matrix<Type> invertSparseMatrix(Eigen::SparseMatrix<Type> A){
matrix<Type> I(A.rows(),A.cols());
I.setIdentity();
Eigen::SimplicialLDLT< Eigen::SparseMatrix<Type> > ldlt(A);
matrix<Type> ans = ldlt.solve(I);
return ans;
}
MatrixXd MultivariateFNormalSufficient::compute_PW_direct() const
{
//Eigen::LLT<MatrixXd, Eigen::Upper> ldlt(get_ldlt());
Eigen::LDLT<MatrixXd, Eigen::Upper> ldlt(get_ldlt());
MatrixXd tmp(ldlt.solve(get_W()));
return tmp;
}
double GaussianProcessInterpolationRestraint::get_logdet_hessian() const {
// compute ldlt
IMP_Eigen::LDLT<IMP_Eigen::MatrixXd, IMP_Eigen::Upper> ldlt(get_hessian());
if (!ldlt.isPositive())
IMP_THROW("Hessian matrix is not positive definite!", ModelException);
return ldlt.vectorD().array().abs().log().sum();
}
std::vector<double> MultivariateFNormalSufficient::get_norms() const
{
if (!flag_norms_)
{
//Eigen::LLT<MatrixXd, Eigen::Upper> ldlt(get_ldlt());
Eigen::LDLT<MatrixXd, Eigen::Upper> ldlt(get_ldlt());
// determinant and derived constants
//MatrixXd L(ldlt.matrixU());
//std::cout << "L: " << L << std::endl << std::endl;
//std::cout << "D: " << ldlt.vectorD() << std::endl << std::endl;
//double logDetSigma=2*L.diagonal().array().log().sum() ;
double logDetSigma = ldlt.vectorD().array().abs().log().sum();
LOG( "MVN: det(Sigma) = " <<exp(logDetSigma) << std::endl);
double norm=pow(2*PI*SQUARE(factor_), -double(N_*M_)/2.0)
* exp(-double(N_)/2.0*logDetSigma);
double lnorm=double(N_*M_)/2 * log(2*PI*SQUARE(factor_))
+ double(N_)/2 * logDetSigma;
LOG( "MVN: norm = " << norm << " lnorm = "
<< lnorm << std::endl);
const_cast<MultivariateFNormalSufficient *>(this)
->set_norms(norm,lnorm);
}
std::vector<double> norms;
norms.push_back(norm_);
norms.push_back(lnorm_);
return norms;
}
Vector2d leastSquareSolverCalib::pointAlignerLoop( Vector2d &x, MatrixXd &Z, int iterations, Vector2d &result)
{
result=x;
MatrixXd H(2,2);
H.setZero();
MatrixXd b(2,1);
b.setZero();
Vector2d X;
std::cout << Z.transpose()<< std::endl;
for(int i = 0; i < iterations; i++){
std::cout<<"iteration "<<i <<std::endl;
X=x;
for(int j=0;j<Z.cols();j++){
Matrix2d J = computeJacobian(j,Z);
Vector2d e=computeError(j,Z);
std:: cout << "error: "<< e<< std::endl<<std::endl;
H+=J.transpose()*J;
b+=J.transpose()*e;
}
}
Vector2d dx;
std::cout << "H: "<<std::endl<<H<<std::endl<<std::endl<<std::endl;
std::cout << "b: "<<std::endl<<b<<std::endl;
LDLT<MatrixXd> ldlt(-H);
dx=ldlt.solve(b); // using a LDLT factorizationldlt;
return dx;
}
IMP_Eigen::MatrixXd
GaussianProcessInterpolation::get_posterior_covariance_matrix(FloatsList x)
const {
unsigned N(x.size());
IMP_Eigen::MatrixXd Wpri(M_, N);
for (unsigned i = 0; i < N; i++) Wpri.col(i) = get_wx_vector(x[i]);
IMP_Eigen::LDLT<IMP_Eigen::MatrixXd, IMP_Eigen::Upper> ldlt(get_ldlt());
// we can now use covariance_function_ because it is up to date
IMP_Eigen::MatrixXd Wpost((*covariance_function_)(x));
