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: keir (at) google.com (Keir Mierle) 30 // 31 // TODO(keir): Implement a generic "compare sparse matrix implementations" test 32 // suite that can compare all the implementations. Then this file would shrink 33 // in size. 34 35 #include "ceres/dense_sparse_matrix.h" 36 37 #include "gtest/gtest.h" 38 #include "ceres/casts.h" 39 #include "ceres/linear_least_squares_problems.h" 40 #include "ceres/matrix_proto.h" 41 #include "ceres/triplet_sparse_matrix.h" 42 #include "ceres/internal/eigen.h" 43 #include "ceres/internal/scoped_ptr.h" 44 45 namespace ceres { 46 namespace internal { 47 48 void CompareMatrices(const SparseMatrix* a, const SparseMatrix* b) { 49 EXPECT_EQ(a->num_rows(), b->num_rows()); 50 EXPECT_EQ(a->num_cols(), b->num_cols()); 51 52 int num_rows = a->num_rows(); 53 int num_cols = a->num_cols(); 54 55 for (int i = 0; i < num_cols; ++i) { 56 Vector x = Vector::Zero(num_cols); 57 x(i) = 1.0; 58 59 Vector y_a = Vector::Zero(num_rows); 60 Vector y_b = Vector::Zero(num_rows); 61 62 a->RightMultiply(x.data(), y_a.data()); 63 b->RightMultiply(x.data(), y_b.data()); 64 65 EXPECT_EQ((y_a - y_b).norm(), 0); 66 } 67 } 68 69 class DenseSparseMatrixTest : public ::testing::Test { 70 protected : 71 virtual void SetUp() { 72 scoped_ptr<LinearLeastSquaresProblem> problem( 73 CreateLinearLeastSquaresProblemFromId(1)); 74 75 CHECK_NOTNULL(problem.get()); 76 77 tsm.reset(down_cast<TripletSparseMatrix*>(problem->A.release())); 78 dsm.reset(new DenseSparseMatrix(*tsm)); 79 80 num_rows = tsm->num_rows(); 81 num_cols = tsm->num_cols(); 82 } 83 84 int num_rows; 85 int num_cols; 86 87 scoped_ptr<TripletSparseMatrix> tsm; 88 scoped_ptr<DenseSparseMatrix> dsm; 89 }; 90 91 TEST_F(DenseSparseMatrixTest, RightMultiply) { 92 CompareMatrices(tsm.get(), dsm.get()); 93 94 // Try with a not entirely zero vector to verify column interactions, which 95 // could be masked by a subtle bug when using the elementary vectors. 96 Vector a(num_cols); 97 for (int i = 0; i < num_cols; i++) { 98 a(i) = i; 99 } 100 Vector b1 = Vector::Zero(num_rows); 101 Vector b2 = Vector::Zero(num_rows); 102 103 tsm->RightMultiply(a.data(), b1.data()); 104 dsm->RightMultiply(a.data(), b2.data()); 105 106 EXPECT_EQ((b1 - b2).norm(), 0); 107 } 108 109 TEST_F(DenseSparseMatrixTest, LeftMultiply) { 110 for (int i = 0; i < num_rows; ++i) { 111 Vector a = Vector::Zero(num_rows); 112 a(i) = 1.0; 113 114 Vector b1 = Vector::Zero(num_cols); 115 Vector b2 = Vector::Zero(num_cols); 116 117 tsm->LeftMultiply(a.data(), b1.data()); 118 dsm->LeftMultiply(a.data(), b2.data()); 119 120 EXPECT_EQ((b1 - b2).norm(), 0); 121 } 122 123 // Try with a not entirely zero vector to verify column interactions, which 124 // could be masked by a subtle bug when using the elementary vectors. 