internal/ceres/partitioned_matrix_view_test.cc

   1 // Ceres Solver - A fast non-linear least squares minimizer
   2 // Copyright 2015 Google Inc. All rights reserved.
   3 // http://ceres-solver.org/
   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@google.com (Sameer Agarwal)
  30
  31 #include "ceres/partitioned_matrix_view.h"
  32
  33 #include <vector>
  34 #include "ceres/block_structure.h"
  35 #include "ceres/casts.h"
  36 #include "ceres/internal/eigen.h"
  37 #include "ceres/internal/scoped_ptr.h"
  38 #include "ceres/linear_least_squares_problems.h"
  39 #include "ceres/random.h"
  40 #include "ceres/sparse_matrix.h"
  41 #include "glog/logging.h"
  42 #include "gtest/gtest.h"
  43
  44 namespace ceres {
  45 namespace internal {
  46
  47 const double kEpsilon = 1e-14;
  48
  49 class PartitionedMatrixViewTest : public ::testing::Test {
  50  protected :
  51   virtual void SetUp() {
  52     srand(5);
  53     scoped_ptr<LinearLeastSquaresProblem> problem(
  54         CreateLinearLeastSquaresProblemFromId(2));
  55     CHECK_NOTNULL(problem.get());
  56     A_.reset(problem->A.release());
  57
  58     num_cols_ = A_->num_cols();
  59     num_rows_ = A_->num_rows();
  60     num_eliminate_blocks_ = problem->num_eliminate_blocks;
  61     LinearSolver::Options options;
  62     options.elimination_groups.push_back(num_eliminate_blocks_);
  63     pmv_.reset(PartitionedMatrixViewBase::Create(
  64                    options,
  65                    *down_cast<BlockSparseMatrix*>(A_.get())));
  66   }
  67
  68   int num_rows_;
  69   int num_cols_;
  70   int num_eliminate_blocks_;
  71   scoped_ptr<SparseMatrix> A_;
  72   scoped_ptr<PartitionedMatrixViewBase> pmv_;
  73 };
  74
  75 TEST_F(PartitionedMatrixViewTest, DimensionsTest) {
  76   EXPECT_EQ(pmv_->num_col_blocks_e(), num_eliminate_blocks_);
  77   EXPECT_EQ(pmv_->num_col_blocks_f(), num_cols_ - num_eliminate_blocks_);
  78   EXPECT_EQ(pmv_->num_cols_e(), num_eliminate_blocks_);
  79   EXPECT_EQ(pmv_->num_cols_f(), num_cols_ - num_eliminate_blocks_);
  80   EXPECT_EQ(pmv_->num_cols(), A_->num_cols());
  81   EXPECT_EQ(pmv_->num_rows(), A_->num_rows());
  82 }
  83
  84 TEST_F(PartitionedMatrixViewTest, RightMultiplyE) {
  85   Vector x1(pmv_->num_cols_e());
  86   Vector x2(pmv_->num_cols());
  87   x2.setZero();
  88
  89   for (int i = 0; i < pmv_->num_cols_e(); ++i) {
  90     x1(i) = x2(i) = RandDouble();
  91   }
  92
  93   Vector y1 = Vector::Zero(pmv_->num_rows());
  94   pmv_->RightMultiplyE(x1.data(), y1.data());
  95
  96   Vector y2 = Vector::Zero(pmv_->num_rows());
  97   A_->RightMultiply(x2.data(), y2.data());
  98
  99   for (int i = 0; i < pmv_->num_rows(); ++i) {
 100     EXPECT_NEAR(y1(i), y2(i), kEpsilon);
 101   }
 102 }
 103
 104 TEST_F(PartitionedMatrixViewTest, RightMultiplyF) {
 105   Vector x1(pmv_->num_cols_f());
 106   Vector x2 = Vector::Zero(pmv_->num_cols());
 107
 108   for (int i = 0; i < pmv_->num_cols_f(); ++i) {
 109     x1(i) = RandDouble();
 110     x2(i + pmv_->num_cols_e()) = x1(i);
 111   }
 112
 113   Vector y1 = Vector::Zero(pmv_->num_rows());
 114   pmv_->RightMultiplyF(x1.data(), y1.data());
 115
 116   Vector y2 = Vector::Zero(pmv_->num_rows());
 117   A_->RightMultiply(x2.data(), y2.data());
 118
 119   for (int i = 0; i < pmv_->num_rows(); ++i) {
 120     EXPECT_NEAR(y1(i), y2(i), kEpsilon);
 121   }
 122 }
 123
 124 TEST_F(PartitionedMatrixViewTest, LeftMultiply) {
 125   Vector x = Vector::Zero(pmv_->num_rows());
 126   for (int i = 0; i < pmv_->num_rows(); ++i) {
 127     x(i) = RandDouble();
 128   }
 129
 130   Vector y = Vector::Zero(pmv_->num_cols());
 131   Vector y1 = Vector::Zero(pmv_->num_cols_e());
 132   Vector y2 = Vector::Zero(pmv_->num_cols_f());
 133
 134   A_->LeftMultiply(x.data(), y.data());
 135   pmv_->LeftMultiplyE(x.data(), y1.data());
 136   pmv_->LeftMultiplyF(x.data(), y2.data());
 137
 138   for (int i = 0; i < pmv_->num_cols(); ++i) {
 139     EXPECT_NEAR(y(i),
 140                 (i < pmv_->num_cols_e()) ? y1(i) : y2(i - pmv_->num_cols_e()),
 141                 kEpsilon);
 142   }
 143 }
 144
 145 TEST_F(PartitionedMatrixViewTest, BlockDiagonalEtE) {
 146   scoped_ptr<BlockSparseMatrix>
 147       block_diagonal_ee(pmv_->CreateBlockDiagonalEtE());
 148   const CompressedRowBlockStructure* bs  = block_diagonal_ee->block_structure();
 149
 150   EXPECT_EQ(block_diagonal_ee->num_rows(), 2);
 151   EXPECT_EQ(block_diagonal_ee->num_cols(), 2);
 152   EXPECT_EQ(bs->cols.size(), 2);
 153   EXPECT_EQ(bs->rows.size(), 2);
 154
 155   EXPECT_NEAR(block_diagonal_ee->values()[0], 10.0, kEpsilon);
 156   EXPECT_NEAR(block_diagonal_ee->values()[1], 155.0, kEpsilon);
 157 }
 158
 159 TEST_F(PartitionedMatrixViewTest, BlockDiagonalFtF) {
 160   scoped_ptr<BlockSparseMatrix>
 161       block_diagonal_ff(pmv_->CreateBlockDiagonalFtF());
 162   const CompressedRowBlockStructure* bs  = block_diagonal_ff->block_structure();
 163
 164   EXPECT_EQ(block_diagonal_ff->num_rows(), 3);
 165   EXPECT_EQ(block_diagonal_ff->num_cols(), 3);
 166   EXPECT_EQ(bs->cols.size(), 3);
 167   EXPECT_EQ(bs->rows.size(), 3);
 168   EXPECT_NEAR(block_diagonal_ff->values()[0], 70.0, kEpsilon);
 169   EXPECT_NEAR(block_diagonal_ff->values()[1], 17.0, kEpsilon);
 170   EXPECT_NEAR(block_diagonal_ff->values()[2], 37.0, kEpsilon);
 171 }
 172
 173 }  // namespace internal
 174 }  // namespace ceres