return Wpost - Wpri.transpose() * ldlt.solve(Wpri);
}
inline
Eigen::Matrix<T,Eigen::Dynamic,Eigen::Dynamic>
inverse_spd(const Eigen::Matrix<T,Eigen::Dynamic,Eigen::Dynamic>& m) {
validate_square(m,"inverse_spd");
validate_symmetric(m,"inverse_spd");
Eigen::LDLT< Eigen::Matrix<T,Eigen::Dynamic,Eigen::Dynamic> > ldlt(0.5*(m+m.transpose()));
if (ldlt.info() != Eigen::Success)
throw std::domain_error("Error in inverse_spd, LDLT factorization failed");
if (!ldlt.isPositive() || (ldlt.vectorD().array() <= 0).any())
throw std::domain_error("Error in inverse_spd, matrix not positive definite");
return ldlt.solve(Eigen::Matrix<T,Eigen::Dynamic,Eigen::Dynamic>::Identity(m.rows(),m.cols()));
}
MatrixType CovarianceEstimator::Evaluate( const VectorType& input )
{
_psdPort.SetOutput( input );
_psd.Invalidate();
_psdPort.Foreprop();
MatrixType out = _psd.GetOutput();
if( !_covarianceOutput )
{
Eigen::LDLT<MatrixType> ldlt( out );
out = ldlt.solve( MatrixType::Identity( out.rows(), out.cols() ) );
}
return out;
}
MatrixXd MultivariateFNormalSufficient::get_P() const
{
if (!flag_P_)
{
//inverse
//Eigen::LLT<MatrixXd, Eigen::Upper> ldlt(get_ldlt());
Eigen::LDLT<MatrixXd, Eigen::Upper> ldlt(get_ldlt());
LOG( "MVN: solving for inverse" << std::endl);
const_cast<MultivariateFNormalSufficient *>(this)
->set_P(ldlt.solve(MatrixXd::Identity(M_,M_)));
}
return P_;
}
bool gcta::comput_inverse_logdet_LDLT(eigenMatrix &Vi, eigenVector &prev_varcmp, double &logdet)
{
int i=0, n=Vi.cols();
LDLT<eigenMatrix> ldlt(Vi);
eigenVector d=ldlt.vectorD();
if(d.minCoeff()<0){
if(prev_varcmp.minCoeff()>0) return false;
else throw("Error: the matrix V becomes negative-definite because of one of the variance component is negative.\nPlease re-run the analysis without the --reml-no-constrain option.");
}
logdet=0.0;
for(i=0; i<n; i++) logdet+=log(d[i]);
Vi.setIdentity();
ldlt.solveInPlace(Vi);
return true;
}
//Eigen::LLT<MatrixXd, Eigen::Upper>
Eigen::LDLT<MatrixXd, Eigen::Upper>
MultivariateFNormalSufficient::get_ldlt() const
{
if (!flag_ldlt_)
{
LOG( "MVN: computing Cholesky decomposition" << std::endl);
// compute Cholesky decomposition for determinant and inverse
//Eigen::LLT<MatrixXd, Eigen::Upper> ldlt(get_Sigma());
Eigen::LDLT<MatrixXd, Eigen::Upper> ldlt(get_Sigma());
//if (ldlt.info() != Eigen::Success)
CHECK(ldlt.isPositive(),
"Sigma matrix is not positive semidefinite!");
const_cast<MultivariateFNormalSufficient *>(this)->set_ldlt(ldlt);
}
return ldlt_;
}
matrix<Type> discrLyap(matrix<Type> A_){
matrix<Type> I_(A_.rows(),A_.cols());
I_.setIdentity();
/* Sparse representations */
Eigen::SparseMatrix<Type> A=asSparseMatrix(A_);
Eigen::SparseMatrix<Type> I=asSparseMatrix(I_);
/* Kronecker */
Eigen::SparseMatrix<Type> AxA=kronecker(A,A);
Eigen::SparseMatrix<Type> IxI=kronecker(I,I);
matrix<Type> vecI=I_.vec().matrix();
Eigen::SparseMatrix<Type> C=IxI-AxA;
/* Solve system C*vecV=vecI.