125 Vector a(num_rows); 126 for (int i = 0; i < num_rows; i++) { 127 a(i) = i; 128 } 129 Vector b1 = Vector::Zero(num_cols); 130 Vector b2 = Vector::Zero(num_cols); 131 132 tsm->LeftMultiply(a.data(), b1.data()); 133 dsm->LeftMultiply(a.data(), b2.data()); 134 135 EXPECT_EQ((b1 - b2).norm(), 0); 136 } 137 138 TEST_F(DenseSparseMatrixTest, ColumnNorm) { 139 Vector b1 = Vector::Zero(num_cols); 140 Vector b2 = Vector::Zero(num_cols); 141 142 tsm->SquaredColumnNorm(b1.data()); 143 dsm->SquaredColumnNorm(b2.data()); 144 145 EXPECT_EQ((b1 - b2).norm(), 0); 146 } 147 148 TEST_F(DenseSparseMatrixTest, Scale) { 149 Vector scale(num_cols); 150 for (int i = 0; i < num_cols; ++i) { 151 scale(i) = i + 1; 152 } 153 tsm->ScaleColumns(scale.data()); 154 dsm->ScaleColumns(scale.data()); 155 CompareMatrices(tsm.get(), dsm.get()); 156 } 157 158 #ifndef CERES_NO_PROTOCOL_BUFFERS 159 TEST_F(DenseSparseMatrixTest, Serialization) { 160 SparseMatrixProto proto; 161 dsm->ToProto(&proto); 162 163 DenseSparseMatrix n(proto); 164 ASSERT_EQ(dsm->num_rows(), n.num_rows()); 165 ASSERT_EQ(dsm->num_cols(), n.num_cols()); 166 ASSERT_EQ(dsm->num_nonzeros(), n.num_nonzeros()); 167 168 for (int i = 0; i < n.num_rows() + 1; ++i) { 169 ASSERT_EQ(dsm->values()[i], proto.dense_matrix().values(i)); 170 } 171 } 172 #endif 173 174 TEST_F(DenseSparseMatrixTest, ToDenseMatrix) { 175 Matrix tsm_dense; 176 Matrix dsm_dense; 177 178 tsm->ToDenseMatrix(&tsm_dense); 179 dsm->ToDenseMatrix(&dsm_dense); 180 181 EXPECT_EQ((tsm_dense - dsm_dense).norm(), 0.0); 182 } 183 184 // TODO(keir): Make this work without protocol buffers. 185 #ifndef CERES_NO_PROTOCOL_BUFFERS 186 TEST_F(DenseSparseMatrixTest, AppendDiagonal) { 187 DenseSparseMatrixProto proto; 188 proto.set_num_rows(3); 189 proto.set_num_cols(3); 190 for (int i = 0; i < 9; ++i) { 191 proto.add_values(i); 192 } 193 SparseMatrixProto outer_proto; 194 *outer_proto.mutable_dense_matrix() = proto; 195 196 DenseSparseMatrix dsm(outer_proto); 197 198 double diagonal[] = { 10, 11, 12 }; 199 dsm.AppendDiagonal(diagonal); 200 201 // Verify the diagonal got added. 202 Matrix m = dsm.matrix(); 203 EXPECT_EQ(6, m.rows()); 204 EXPECT_EQ(3, m.cols()); 205 for (int i = 0; i < 3; ++i) { 206 for (int j = 0; j < 3; ++j) { 207 EXPECT_EQ(3 * i + j, m(i, j)); 208 if (i == j) { 209 EXPECT_EQ(10 + i, m(i + 3, j)); 210 } else { 211 EXPECT_EQ(0, m(i + 3, j)); 212 } 213 } 214 } 215 216 // Verify the diagonal gets removed. 217 dsm.RemoveDiagonal(); 218 m = dsm.matrix(); 219 220 EXPECT_EQ(3, m.rows()); 221 EXPECT_EQ(3, m.cols()); 222 223 for (int i = 0; i < 3; ++i) { 224 for (int j = 0; j < 3; ++j) { 225 EXPECT_EQ(3 * i + j, m(i, j)); 226 } 227 } 228 } 229 #endif 230 231 } // namespace internal 232 } // namespace ceres 233