jacobisvd.cpp
Go to the documentation of this file.
00001 // This file is part of Eigen, a lightweight C++ template library
00002 // for linear algebra.
00003 //
00004 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
00005 // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
00006 //
00007 // Eigen is free software; you can redistribute it and/or
00008 // modify it under the terms of the GNU Lesser General Public
00009 // License as published by the Free Software Foundation; either
00010 // version 3 of the License, or (at your option) any later version.
00011 //
00012 // Alternatively, you can redistribute it and/or
00013 // modify it under the terms of the GNU General Public License as
00014 // published by the Free Software Foundation; either version 2 of
00015 // the License, or (at your option) any later version.
00016 //
00017 // Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
00018 // WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
00019 // FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
00020 // GNU General Public License for more details.
00021 //
00022 // You should have received a copy of the GNU Lesser General Public
00023 // License and a copy of the GNU General Public License along with
00024 // Eigen. If not, see <http://www.gnu.org/licenses/>.
00025 
00026 // discard stack allocation as that too bypasses malloc
00027 #define EIGEN_STACK_ALLOCATION_LIMIT 0
00028 #define EIGEN_RUNTIME_NO_MALLOC
00029 #include "main.h"
00030 #include <Eigen/SVD>
00031 
00032 template<typename MatrixType, int QRPreconditioner>
00033 void jacobisvd_check_full(const MatrixType& m, const JacobiSVD<MatrixType, QRPreconditioner>& svd)
00034 {
00035   typedef typename MatrixType::Index Index;
00036   Index rows = m.rows();
00037   Index cols = m.cols();
00038 
00039   enum {
00040     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
00041     ColsAtCompileTime = MatrixType::ColsAtCompileTime
00042   };
00043 
00044   typedef typename MatrixType::Scalar Scalar;
00045   typedef typename NumTraits<Scalar>::Real RealScalar;
00046   typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime> MatrixUType;
00047   typedef Matrix<Scalar, ColsAtCompileTime, ColsAtCompileTime> MatrixVType;
00048   typedef Matrix<Scalar, RowsAtCompileTime, 1> ColVectorType;
00049   typedef Matrix<Scalar, ColsAtCompileTime, 1> InputVectorType;
00050 
00051   MatrixType sigma = MatrixType::Zero(rows,cols);
00052   sigma.diagonal() = svd.singularValues().template cast<Scalar>();
00053   MatrixUType u = svd.matrixU();
00054   MatrixVType v = svd.matrixV();
00055 
00056   VERIFY_IS_APPROX(m, u * sigma * v.adjoint());
00057   VERIFY_IS_UNITARY(u);
00058   VERIFY_IS_UNITARY(v);
00059 }
00060 
00061 template<typename MatrixType, int QRPreconditioner>
00062 void jacobisvd_compare_to_full(const MatrixType& m,
00063                                unsigned int computationOptions,
00064                                const JacobiSVD<MatrixType, QRPreconditioner>& referenceSvd)
00065 {
00066   typedef typename MatrixType::Index Index;
00067   Index rows = m.rows();
00068   Index cols = m.cols();
00069   Index diagSize = std::min(rows, cols);
00070 
00071   JacobiSVD<MatrixType, QRPreconditioner> svd(m, computationOptions);
00072 
00073   VERIFY_IS_APPROX(svd.singularValues(), referenceSvd.singularValues());
00074   if(computationOptions & ComputeFullU)
00075     VERIFY_IS_APPROX(svd.matrixU(), referenceSvd.matrixU());
00076   if(computationOptions & ComputeThinU)
00077     VERIFY_IS_APPROX(svd.matrixU(), referenceSvd.matrixU().leftCols(diagSize));
00078   if(computationOptions & ComputeFullV)