Note: No assumptions on symmetry or eigenvalue-range of C. Therefore
rewrite it as (C'*C)*vecV=(C*vecI). Since (C'*C) is symmetric, the
LDL'-factorization can be applied.
*/
Eigen::SimplicialLDLT< Eigen::SparseMatrix<Type> > ldlt(C.transpose()*C);
matrix<Type> z = ldlt.solve(C.transpose()*vecI);
matrix<Type> V(A_.rows(),A_.cols());
for(int i=0;i<z.size();i++)V(i)=z(i);
return V;
}
//Eigen::LLT<MatrixXd, Eigen::Upper>
Eigen::LDLT<MatrixXd, Eigen::Upper>
MultivariateFNormalSufficient::get_ldlt() const
{
if (!flag_ldlt_)
{
timer_.start(CHOLESKY);
IMP_LOG(TERSE, "MVN: computing Cholesky decomposition" << std::endl);
// compute Cholesky decomposition for determinant and inverse
//Eigen::LLT<MatrixXd, Eigen::Upper> ldlt(get_Sigma());
Eigen::LDLT<MatrixXd, Eigen::Upper> ldlt(get_Sigma());
//if (ldlt.info() != Eigen::Success)
if (!ldlt.isPositive())
{
std::cout << "Sigma" << std::endl;
std::cout << get_Sigma() << std::endl;
IMP_THROW("Sigma matrix is not positive semidefinite!",
ModelException);
}
const_cast<MultivariateFNormalSufficient *>(this)->set_ldlt(ldlt);
timer_.stop(CHOLESKY);
}
return ldlt_;
}
template<typename Scalar> void sparse_solvers(int rows, int cols)
{
double density = std::max(8./(rows*cols), 0.01);
typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
typedef Matrix<Scalar,Dynamic,1> DenseVector;
// Scalar eps = 1e-6;
DenseVector vec1 = DenseVector::Random(rows);
std::vector<Vector2i> zeroCoords;
std::vector<Vector2i> nonzeroCoords;
// test triangular solver
{
DenseVector vec2 = vec1, vec3 = vec1;
SparseMatrix<Scalar> m2(rows, cols);
DenseMatrix refMat2 = DenseMatrix::Zero(rows, cols);
// lower
initSparse<Scalar>(density, refMat2, m2, ForceNonZeroDiag|MakeLowerTriangular, &zeroCoords, &nonzeroCoords);
VERIFY_IS_APPROX(refMat2.template marked<LowerTriangular>().solveTriangular(vec2),
m2.template marked<LowerTriangular>().solveTriangular(vec3));
// lower - transpose
initSparse<Scalar>(density, refMat2, m2, ForceNonZeroDiag|MakeLowerTriangular, &zeroCoords, &nonzeroCoords);
VERIFY_IS_APPROX(refMat2.template marked<LowerTriangular>().transpose().solveTriangular(vec2),
m2.template marked<LowerTriangular>().transpose().solveTriangular(vec3));
// upper
initSparse<Scalar>(density, refMat2, m2, ForceNonZeroDiag|MakeUpperTriangular, &zeroCoords, &nonzeroCoords);
VERIFY_IS_APPROX(refMat2.template marked<UpperTriangular>().solveTriangular(vec2),
m2.template marked<UpperTriangular>().solveTriangular(vec3));
// upper - transpose
initSparse<Scalar>(density, refMat2, m2, ForceNonZeroDiag|MakeUpperTriangular, &zeroCoords, &nonzeroCoords);
VERIFY_IS_APPROX(refMat2.template marked<UpperTriangular>().transpose().solveTriangular(vec2),
m2.template marked<UpperTriangular>().transpose().solveTriangular(vec3));
}
// test LLT
{
// TODO fix the issue with complex (see SparseLLT::solveInPlace)
SparseMatrix<Scalar> m2(rows, cols);
DenseMatrix refMat2(rows, cols);
