1 // Ceres Solver - A fast non-linear least squares minimizer 2 // Copyright 2010, 2011, 2012 Google Inc. All rights reserved. 3 // http://code.google.com/p/ceres-solver/ 4 // 5 // Redistribution and use in source and binary forms, with or without 6 // modification, are permitted provided that the following conditions are met: 7 // 8 // * Redistributions of source code must retain the above copyright notice, 9 // this list of conditions and the following disclaimer. 10 // * Redistributions in binary form must reproduce the above copyright notice, 11 // this list of conditions and the following disclaimer in the documentation 12 // and/or other materials provided with the distribution. 13 // * Neither the name of Google Inc. nor the names of its contributors may be 14 // used to endorse or promote products derived from this software without 15 // specific prior written permission. 16 // 17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 18 // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 // ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 21 // LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 22 // CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 23 // SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 24 // INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 25 // CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 26 // ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 27 // POSSIBILITY OF SUCH DAMAGE. 28 // 29 // Author: sameeragarwal (at) google.com (Sameer Agarwal) 30 31 #include <cstddef> 32 #include "ceres/block_sparse_matrix.h" 33 #include "ceres/block_structure.h" 34 #include "ceres/casts.h" 35 #include "ceres/internal/scoped_ptr.h" 36 #include "ceres/linear_least_squares_problems.h" 37 #include "ceres/linear_solver.h" 38 #include "ceres/schur_complement_solver.h" 39 #include "ceres/triplet_sparse_matrix.h" 40 #include "ceres/types.h" 41 #include "glog/logging.h" 42 #include "gtest/gtest.h" 43 44 namespace ceres { 45 namespace internal { 46 47 class SchurComplementSolverTest : public ::testing::Test { 48 protected: 49 void SetUpFromProblemId(int problem_id) { 50 scoped_ptr<LinearLeastSquaresProblem> problem( 51 CreateLinearLeastSquaresProblemFromId(problem_id)); 52 53 CHECK_NOTNULL(problem.get()); 54 A.reset(down_cast<BlockSparseMatrix*>(problem->A.release())); 55 b.reset(problem->b.release()); 56 D.reset(problem->D.release()); 57 58 num_cols = A->num_cols(); 59 num_rows = A->num_rows(); 60 num_eliminate_blocks = problem->num_eliminate_blocks; 61 62 x.reset(new double[num_cols]); 63 sol.reset(new double[num_cols]); 64 sol_d.reset(new double[num_cols]); 65 66 LinearSolver::Options options; 67 options.type = DENSE_QR; 68 69 scoped_ptr<LinearSolver> qr(LinearSolver::Create(options)); 70 71 TripletSparseMatrix triplet_A(A->num_rows(), 72 A->num_cols(), 73 A->num_nonzeros()); 74 A->ToTripletSparseMatrix(&triplet_A); 75 76 // Gold standard solutions using dense QR factorization. 77 DenseSparseMatrix dense_A(triplet_A); 78 qr->Solve(&dense_A, b.get(), LinearSolver::PerSolveOptions(), sol.get()); 79 80 // Gold standard solution with appended diagonal. 81 LinearSolver::PerSolveOptions per_solve_options; 82 per_solve_options.D = D.get(); 83 qr->Solve(&dense_A, b.get(), per_solve_options, sol_d.get()); 84 } 85 86 void ComputeAndCompareSolutions( 87 int problem_id, 88 bool regularization, 89 ceres::LinearSolverType linear_solver_type, 90 ceres::DenseLinearAlgebraLibraryType dense_linear_algebra_library_type, 91 ceres::SparseLinearAlgebraLibraryType sparse_linear_algebra_library_type, 92 bool use_postordering) { 93 SetUpFromProblemId(problem_id); 94 LinearSolver::Options options; 95 options.elimination_groups.push_back(num_eliminate_blocks); 96 options.elimination_groups.push_back( 97 A->block_structure()->cols.size() - num_eliminate_blocks); 98 options.type = linear_solver_type; 