00079     VERIFY_IS_APPROX(svd.matrixV(), referenceSvd.matrixV());
00080   if(computationOptions & ComputeThinV)
00081     VERIFY_IS_APPROX(svd.matrixV(), referenceSvd.matrixV().leftCols(diagSize));
00082 }
00083 
00084 template<typename MatrixType, int QRPreconditioner>
00085 void jacobisvd_solve(const MatrixType& m, unsigned int computationOptions)
00086 {
00087   typedef typename MatrixType::Scalar Scalar;
00088   typedef typename MatrixType::Index Index;
00089   Index rows = m.rows();
00090   Index cols = m.cols();
00091 
00092   enum {
00093     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
00094     ColsAtCompileTime = MatrixType::ColsAtCompileTime
00095   };
00096 
00097   typedef Matrix<Scalar, RowsAtCompileTime, Dynamic> RhsType;
00098   typedef Matrix<Scalar, ColsAtCompileTime, Dynamic> SolutionType;
00099 
00100   RhsType rhs = RhsType::Random(rows, internal::random<Index>(1, cols));
00101   JacobiSVD<MatrixType, QRPreconditioner> svd(m, computationOptions);
00102   SolutionType x = svd.solve(rhs);
00103   // evaluate normal equation which works also for least-squares solutions
00104   VERIFY_IS_APPROX(m.adjoint()*m*x,m.adjoint()*rhs);
00105 }
00106 
00107 template<typename MatrixType, int QRPreconditioner>
00108 void jacobisvd_test_all_computation_options(const MatrixType& m)
00109 {
00110   if (QRPreconditioner == NoQRPreconditioner && m.rows() != m.cols())
00111     return;
00112   JacobiSVD<MatrixType, QRPreconditioner> fullSvd(m, ComputeFullU|ComputeFullV);
00113 
00114   jacobisvd_check_full(m, fullSvd);
00115   jacobisvd_solve<MatrixType, QRPreconditioner>(m, ComputeFullU | ComputeFullV);
00116 
00117   if(QRPreconditioner == FullPivHouseholderQRPreconditioner)
00118     return;
00119 
00120   jacobisvd_compare_to_full(m, ComputeFullU, fullSvd);
00121   jacobisvd_compare_to_full(m, ComputeFullV, fullSvd);
00122   jacobisvd_compare_to_full(m, 0, fullSvd);
00123 
00124   if (MatrixType::ColsAtCompileTime == Dynamic) {
00125     // thin U/V are only available with dynamic number of columns
00126     jacobisvd_compare_to_full(m, ComputeFullU|ComputeThinV, fullSvd);
00127     jacobisvd_compare_to_full(m,              ComputeThinV, fullSvd);
00128     jacobisvd_compare_to_full(m, ComputeThinU|ComputeFullV, fullSvd);
00129     jacobisvd_compare_to_full(m, ComputeThinU             , fullSvd);
00130     jacobisvd_compare_to_full(m, ComputeThinU|ComputeThinV, fullSvd);
00131     jacobisvd_solve<MatrixType, QRPreconditioner>(m, ComputeFullU | ComputeThinV);
00132     jacobisvd_solve<MatrixType, QRPreconditioner>(m, ComputeThinU | ComputeFullV);
00133     jacobisvd_solve<MatrixType, QRPreconditioner>(m, ComputeThinU | ComputeThinV);
00134   }
00135 }
00136 
00137 template<typename MatrixType>
00138 void jacobisvd(const MatrixType& a = MatrixType(), bool pickrandom = true)
00139 {
00140   MatrixType m = pickrandom ? MatrixType::Random(a.rows(), a.cols()) : a;
00141 
00142   jacobisvd_test_all_computation_options<MatrixType, FullPivHouseholderQRPreconditioner>(m);
00143   jacobisvd_test_all_computation_options<MatrixType, ColPivHouseholderQRPreconditioner>(m);
00144   jacobisvd_test_all_computation_options<MatrixType, HouseholderQRPreconditioner>(m);
00145   jacobisvd_test_all_computation_options<MatrixType, NoQRPreconditioner>(m);
00146 }
00147 
00148 template<typename MatrixType> void jacobisvd_verify_assert(const MatrixType& m)
00149 {
00150   typedef typename MatrixType::Scalar Scalar;
00151   typedef typename MatrixType::Index Index;
00152   Index rows = m.rows();
00153   Index cols = m.cols();
00154 
00155   enum {
00156     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