DenseVector b = DenseVector::Random(cols);
DenseVector refX(cols), x(cols);
initSPD(density, refMat2, m2);
refMat2.llt().solve(b, &refX);
typedef SparseMatrix<Scalar,LowerTriangular|SelfAdjoint> SparseSelfAdjointMatrix;
if (!NumTraits<Scalar>::IsComplex)
{
x = b;
SparseLLT<SparseSelfAdjointMatrix> (m2).solveInPlace(x);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LLT: default");
}
#ifdef EIGEN_CHOLMOD_SUPPORT
x = b;
SparseLLT<SparseSelfAdjointMatrix,Cholmod>(m2).solveInPlace(x);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LLT: cholmod");
#endif
if (!NumTraits<Scalar>::IsComplex)
{
#ifdef EIGEN_TAUCS_SUPPORT
x = b;
SparseLLT<SparseSelfAdjointMatrix,Taucs>(m2,IncompleteFactorization).solveInPlace(x);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LLT: taucs (IncompleteFactorization)");
x = b;
SparseLLT<SparseSelfAdjointMatrix,Taucs>(m2,SupernodalMultifrontal).solveInPlace(x);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LLT: taucs (SupernodalMultifrontal)");
x = b;
SparseLLT<SparseSelfAdjointMatrix,Taucs>(m2,SupernodalLeftLooking).solveInPlace(x);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LLT: taucs (SupernodalLeftLooking)");
#endif
}
}
// test LDLT
if (!NumTraits<Scalar>::IsComplex)
{
// TODO fix the issue with complex (see SparseLDLT::solveInPlace)
SparseMatrix<Scalar> m2(rows, cols);
DenseMatrix refMat2(rows, cols);
DenseVector b = DenseVector::Random(cols);
DenseVector refX(cols), x(cols);
//initSPD(density, refMat2, m2);
initSparse<Scalar>(density, refMat2, m2, ForceNonZeroDiag|MakeUpperTriangular, 0, 0);
refMat2 += refMat2.adjoint();
refMat2.diagonal() *= 0.5;
refMat2.ldlt().solve(b, &refX);
typedef SparseMatrix<Scalar,UpperTriangular|SelfAdjoint> SparseSelfAdjointMatrix;
x = b;
SparseLDLT<SparseSelfAdjointMatrix> ldlt(m2);
if (ldlt.succeeded())
ldlt.solveInPlace(x);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LDLT: default");
}
// test LU
{
static int count = 0;
SparseMatrix<Scalar> m2(rows, cols);
DenseMatrix refMat2(rows, cols);
DenseVector b = DenseVector::Random(cols);
DenseVector refX(cols), x(cols);
initSparse<Scalar>(density, refMat2, m2, ForceNonZeroDiag, &zeroCoords, &nonzeroCoords);
LU<DenseMatrix> refLu(refMat2);
refLu.solve(b, &refX);
#if defined(EIGEN_SUPERLU_SUPPORT) || defined(EIGEN_UMFPACK_SUPPORT)
Scalar refDet = refLu.determinant();
#endif
x.setZero();
// // SparseLU<SparseMatrix<Scalar> > (m2).solve(b,&x);
// // VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LU: default");
#ifdef EIGEN_SUPERLU_SUPPORT
{
x.setZero();
SparseLU<SparseMatrix<Scalar>,SuperLU> slu(m2);
if (slu.succeeded())
{
if (slu.solve(b,&x)) {
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LU: SuperLU");
}
// std::cerr << refDet << " == " << slu.determinant() << "\n";
if (count==0) {
VERIFY_IS_APPROX(refDet,slu.determinant()); // FIXME det is not very stable for complex
}
}
}
#endif
