• 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 #include <iostream>
11 #include <fstream>
12 #include <iomanip>
13 
14 #include "main.h"
15 #include <Eigen/LevenbergMarquardt>
16 
17 using namespace std;
18 using namespace Eigen;
19 
20 template <typename Scalar>
21 struct sparseGaussianTest : SparseFunctor<Scalar, int>
22 {
23   typedef Matrix<Scalar,Dynamic,1> VectorType;
24   typedef SparseFunctor<Scalar,int> Base;
25   typedef typename Base::JacobianType JacobianType;
sparseGaussianTestsparseGaussianTest26   sparseGaussianTest(int inputs, int values) : SparseFunctor<Scalar,int>(inputs,values)
27   { }
28 
modelsparseGaussianTest29   VectorType model(const VectorType& uv, VectorType& x)
30   {
31     VectorType y; //Change this to use expression template
32     int m = Base::values();
33     int n = Base::inputs();
34     eigen_assert(uv.size()%2 == 0);
35     eigen_assert(uv.size() == n);
36     eigen_assert(x.size() == m);
37     y.setZero(m);
38     int half = n/2;
39     VectorBlock<const VectorType> u(uv, 0, half);
40     VectorBlock<const VectorType> v(uv, half, half);
41     Scalar coeff;
42     for (int j = 0; j < m; j++)
43     {
44       for (int i = 0; i < half; i++)
45       {
46         coeff = (x(j)-i)/v(i);
47         coeff *= coeff;
48         if (coeff < 1. && coeff > 0.)
49           y(j) += u(i)*std::pow((1-coeff), 2);
50       }
51     }
52     return y;
53   }
initPointssparseGaussianTest54   void initPoints(VectorType& uv_ref, VectorType& x)
55   {
56     m_x = x;
57     m_y = this->model(uv_ref,x);
58   }
operator ()sparseGaussianTest59   int operator()(const VectorType& uv, VectorType& fvec)
60   {
61     int m = Base::values();
62     int n = Base::inputs();
63     eigen_assert(uv.size()%2 == 0);
64     eigen_assert(uv.size() == n);
65     int half = n/2;
66     VectorBlock<const VectorType> u(uv, 0, half);
67     VectorBlock<const VectorType> v(uv, half, half);
68     fvec = m_y;
69     Scalar coeff;
70     for (int j = 0; j < m; j++)
71     {
72       for (int i = 0; i < half; i++)
73       {
74         coeff = (m_x(j)-i)/v(i);
75         coeff *= coeff;
76         if (coeff < 1. && coeff > 0.)
77           fvec(j) -= u(i)*std::pow((1-coeff), 2);
78       }
79     }
80     return 0;
81   }
82 
dfsparseGaussianTest83   int df(const VectorType& uv, JacobianType& fjac)
84   {
85     int m = Base::values();
86     int n = Base::inputs();
87     eigen_assert(n == uv.size());
88     eigen_assert(fjac.rows() == m);
89     eigen_assert(fjac.cols() == n);
90     int half = n/2;
91     VectorBlock<const VectorType> u(uv, 0, half);
92     VectorBlock<const VectorType> v(uv, half, half);
93     Scalar coeff;
94 
95     //Derivatives with respect to u
96     for (int col = 0; col < half; col++)
97     {
98       for (int row = 0; row < m; row++)
99       {
100         coeff = (m_x(row)-col)/v(col);
101           coeff = coeff*coeff;
102         if(coeff < 1. && coeff > 0.)
103         {
104           fjac.coeffRef(row,col) = -(1-coeff)*(1-coeff);
105         }
106       }
107     }
108     //Derivatives with respect to v
109     for (int col = 0; col < half; col++)
110     {
111       for (int row = 0; row < m; row++)
112       {
113         coeff = (m_x(row)-col)/v(col);
114         coeff = coeff*coeff;
115         if(coeff < 1. && coeff > 0.)
116         {
117           fjac.coeffRef(row,col+half) = -4 * (u(col)/v(col))*coeff*(1-coeff);
118         }
119       }
120     }
121     return 0;
122   }
123 
124   VectorType m_x, m_y; //Data points
125 };
126 
127 
128 template<typename T>
test_sparseLM_T()129 void test_sparseLM_T()
130 {
131   typedef Matrix<T,Dynamic,1> VectorType;
132 
133   int inputs = 10;
134   int values = 2000;
135   sparseGaussianTest<T> sparse_gaussian(inputs, values);
136   VectorType uv(inputs),uv_ref(inputs);
137   VectorType x(values);
138   // Generate the reference solution
139   uv_ref << -2, 1, 4 ,8, 6, 1.8, 1.2, 1.1, 1.9 , 3;
140   //Generate the reference data points
141   x.setRandom();
142   x = 10*x;
143   x.array() += 10;
144   sparse_gaussian.initPoints(uv_ref, x);
145 
146 
147   // Generate the initial parameters
148   VectorBlock<VectorType> u(uv, 0, inputs/2);
149   VectorBlock<VectorType> v(uv, inputs/2, inputs/2);
150   v.setOnes();
151   //Generate u or Solve for u from v
152   u.setOnes();
153 
154   // Solve the optimization problem
155   LevenbergMarquardt<sparseGaussianTest<T> > lm(sparse_gaussian);
156   int info;
157 //   info = lm.minimize(uv);
158 
159   VERIFY_IS_EQUAL(info,1);
160     // Do a step by step solution and save the residual
161   int maxiter = 200;
162   int iter = 0;
163   MatrixXd Err(values, maxiter);
164   MatrixXd Mod(values, maxiter);
165   LevenbergMarquardtSpace::Status status;
166   status = lm.minimizeInit(uv);
167   if (status==LevenbergMarquardtSpace::ImproperInputParameters)
168       return ;
169 
170 }
test_sparseLM()171 void test_sparseLM()
172 {
173   CALL_SUBTEST_1(test_sparseLM_T<double>());
174 
175   // CALL_SUBTEST_2(test_sparseLM_T<std::complex<double>());
176 }
177