• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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@google.com (Sameer Agarwal)
30 
31 #include "ceres/compressed_row_sparse_matrix.h"
32 
33 #include <numeric>
34 #include "ceres/casts.h"
35 #include "ceres/crs_matrix.h"
36 #include "ceres/cxsparse.h"
37 #include "ceres/internal/eigen.h"
38 #include "ceres/internal/scoped_ptr.h"
39 #include "ceres/linear_least_squares_problems.h"
40 #include "ceres/random.h"
41 #include "ceres/triplet_sparse_matrix.h"
42 #include "glog/logging.h"
43 #include "gtest/gtest.h"
44 
45 namespace ceres {
46 namespace internal {
47 
CompareMatrices(const SparseMatrix * a,const SparseMatrix * b)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 CompressedRowSparseMatrixTest : public ::testing::Test {
70  protected :
SetUp()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     crsm.reset(new CompressedRowSparseMatrix(*tsm));
79 
80     num_rows = tsm->num_rows();
81     num_cols = tsm->num_cols();
82 
83     vector<int>* row_blocks = crsm->mutable_row_blocks();
84     row_blocks->resize(num_rows);
85     std::fill(row_blocks->begin(), row_blocks->end(), 1);
86 
87     vector<int>* col_blocks = crsm->mutable_col_blocks();
88     col_blocks->resize(num_cols);
89     std::fill(col_blocks->begin(), col_blocks->end(), 1);
90   }
91 
92   int num_rows;
93   int num_cols;
94 
95   scoped_ptr<TripletSparseMatrix> tsm;
96   scoped_ptr<CompressedRowSparseMatrix> crsm;
97 };
98 
TEST_F(CompressedRowSparseMatrixTest,RightMultiply)99 TEST_F(CompressedRowSparseMatrixTest, RightMultiply) {
100   CompareMatrices(tsm.get(), crsm.get());
101 }
102 
TEST_F(CompressedRowSparseMatrixTest,LeftMultiply)103 TEST_F(CompressedRowSparseMatrixTest, LeftMultiply) {
104   for (int i = 0; i < num_rows; ++i) {
105     Vector a = Vector::Zero(num_rows);
106     a(i) = 1.0;
107 
108     Vector b1 = Vector::Zero(num_cols);
109     Vector b2 = Vector::Zero(num_cols);
110 
111     tsm->LeftMultiply(a.data(), b1.data());
112     crsm->LeftMultiply(a.data(), b2.data());
113 
114     EXPECT_EQ((b1 - b2).norm(), 0);
115   }
116 }
117 
TEST_F(CompressedRowSparseMatrixTest,ColumnNorm)118 TEST_F(CompressedRowSparseMatrixTest, ColumnNorm) {
119   Vector b1 = Vector::Zero(num_cols);
120   Vector b2 = Vector::Zero(num_cols);
121 
122   tsm->SquaredColumnNorm(b1.data());
123   crsm->SquaredColumnNorm(b2.data());
124 
125   EXPECT_EQ((b1 - b2).norm(), 0);
126 }
127 
TEST_F(CompressedRowSparseMatrixTest,Scale)128 TEST_F(CompressedRowSparseMatrixTest, Scale) {
129   Vector scale(num_cols);
130   for (int i = 0; i < num_cols; ++i) {
131     scale(i) = i + 1;
132   }
133 
134   tsm->ScaleColumns(scale.data());
135   crsm->ScaleColumns(scale.data());
136   CompareMatrices(tsm.get(), crsm.get());
137 }
138 
TEST_F(CompressedRowSparseMatrixTest,DeleteRows)139 TEST_F(CompressedRowSparseMatrixTest, DeleteRows) {
140   // Clear the row and column blocks as these are purely scalar tests.
141   crsm->mutable_row_blocks()->clear();
142   crsm->mutable_col_blocks()->clear();
143   for (int i = 0; i < num_rows; ++i) {
144     tsm->Resize(num_rows - i, num_cols);
145     crsm->DeleteRows(crsm->num_rows() - tsm->num_rows());
146     CompareMatrices(tsm.get(), crsm.get());
147   }
148 }
149 
TEST_F(CompressedRowSparseMatrixTest,AppendRows)150 TEST_F(CompressedRowSparseMatrixTest, AppendRows) {
151   // Clear the row and column blocks as these are purely scalar tests.