#ifdef EIGEN_UMFPACK_SUPPORT
{
// check solve
x.setZero();
SparseLU<SparseMatrix<Scalar>,UmfPack> slu(m2);
if (slu.succeeded()) {
if (slu.solve(b,&x)) {
if (count==0) {
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LU: umfpack"); // FIXME solve is not very stable for complex
}
}
VERIFY_IS_APPROX(refDet,slu.determinant());
// TODO check the extracted data
//std::cerr << slu.matrixL() << "\n";
}
}
#endif
count++;
}
}
void GaussianProcessInterpolation::compute_Omi() {
// get inverse
IMP_LOG_TERSE(" compute_Omi: inverse" << std::endl);
IMP_Eigen::LDLT<IMP_Eigen::MatrixXd, IMP_Eigen::Upper> ldlt(get_ldlt());
Omi_ = ldlt.solve(IMP_Eigen::MatrixXd::Identity(M_, M_));
}
IMP_Eigen::MatrixXd GaussianProcessInterpolation::get_dcov_dOm(Floats q) const {
IMP_Eigen::VectorXd wq(get_wx_vector(q));
IMP_Eigen::LDLT<IMP_Eigen::MatrixXd, IMP_Eigen::Upper> ldlt(get_ldlt());
IMP_Eigen::VectorXd ret(ldlt.solve(wq));
return ret * ret.transpose();
}
template<typename Scalar> void sparse_ldlt(int rows, int cols)
{
static bool odd = true;
odd = !odd;
double density = (std::max)(8./(rows*cols), 0.01);
typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
typedef Matrix<Scalar,Dynamic,1> DenseVector;
SparseMatrix<Scalar> m2(rows, cols);
DenseMatrix refMat2(rows, cols);
DenseVector b = DenseVector::Random(cols);
DenseVector refX(cols), x(cols);
initSparse<Scalar>(density, refMat2, m2, ForceNonZeroDiag|MakeUpperTriangular, 0, 0);
SparseMatrix<Scalar> m3 = m2 * m2.adjoint(), m3_lo(rows,rows), m3_up(rows,rows);
DenseMatrix refMat3 = refMat2 * refMat2.adjoint();
refX = refMat3.template selfadjointView<Upper>().ldlt().solve(b);
typedef SparseMatrix<Scalar,Upper|SelfAdjoint> SparseSelfAdjointMatrix;
x = b;
SparseLDLT<SparseSelfAdjointMatrix> ldlt(m3);
if (ldlt.succeeded())
ldlt.solveInPlace(x);
else
std::cerr << "warning LDLT failed\n";
VERIFY_IS_APPROX(refMat3.template selfadjointView<Upper>() * x, b);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LDLT: default");
#ifdef EIGEN_CHOLMOD_SUPPORT
{
x = b;
SparseLDLT<SparseSelfAdjointMatrix, Cholmod> ldlt2(m3);
if (ldlt2.succeeded())
{
ldlt2.solveInPlace(x);
VERIFY_IS_APPROX(refMat3.template selfadjointView<Upper>() * x, b);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LDLT: cholmod solveInPlace");
x = ldlt2.solve(b);
VERIFY_IS_APPROX(refMat3.template selfadjointView<Upper>() * x, b);
VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LDLT: cholmod solve");
}
else
std::cerr << "warning LDLT failed\n";
}
#endif
// new Simplicial LLT
// new API
{
SparseMatrix<Scalar> m2(rows, cols);
DenseMatrix refMat2(rows, cols);
DenseVector b = DenseVector::Random(cols);
DenseVector ref_x(cols), x(cols);
DenseMatrix B = DenseMatrix::Random(rows,cols);
DenseMatrix ref_X(rows,cols), X(rows,cols);
initSparse<Scalar>(density, refMat2, m2, ForceNonZeroDiag|MakeLowerTriangular, 0, 0);
for(int i=0; i<rows; ++i)