99 options.dense_linear_algebra_library_type = 100 dense_linear_algebra_library_type; 101 options.sparse_linear_algebra_library_type = 102 sparse_linear_algebra_library_type; 103 options.use_postordering = use_postordering; 104 105 scoped_ptr<LinearSolver> solver(LinearSolver::Create(options)); 106 107 LinearSolver::PerSolveOptions per_solve_options; 108 LinearSolver::Summary summary; 109 if (regularization) { 110 per_solve_options.D = D.get(); 111 } 112 113 summary = solver->Solve(A.get(), b.get(), per_solve_options, x.get()); 114 115 if (regularization) { 116 for (int i = 0; i < num_cols; ++i) { 117 ASSERT_NEAR(sol_d.get()[i], x[i], 1e-10); 118 } 119 } else { 120 for (int i = 0; i < num_cols; ++i) { 121 ASSERT_NEAR(sol.get()[i], x[i], 1e-10); 122 } 123 } 124 } 125 126 int num_rows; 127 int num_cols; 128 int num_eliminate_blocks; 129 130 scoped_ptr<BlockSparseMatrix> A; 131 scoped_array<double> b; 132 scoped_array<double> x; 133 scoped_array<double> D; 134 scoped_array<double> sol; 135 scoped_array<double> sol_d; 136 }; 137 138 TEST_F(SchurComplementSolverTest, EigenBasedDenseSchurWithSmallProblem) { 139 ComputeAndCompareSolutions(2, false, DENSE_SCHUR, EIGEN, SUITE_SPARSE, true); 140 ComputeAndCompareSolutions(2, true, DENSE_SCHUR, EIGEN, SUITE_SPARSE, true); 141 } 142 143 TEST_F(SchurComplementSolverTest, EigenBasedDenseSchurWithLargeProblem) { 144 ComputeAndCompareSolutions(3, false, DENSE_SCHUR, EIGEN, SUITE_SPARSE, true); 145 ComputeAndCompareSolutions(3, true, DENSE_SCHUR, EIGEN, SUITE_SPARSE, true); 146 } 147 148 #ifndef CERES_NO_LAPACK 149 TEST_F(SchurComplementSolverTest, LAPACKBasedDenseSchurWithSmallProblem) { 150 ComputeAndCompareSolutions(2, false, DENSE_SCHUR, LAPACK, SUITE_SPARSE, true); 151 ComputeAndCompareSolutions(2, true, DENSE_SCHUR, LAPACK, SUITE_SPARSE, true); 152 } 153 154 TEST_F(SchurComplementSolverTest, LAPACKBasedDenseSchurWithLargeProblem) { 155 ComputeAndCompareSolutions(3, false, DENSE_SCHUR, LAPACK, SUITE_SPARSE, true); 156 ComputeAndCompareSolutions(3, true, DENSE_SCHUR, LAPACK, SUITE_SPARSE, true); 157 } 158 #endif 159 160 #ifndef CERES_NO_SUITESPARSE 161 TEST_F(SchurComplementSolverTest, 162 SparseSchurWithSuiteSparseSmallProblemNoPostOrdering) { 163 ComputeAndCompareSolutions( 164 2, false, SPARSE_SCHUR, EIGEN, SUITE_SPARSE, false); 165 ComputeAndCompareSolutions(2, true, SPARSE_SCHUR, EIGEN, SUITE_SPARSE, false); 166 } 167 168 TEST_F(SchurComplementSolverTest, 169 SparseSchurWithSuiteSparseSmallProblemPostOrdering) { 170 ComputeAndCompareSolutions(2, false, SPARSE_SCHUR, EIGEN, SUITE_SPARSE, true); 171 ComputeAndCompareSolutions(2, true, SPARSE_SCHUR, EIGEN, SUITE_SPARSE, true); 172 } 173 174 TEST_F(SchurComplementSolverTest, 175 SparseSchurWithSuiteSparseLargeProblemNoPostOrdering) { 176 ComputeAndCompareSolutions( 177 3, false, SPARSE_SCHUR, EIGEN, SUITE_SPARSE, false); 178 ComputeAndCompareSolutions(3, true, SPARSE_SCHUR, EIGEN, SUITE_SPARSE, false); 179 } 180 181 TEST_F(SchurComplementSolverTest, 182 SparseSchurWithSuiteSparseLargeProblemPostOrdering) { 183 ComputeAndCompareSolutions(3, false, SPARSE_SCHUR, EIGEN, SUITE_SPARSE, true); 184 ComputeAndCompareSolutions(3, true, SPARSE_SCHUR, EIGEN, SUITE_SPARSE, true); 185 } 186 #endif // CERES_NO_SUITESPARSE 187 188 #ifndef CERES_NO_CXSPARSE 189 TEST_F(SchurComplementSolverTest, 190 SparseSchurWithSuiteSparseSmallProblem) { 191 ComputeAndCompareSolutions(2, false, SPARSE_SCHUR, EIGEN, CX_SPARSE, true); 192 ComputeAndCompareSolutions(2, true, SPARSE_SCHUR, EIGEN, CX_SPARSE, true); 193 } 194 195 TEST_F(SchurComplementSolverTest, 196 SparseSchurWithSuiteSparseLargeProblem) { 197 ComputeAndCompareSolutions(3, false, SPARSE_SCHUR, EIGEN, CX_SPARSE, true); 198 ComputeAndCompareSolutions(3, true, SPARSE_SCHUR, EIGEN, CX_SPARSE, true); 199 } 200 #endif // CERES_NO_CXSPARSE 201 202 } // namespace internal 203 } // namespace ceres 204