152   crsm->mutable_row_blocks()->clear();
153   crsm->mutable_col_blocks()->clear();
154 
155   for (int i = 0; i < num_rows; ++i) {
156     TripletSparseMatrix tsm_appendage(*tsm);
157     tsm_appendage.Resize(i, num_cols);
158 
159     tsm->AppendRows(tsm_appendage);
160     CompressedRowSparseMatrix crsm_appendage(tsm_appendage);
161     crsm->AppendRows(crsm_appendage);
162 
163     CompareMatrices(tsm.get(), crsm.get());
164   }
165 }
166 
TEST_F(CompressedRowSparseMatrixTest,AppendAndDeleteBlockDiagonalMatrix)167 TEST_F(CompressedRowSparseMatrixTest, AppendAndDeleteBlockDiagonalMatrix) {
168   int num_diagonal_rows = crsm->num_cols();
169 
170   scoped_array<double> diagonal(new double[num_diagonal_rows]);
171   for (int i = 0; i < num_diagonal_rows; ++i) {
172     diagonal[i] =i;
173   }
174 
175   vector<int> row_and_column_blocks;
176   row_and_column_blocks.push_back(1);
177   row_and_column_blocks.push_back(2);
178   row_and_column_blocks.push_back(2);
179 
180   const vector<int> pre_row_blocks = crsm->row_blocks();
181   const vector<int> pre_col_blocks = crsm->col_blocks();
182 
183   scoped_ptr<CompressedRowSparseMatrix> appendage(
184       CompressedRowSparseMatrix::CreateBlockDiagonalMatrix(
185           diagonal.get(), row_and_column_blocks));
186   LOG(INFO) << appendage->row_blocks().size();
187 
188   crsm->AppendRows(*appendage);
189 
190   const vector<int> post_row_blocks = crsm->row_blocks();
191   const vector<int> post_col_blocks = crsm->col_blocks();
192 
193   vector<int> expected_row_blocks = pre_row_blocks;
194   expected_row_blocks.insert(expected_row_blocks.end(),
195                              row_and_column_blocks.begin(),
196                              row_and_column_blocks.end());
197 
198   vector<int> expected_col_blocks = pre_col_blocks;
199 
200   EXPECT_EQ(expected_row_blocks, crsm->row_blocks());
201   EXPECT_EQ(expected_col_blocks, crsm->col_blocks());
202 
203   crsm->DeleteRows(num_diagonal_rows);
204   EXPECT_EQ(crsm->row_blocks(), pre_row_blocks);
205   EXPECT_EQ(crsm->col_blocks(), pre_col_blocks);
206 }
207 
TEST_F(CompressedRowSparseMatrixTest,ToDenseMatrix)208 TEST_F(CompressedRowSparseMatrixTest, ToDenseMatrix) {
209   Matrix tsm_dense;
210   Matrix crsm_dense;
211 
212   tsm->ToDenseMatrix(&tsm_dense);
213   crsm->ToDenseMatrix(&crsm_dense);
214 
215   EXPECT_EQ((tsm_dense - crsm_dense).norm(), 0.0);
216 }
217 
TEST_F(CompressedRowSparseMatrixTest,ToCRSMatrix)218 TEST_F(CompressedRowSparseMatrixTest, ToCRSMatrix) {
219   CRSMatrix crs_matrix;
220   crsm->ToCRSMatrix(&crs_matrix);
221   EXPECT_EQ(crsm->num_rows(), crs_matrix.num_rows);
222   EXPECT_EQ(crsm->num_cols(), crs_matrix.num_cols);
223   EXPECT_EQ(crsm->num_rows() + 1, crs_matrix.rows.size());
224   EXPECT_EQ(crsm->num_nonzeros(), crs_matrix.cols.size());
225   EXPECT_EQ(crsm->num_nonzeros(), crs_matrix.values.size());
226 
227   for (int i = 0; i < crsm->num_rows() + 1; ++i) {
228     EXPECT_EQ(crsm->rows()[i], crs_matrix.rows[i]);
229   }
230 
231   for (int i = 0; i < crsm->num_nonzeros(); ++i) {
232     EXPECT_EQ(crsm->cols()[i], crs_matrix.cols[i]);
233     EXPECT_EQ(crsm->values()[i], crs_matrix.values[i]);
234   }
235 }
236 
TEST(CompressedRowSparseMatrix,CreateBlockDiagonalMatrix)237 TEST(CompressedRowSparseMatrix, CreateBlockDiagonalMatrix) {
238   vector<int> blocks;
239   blocks.push_back(1);
240   blocks.push_back(2);
241   blocks.push_back(2);
242 
243   Vector diagonal(5);
244   for (int i = 0; i < 5; ++i) {
245     diagonal(i) = i + 1;
246   }
247 
248   scoped_ptr<CompressedRowSparseMatrix> matrix(
249       CompressedRowSparseMatrix::CreateBlockDiagonalMatrix(
250           diagonal.data(), blocks));
251 
252   EXPECT_EQ(matrix->num_rows(), 5);
253   EXPECT_EQ(matrix->num_cols(), 5);
254   EXPECT_EQ(matrix->num_nonzeros(), 9);
