test_lmder(dense_gaussian, uv);
178 
179   // Solve step by step
180   v.setOnes(); u.setOnes();
181   test_minimizeSteps(dense_gaussian, uv);
182 
183 }
184 
test_denseLM()185 void test_denseLM()
186 {
187   CALL_SUBTEST_2(test_denseLM_T<double>());
188 
189   // CALL_SUBTEST_2(test_sparseLM_T<std::complex<double>());
190 }
191