00157     ColsAtCompileTime = MatrixType::ColsAtCompileTime
00158   };
00159 
00160   typedef Matrix<Scalar, RowsAtCompileTime, 1> RhsType;
00161 
00162   RhsType rhs(rows);
00163 
00164   JacobiSVD<MatrixType> svd;
00165   VERIFY_RAISES_ASSERT(svd.matrixU())
00166   VERIFY_RAISES_ASSERT(svd.singularValues())
00167   VERIFY_RAISES_ASSERT(svd.matrixV())
00168   VERIFY_RAISES_ASSERT(svd.solve(rhs))
00169 
00170   MatrixType a = MatrixType::Zero(rows, cols);
00171   a.setZero();
00172   svd.compute(a, 0);
00173   VERIFY_RAISES_ASSERT(svd.matrixU())
00174   VERIFY_RAISES_ASSERT(svd.matrixV())
00175   svd.singularValues();
00176   VERIFY_RAISES_ASSERT(svd.solve(rhs))
00177 
00178   if (ColsAtCompileTime == Dynamic)
00179   {
00180     svd.compute(a, ComputeThinU);
00181     svd.matrixU();
00182     VERIFY_RAISES_ASSERT(svd.matrixV())
00183     VERIFY_RAISES_ASSERT(svd.solve(rhs))
00184 
00185     svd.compute(a, ComputeThinV);
00186     svd.matrixV();
00187     VERIFY_RAISES_ASSERT(svd.matrixU())
00188     VERIFY_RAISES_ASSERT(svd.solve(rhs))
00189 
00190     JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner> svd_fullqr;
00191     VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeFullU|ComputeThinV))
00192     VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeThinU|ComputeThinV))
00193     VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeThinU|ComputeFullV))
00194   }
00195   else
00196   {
00197     VERIFY_RAISES_ASSERT(svd.compute(a, ComputeThinU))
00198     VERIFY_RAISES_ASSERT(svd.compute(a, ComputeThinV))
00199   }
00200 }
00201 
00202 template<typename MatrixType>
00203 void jacobisvd_method()
00204 {
00205   enum { Size = MatrixType::RowsAtCompileTime };
00206   typedef typename MatrixType::RealScalar RealScalar;
00207   typedef Matrix<RealScalar, Size, 1> RealVecType;
00208   MatrixType m = MatrixType::Identity();
00209   VERIFY_IS_APPROX(m.jacobiSvd().singularValues(), RealVecType::Ones());
00210   VERIFY_RAISES_ASSERT(m.jacobiSvd().matrixU());
00211   VERIFY_RAISES_ASSERT(m.jacobiSvd().matrixV());
00212   VERIFY_IS_APPROX(m.jacobiSvd(ComputeFullU|ComputeFullV).solve(m), m);
00213 }
00214 
00215 // work around stupid msvc error when constructing at compile time an expression that involves
00216 // a division by zero, even if the numeric type has floating point
00217 template<typename Scalar>
00218 EIGEN_DONT_INLINE Scalar zero() { return Scalar(0); }
00219 
00220 // workaround aggressive optimization in ICC
00221 template<typename T> EIGEN_DONT_INLINE  T sub(T a, T b) { return a - b; }
00222 
00223 template<typename MatrixType>
00224 void jacobisvd_inf_nan()
00225 {
00226   // all this function does is verify we don't iterate infinitely on nan/inf values
00227 
00228   JacobiSVD<MatrixType> svd;
00229   typedef typename MatrixType::Scalar Scalar;
00230   Scalar some_inf = Scalar(1) / zero<Scalar>();
00231   VERIFY(sub(some_inf, some_inf) != sub(some_inf, some_inf));
00232   svd.compute(MatrixType::Constant(10,10,some_inf), ComputeFullU | ComputeFullV);
00233 
00234   Scalar some_nan = zero<Scalar>() / zero<Scalar>();
00235   VERIFY(some_nan != some_nan);
00236   svd.compute(MatrixType::Constant(10,10,some_nan), ComputeFullU | ComputeFullV);
00237 
00238   MatrixType m = MatrixType::Zero(10,10);
00239   m(internal::random<int>(0,9), internal::random<int>(0,9)) = some_inf;
00240   svd.compute(m, ComputeFullU | ComputeFullV);
00241 
00242   m = MatrixType::Zero(10,10);
00243   m(internal::random<int>(0,9), internal::random<int>(0,9)) = some_nan;
00244   svd.compute(m, ComputeFullU | ComputeFullV);
00245 }
00246 
00247 void jacobisvd_preallocate()
00248 {