255   EXPECT_EQ(blocks, matrix->row_blocks());
256   EXPECT_EQ(blocks, matrix->col_blocks());
257 
258   Vector x(5);
259   Vector y(5);
260 
261   x.setOnes();
262   y.setZero();
263   matrix->RightMultiply(x.data(), y.data());
264   for (int i = 0; i < diagonal.size(); ++i) {
265     EXPECT_EQ(y[i], diagonal[i]);
266   }
267 
268   y.setZero();
269   matrix->LeftMultiply(x.data(), y.data());
270   for (int i = 0; i < diagonal.size(); ++i) {
271     EXPECT_EQ(y[i], diagonal[i]);
272   }
273 
274   Matrix dense;
275   matrix->ToDenseMatrix(&dense);
276   EXPECT_EQ((dense.diagonal() - diagonal).norm(), 0.0);
277 }
278 
279 class SolveLowerTriangularTest : public ::testing::Test {
280  protected:
SetUp()281   void SetUp() {
282     matrix_.reset(new CompressedRowSparseMatrix(4, 4, 7));
283     int* rows = matrix_->mutable_rows();
284     int* cols = matrix_->mutable_cols();
285     double* values = matrix_->mutable_values();
286 
287     rows[0] = 0;
288     cols[0] = 0;
289     values[0] = 0.50754;
290 
291     rows[1] = 1;
292     cols[1] = 1;
293     values[1] = 0.80483;
294 
295     rows[2] = 2;
296     cols[2] = 1;
297     values[2] = 0.14120;
298     cols[3] = 2;
299     values[3] = 0.3;
300 
301     rows[3] = 4;
302     cols[4] = 0;
303     values[4] = 0.77696;
304     cols[5] = 1;
305     values[5] = 0.41860;
306     cols[6] = 3;
307     values[6] = 0.88979;
308 
309     rows[4] = 7;
310   }
311 
312   scoped_ptr<CompressedRowSparseMatrix> matrix_;
313 };
314 
TEST_F(SolveLowerTriangularTest,SolveInPlace)315 TEST_F(SolveLowerTriangularTest, SolveInPlace) {
316   double rhs_and_solution[] = {1.0, 1.0, 2.0, 2.0};
317   double expected[] = {1.970288,  1.242498,  6.081864, -0.057255};
318   matrix_->SolveLowerTriangularInPlace(rhs_and_solution);
319   for (int i = 0; i < 4; ++i) {
320     EXPECT_NEAR(rhs_and_solution[i], expected[i], 1e-4) << i;
321   }
322 }
323 
TEST_F(SolveLowerTriangularTest,TransposeSolveInPlace)324 TEST_F(SolveLowerTriangularTest, TransposeSolveInPlace) {
325   double rhs_and_solution[] = {1.0, 1.0, 2.0, 2.0};
326   const double expected[] = { -1.4706, -1.0962, 6.6667, 2.2477};
327 
328   matrix_->SolveLowerTriangularTransposeInPlace(rhs_and_solution);
329   for (int i = 0; i < 4; ++i) {
330     EXPECT_NEAR(rhs_and_solution[i], expected[i], 1e-4) << i;
331   }
332 }
333 
TEST(CompressedRowSparseMatrix,Transpose)334 TEST(CompressedRowSparseMatrix, Transpose) {
335   //  0  1  0  2  3  0
336   //  4  6  7  0  0  8
337   //  9 10  0 11 12  0
338   // 13  0 14 15  9  0
339   //  0 16 17  0  0  0
340 
341   // Block structure:
342   //  A  A  A  A  B  B
343   //  A  A  A  A  B  B
344   //  A  A  A  A  B  B
345   //  C  C  C  C  D  D
346   //  C  C  C  C  D  D
347   //  C  C  C  C  D  D
348 
349   CompressedRowSparseMatrix matrix(5, 6, 30);
350   int* rows = matrix.mutable_rows();
351   int* cols = matrix.mutable_cols();
352   double* values = matrix.mutable_values();
353   matrix.mutable_row_blocks()->push_back(3);
354   matrix.mutable_row_blocks()->push_back(3);
355   matrix.mutable_col_blocks()->push_back(4);
356   matrix.mutable_col_blocks()->push_back(2);
357 
358   rows[0] = 0;
359   cols[0] = 1;
360   cols[1] = 3;
361   cols[2] = 4;
362 
363   rows[1] = 3;
364   cols[3] = 0;
365   cols[4] = 1;
366   cols[5] = 2;
367   cols[6] = 5;
368 
369 
370   rows[2] = 7;
371   cols[7] = 0;
372   cols[8] = 1;
373   cols[9] = 3;
374   cols[10] = 4;
375 
376   rows[3] = 11;
377   cols[11] = 0;
378   cols[12] = 2;
379   cols[13] = 3;
380   cols[14] = 4;
381 
382   rows[4] = 15;
383   cols[15] = 1;
384   cols[16] = 2;
385   rows[5] = 17;
386 
387   copy(values, values + 17, cols);
388 
389   scoped_ptr<CompressedRowSparseMatrix> transpose(matrix.Transpose());
390 
391   ASSERT_EQ(transpose->row_blocks().size(), matrix.col_blocks().size());
