Carlos Hernandez | 7faaa9f | 2014-08-05 17:53:32 -0700 | [diff] [blame^] | 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; |
| 26 | sparseGaussianTest(int inputs, int values) : SparseFunctor<Scalar,int>(inputs,values) |
| 27 | { } |
| 28 | |
| 29 | 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 | } |
| 54 | void initPoints(VectorType& uv_ref, VectorType& x) |
| 55 | { |
| 56 | m_x = x; |
| 57 | m_y = this->model(uv_ref,x); |
| 58 | } |
| 59 | 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 | |
| 83 | 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> |
| 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 | } |
| 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 | } |