internal/ceres/dynamic_sparse_normal_cholesky_solver_test.cc

   1 // Ceres Solver - A fast non-linear least squares minimizer
   2 // Copyright 2017 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/casts.h"
  32 #include "ceres/compressed_row_sparse_matrix.h"
  33 #include "ceres/internal/scoped_ptr.h"
  34 #include "ceres/linear_least_squares_problems.h"
  35 #include "ceres/linear_solver.h"
  36 #include "ceres/triplet_sparse_matrix.h"
  37 #include "ceres/types.h"
  38 #include "glog/logging.h"
  39 #include "gtest/gtest.h"
  40
  41 #include "Eigen/Cholesky"
  42
  43 namespace ceres {
  44 namespace internal {
  45
  46 // TODO(sameeragarwal): These tests needs to be re-written to be more
  47 // thorough, they do not really test the dynamic nature of the
  48 // sparsity.
  49 class DynamicSparseNormalCholeskySolverTest : public ::testing::Test {
  50  protected:
  51   virtual void SetUp() {
  52     scoped_ptr<LinearLeastSquaresProblem> problem(
  53         CreateLinearLeastSquaresProblemFromId(1));
  54     A_.reset(CompressedRowSparseMatrix::FromTripletSparseMatrix(
  55         *down_cast<TripletSparseMatrix*>(problem->A.get())));
  56     b_.reset(problem->b.release());
  57     D_.reset(problem->D.release());
  58   }
  59
  60   void TestSolver(const LinearSolver::Options& options, double* D) {
  61     Matrix dense_A;
  62     A_->ToDenseMatrix(&dense_A);
  63     Matrix lhs = dense_A.transpose() * dense_A;
  64     if (D != NULL) {
  65       lhs += (ConstVectorRef(D, A_->num_cols()).array() *
  66               ConstVectorRef(D, A_->num_cols()).array())
  67                  .matrix()
  68                  .asDiagonal();
  69     }
  70
  71     Vector rhs(A_->num_cols());
  72     rhs.setZero();
  73     A_->LeftMultiply(b_.get(), rhs.data());
  74     Vector expected_solution = lhs.llt().solve(rhs);
  75
  76     scoped_ptr<LinearSolver> solver(LinearSolver::Create(options));
  77     LinearSolver::PerSolveOptions per_solve_options;
  78     per_solve_options.D = D;
  79     Vector actual_solution(A_->num_cols());
  80     LinearSolver::Summary summary;
  81     summary = solver->Solve(
  82         A_.get(), b_.get(), per_solve_options, actual_solution.data());
  83
  84     EXPECT_EQ(summary.termination_type, LINEAR_SOLVER_SUCCESS);
  85
  86     for (int i = 0; i < A_->num_cols(); ++i) {
  87       EXPECT_NEAR(expected_solution(i), actual_solution(i), 1e-8)
  88           << "\nExpected: " << expected_solution.transpose()
  89           << "\nActual: " << actual_solution.transpose();
  90     }
  91   }
  92
  93   void TestSolver(
  94       const SparseLinearAlgebraLibraryType sparse_linear_algebra_library_type) {
  95     LinearSolver::Options options;
  96     options.type = SPARSE_NORMAL_CHOLESKY;
  97     options.dynamic_sparsity = true;
  98     options.sparse_linear_algebra_library_type =
  99         sparse_linear_algebra_library_type;
 100     TestSolver(options, NULL);
 101     TestSolver(options, D_.get());
 102   }
 103
 104   scoped_ptr<CompressedRowSparseMatrix> A_;
 105   scoped_array<double> b_;
 106   scoped_array<double> D_;
 107 };
 108
 109 #ifndef CERES_NO_SUITESPARSE
 110 TEST_F(DynamicSparseNormalCholeskySolverTest, SuiteSparse) {
 111   TestSolver(SUITE_SPARSE);
 112 }
 113 #endif
 114
 115 #ifndef CERES_NO_CXSPARSE
 116 TEST_F(DynamicSparseNormalCholeskySolverTest, CXSparse) {
 117   TestSolver(CX_SPARSE);
 118 }
 119 #endif
 120
 121 #ifdef CERES_USE_EIGEN_SPARSE
 122 TEST_F(DynamicSparseNormalCholeskySolverTest, Eigen) {
 123   TestSolver(EIGEN_SPARSE);
 124 }
 125 #endif  // CERES_USE_EIGEN_SPARSE
 126
 127 }  // namespace internal
 128 }  // namespace ceres