392   for (int i = 0; i < transpose->row_blocks().size(); ++i) {
393     EXPECT_EQ(transpose->row_blocks()[i], matrix.col_blocks()[i]);
394   }
395 
396   ASSERT_EQ(transpose->col_blocks().size(), matrix.row_blocks().size());
397   for (int i = 0; i < transpose->col_blocks().size(); ++i) {
398     EXPECT_EQ(transpose->col_blocks()[i], matrix.row_blocks()[i]);
399   }
400 
401   Matrix dense_matrix;
402   matrix.ToDenseMatrix(&dense_matrix);
403 
404   Matrix dense_transpose;
405   transpose->ToDenseMatrix(&dense_transpose);
406   EXPECT_NEAR((dense_matrix - dense_transpose.transpose()).norm(), 0.0, 1e-14);
407 }
408 
409 #ifndef CERES_NO_CXSPARSE
410 
411 struct RandomMatrixOptions {
412   int num_row_blocks;
413   int min_row_block_size;
414   int max_row_block_size;
415   int num_col_blocks;
416   int min_col_block_size;
417   int max_col_block_size;
418   double block_density;
419 };
420 
CreateRandomCompressedRowSparseMatrix(const RandomMatrixOptions & options)421 CompressedRowSparseMatrix* CreateRandomCompressedRowSparseMatrix(
422     const RandomMatrixOptions& options) {
423   vector<int> row_blocks;
424   for (int i = 0; i < options.num_row_blocks; ++i) {
425     const int delta_block_size =
426         Uniform(options.max_row_block_size - options.min_row_block_size);
427     row_blocks.push_back(options.min_row_block_size + delta_block_size);
428   }
429 
430   vector<int> col_blocks;
431   for (int i = 0; i < options.num_col_blocks; ++i) {
432     const int delta_block_size =
433         Uniform(options.max_col_block_size - options.min_col_block_size);
434     col_blocks.push_back(options.min_col_block_size + delta_block_size);
435   }
436 
437   vector<int> rows;
438   vector<int> cols;
439   vector<double> values;
440 
441   while (values.size() == 0) {
442     int row_block_begin = 0;
443     for (int r = 0; r < options.num_row_blocks; ++r) {
444       int col_block_begin = 0;
445       for (int c = 0; c < options.num_col_blocks; ++c) {
446         if (RandDouble() <= options.block_density) {
447           for (int i = 0; i < row_blocks[r]; ++i) {
448             for (int j = 0; j < col_blocks[c]; ++j) {
449               rows.push_back(row_block_begin + i);
450               cols.push_back(col_block_begin + j);
451               values.push_back(RandNormal());
452             }
453           }
454         }
455         col_block_begin += col_blocks[c];
456       }
457       row_block_begin += row_blocks[r];
458     }
459   }
460 
461   const int num_rows = std::accumulate(row_blocks.begin(), row_blocks.end(), 0);
462   const int num_cols = std::accumulate(col_blocks.begin(), col_blocks.end(), 0);
463   const int num_nonzeros = values.size();
464 
465   TripletSparseMatrix tsm(num_rows, num_cols, num_nonzeros);
466   std::copy(rows.begin(), rows.end(), tsm.mutable_rows());
467   std::copy(cols.begin(), cols.end(), tsm.mutable_cols());
468   std::copy(values.begin(), values.end(), tsm.mutable_values());
469   tsm.set_num_nonzeros(num_nonzeros);
470   CompressedRowSparseMatrix* matrix = new CompressedRowSparseMatrix(tsm);
471   (*matrix->mutable_row_blocks())  = row_blocks;
472   (*matrix->mutable_col_blocks())  = col_blocks;
473   return matrix;
474 }
475 
ToDenseMatrix(const cs_di * matrix,Matrix * dense_matrix)476 void ToDenseMatrix(const cs_di* matrix, Matrix* dense_matrix) {
477   dense_matrix->resize(matrix->m, matrix->n);
478   dense_matrix->setZero();
479 
480   for (int c = 0; c < matrix->n; ++c) {
481    for (int idx = matrix->p[c]; idx < matrix->p[c + 1]; ++idx) {
482      const int r = matrix->i[idx];
483      (*dense_matrix)(r, c) = matrix->x[idx];
484    }
485  }
486 }
487 
TEST(CompressedRowSparseMatrix,ComputeOuterProduct)488 TEST(CompressedRowSparseMatrix, ComputeOuterProduct) {
489   // "Randomly generated seed."