00249   Vector3f v(3.f, 2.f, 1.f);
00250   MatrixXf m = v.asDiagonal();
00251 
00252   internal::set_is_malloc_allowed(false);
00253   VERIFY_RAISES_ASSERT(VectorXf v(10);)
00254   JacobiSVD<MatrixXf> svd;
00255   internal::set_is_malloc_allowed(true);
00256   svd.compute(m);
00257   VERIFY_IS_APPROX(svd.singularValues(), v);
00258 
00259   JacobiSVD<MatrixXf> svd2(3,3);
00260   internal::set_is_malloc_allowed(false);
00261   svd2.compute(m);
00262   internal::set_is_malloc_allowed(true);
00263   VERIFY_IS_APPROX(svd2.singularValues(), v);
00264   VERIFY_RAISES_ASSERT(svd2.matrixU());
00265   VERIFY_RAISES_ASSERT(svd2.matrixV());
00266   svd2.compute(m, ComputeFullU | ComputeFullV);
00267   VERIFY_IS_APPROX(svd2.matrixU(), Matrix3f::Identity());
00268   VERIFY_IS_APPROX(svd2.matrixV(), Matrix3f::Identity());
00269   internal::set_is_malloc_allowed(false);
00270   svd2.compute(m);
00271   internal::set_is_malloc_allowed(true);
00272 
00273   JacobiSVD<MatrixXf> svd3(3,3,ComputeFullU|ComputeFullV);
00274   internal::set_is_malloc_allowed(false);
00275   svd2.compute(m);
00276   internal::set_is_malloc_allowed(true);
00277   VERIFY_IS_APPROX(svd2.singularValues(), v);
00278   VERIFY_IS_APPROX(svd2.matrixU(), Matrix3f::Identity());
00279   VERIFY_IS_APPROX(svd2.matrixV(), Matrix3f::Identity());
00280   internal::set_is_malloc_allowed(false);
00281   svd2.compute(m, ComputeFullU|ComputeFullV);
00282   internal::set_is_malloc_allowed(true);
00283 
00284   
00285 }
00286 
00287 void test_jacobisvd()
00288 {
00289   CALL_SUBTEST_3(( jacobisvd_verify_assert(Matrix3f()) ));
00290   CALL_SUBTEST_4(( jacobisvd_verify_assert(Matrix4d()) ));
00291   CALL_SUBTEST_7(( jacobisvd_verify_assert(MatrixXf(10,12)) ));
00292   CALL_SUBTEST_8(( jacobisvd_verify_assert(MatrixXcd(7,5)) ));
00293 
00294   for(int i = 0; i < g_repeat; i++) {
00295     Matrix2cd m;
00296     m << 0, 1,
00297          0, 1;
00298     CALL_SUBTEST_1(( jacobisvd(m, false) ));
00299     m << 1, 0,
00300          1, 0;
00301     CALL_SUBTEST_1(( jacobisvd(m, false) ));
00302 
00303     Matrix2d n;
00304     n << 0, 0,
00305          0, 0;
00306     CALL_SUBTEST_2(( jacobisvd(n, false) ));
00307     n << 0, 0,
00308          0, 1;
00309     CALL_SUBTEST_2(( jacobisvd(n, false) ));
00310     
00311     CALL_SUBTEST_3(( jacobisvd<Matrix3f>() ));
00312     CALL_SUBTEST_4(( jacobisvd<Matrix4d>() ));
00313     CALL_SUBTEST_5(( jacobisvd<Matrix<float,3,5> >() ));
00314     CALL_SUBTEST_6(( jacobisvd<Matrix<double,Dynamic,2> >(Matrix<double,Dynamic,2>(10,2)) ));
00315 
00316     int r = internal::random<int>(1, 30),
00317         c = internal::random<int>(1, 30);
00318     CALL_SUBTEST_7(( jacobisvd<MatrixXf>(MatrixXf(r,c)) ));
00319     CALL_SUBTEST_8(( jacobisvd<MatrixXcd>(MatrixXcd(r,c)) ));
00320     (void) r;
00321     (void) c;
00322 
00323     // Test on inf/nan matrix
00324     CALL_SUBTEST_7( jacobisvd_inf_nan<MatrixXf>() );
00325   }
00326 
00327   CALL_SUBTEST_7(( jacobisvd<MatrixXf>(MatrixXf(internal::random<int>(100, 150), internal::random<int>(100, 150))) ));
00328   CALL_SUBTEST_8(( jacobisvd<MatrixXcd>(MatrixXcd(internal::random<int>(80, 100), internal::random<int>(80, 100))) ));
00329 
00330   // test matrixbase method
00331   CALL_SUBTEST_1(( jacobisvd_method<Matrix2cd>() ));
00332   CALL_SUBTEST_3(( jacobisvd_method<Matrix3f>() ));
00333 
00334   // Test problem size constructors
00335   CALL_SUBTEST_7( JacobiSVD<MatrixXf>(10,10) );
00336 
00337   // Check that preallocation avoids subsequent mallocs
00338   CALL_SUBTEST_9( jacobisvd_preallocate() );
00339 }


libicr
Author(s): Robert Krug
autogenerated on Mon Jan 6 2014 11:32:51