490   SetRandomState(29823);
491   int kMaxNumRowBlocks = 10;
492   int kMaxNumColBlocks = 10;
493   int kNumTrials = 10;
494 
495   CXSparse cxsparse;
496   const double kTolerance = 1e-18;
497 
498   // Create a random matrix, compute its outer product using CXSParse
499   // and ComputeOuterProduct. Convert both matrices to dense matrices
500   // and compare their upper triangular parts. They should be within
501   // kTolerance of each other.
502   for (int num_row_blocks = 1;
503        num_row_blocks < kMaxNumRowBlocks;
504        ++num_row_blocks) {
505     for (int num_col_blocks = 1;
506          num_col_blocks < kMaxNumColBlocks;
507          ++num_col_blocks) {
508       for (int trial = 0; trial < kNumTrials; ++trial) {
509 
510 
511         RandomMatrixOptions options;
512         options.num_row_blocks = num_row_blocks;
513         options.num_col_blocks = num_col_blocks;
514         options.min_row_block_size = 1;
515         options.max_row_block_size = 5;
516         options.min_col_block_size = 1;
517         options.max_col_block_size = 10;
518         options.block_density = std::max(0.1, RandDouble());
519 
520         VLOG(2) << "num row blocks: " << options.num_row_blocks;
521         VLOG(2) << "num col blocks: " << options.num_col_blocks;
522         VLOG(2) << "min row block size: " << options.min_row_block_size;
523         VLOG(2) << "max row block size: " << options.max_row_block_size;
524         VLOG(2) << "min col block size: " << options.min_col_block_size;
525         VLOG(2) << "max col block size: " << options.max_col_block_size;
526         VLOG(2) << "block density: " << options.block_density;
527 
528         scoped_ptr<CompressedRowSparseMatrix> matrix(
529             CreateRandomCompressedRowSparseMatrix(options));
530 
531         cs_di cs_matrix_transpose = cxsparse.CreateSparseMatrixTransposeView(matrix.get());
532         cs_di* cs_matrix = cxsparse.TransposeMatrix(&cs_matrix_transpose);
533         cs_di* expected_outer_product =
534             cxsparse.MatrixMatrixMultiply(&cs_matrix_transpose, cs_matrix);
535 
536         vector<int> program;
537         scoped_ptr<CompressedRowSparseMatrix> outer_product(
538             CompressedRowSparseMatrix::CreateOuterProductMatrixAndProgram(
539                 *matrix, &program));
540         CompressedRowSparseMatrix::ComputeOuterProduct(*matrix,
541                                                        program,
542                                                        outer_product.get());
543 
544         cs_di actual_outer_product =
545             cxsparse.CreateSparseMatrixTransposeView(outer_product.get());
546 
547         ASSERT_EQ(actual_outer_product.m, actual_outer_product.n);
548         ASSERT_EQ(expected_outer_product->m, expected_outer_product->n);
549         ASSERT_EQ(actual_outer_product.m, expected_outer_product->m);
550 
551         Matrix actual_matrix;
552         Matrix expected_matrix;
553 
554         ToDenseMatrix(expected_outer_product, &expected_matrix);
555         expected_matrix.triangularView<Eigen::StrictlyLower>().setZero();
556 
557         ToDenseMatrix(&actual_outer_product, &actual_matrix);
558         const double diff_norm = (actual_matrix - expected_matrix).norm() / expected_matrix.norm();
559         ASSERT_NEAR(diff_norm, 0.0, kTolerance)
560             << "expected: \n"
561             << expected_matrix
562             << "\nactual: \n"
563             << actual_matrix;
564 
565         cxsparse.Free(cs_matrix);
566         cxsparse.Free(expected_outer_product);
567       }
568     }
569   }
570 }
571 
572 #endif  // CERES_NO_CXSPARSE
573 
574 }  // namespace internal
575 }  // namespace ceres
576