m2.coeffRef(i,i) = refMat2(i,i) = internal::abs(internal::real(refMat2(i,i)));
SparseMatrix<Scalar> m3 = m2 * m2.adjoint(), m3_lo(rows,rows), m3_up(rows,rows);
DenseMatrix refMat3 = refMat2 * refMat2.adjoint();
m3_lo.template selfadjointView<Lower>().rankUpdate(m2,0);
m3_up.template selfadjointView<Upper>().rankUpdate(m2,0);
// with a single vector as the rhs
ref_x = refMat3.template selfadjointView<Lower>().llt().solve(b);
x = SimplicialCholesky<SparseMatrix<Scalar>, Lower>().setMode(odd ? SimplicialCholeskyLLt : SimplicialCholeskyLDLt).compute(m3).solve(b);
VERIFY(ref_x.isApprox(x,test_precision<Scalar>()) && "SimplicialCholesky: solve, full storage, lower, single dense rhs");
x = SimplicialCholesky<SparseMatrix<Scalar>, Upper>().setMode(odd ? SimplicialCholeskyLLt : SimplicialCholeskyLDLt).compute(m3).solve(b);
VERIFY(ref_x.isApprox(x,test_precision<Scalar>()) && "SimplicialCholesky: solve, full storage, upper, single dense rhs");
x = SimplicialCholesky<SparseMatrix<Scalar>, Lower>(m3_lo).solve(b);
VERIFY(ref_x.isApprox(x,test_precision<Scalar>()) && "SimplicialCholesky: solve, lower only, single dense rhs");
x = SimplicialCholesky<SparseMatrix<Scalar>, Upper>(m3_up).solve(b);
VERIFY(ref_x.isApprox(x,test_precision<Scalar>()) && "SimplicialCholesky: solve, upper only, single dense rhs");
// with multiple rhs
ref_X = refMat3.template selfadjointView<Lower>().llt().solve(B);
X = SimplicialCholesky<SparseMatrix<Scalar>, Lower>().setMode(odd ? SimplicialCholeskyLLt : SimplicialCholeskyLDLt).compute(m3).solve(B);
VERIFY(ref_X.isApprox(X,test_precision<Scalar>()) && "SimplicialCholesky: solve, full storage, lower, multiple dense rhs");
X = SimplicialCholesky<SparseMatrix<Scalar>, Upper>().setMode(odd ? SimplicialCholeskyLLt : SimplicialCholeskyLDLt).compute(m3).solve(B);
VERIFY(ref_X.isApprox(X,test_precision<Scalar>()) && "SimplicialCholesky: solve, full storage, upper, multiple dense rhs");
// with a sparse rhs
// SparseMatrix<Scalar> spB(rows,cols), spX(rows,cols);
// B.diagonal().array() += 1;
// spB = B.sparseView(0.5,1);
//
// ref_X = refMat3.template selfadjointView<Lower>().llt().solve(DenseMatrix(spB));
//
// spX = SimplicialCholesky<SparseMatrix<Scalar>, Lower>(m3).solve(spB);
// VERIFY(ref_X.isApprox(spX.toDense(),test_precision<Scalar>()) && "LLT: cholmod solve, multiple sparse rhs");
//
// spX = SimplicialCholesky<SparseMatrix<Scalar>, Upper>(m3).solve(spB);
// VERIFY(ref_X.isApprox(spX.toDense(),test_precision<Scalar>()) && "LLT: cholmod solve, multiple sparse rhs");
}
// for(int i=0; i<rows; ++i)
// m2.coeffRef(i,i) = refMat2(i,i) = internal::abs(internal::real(refMat2(i,i)));
//
// refX = refMat2.template selfadjointView<Upper>().ldlt().solve(b);
// typedef SparseMatrix<Scalar,Upper|SelfAdjoint> SparseSelfAdjointMatrix;
// x = b;
// SparseLDLT<SparseSelfAdjointMatrix> ldlt(m2);
// if (ldlt.succeeded())
// ldlt.solveInPlace(x);
// else
// std::cerr << "warning LDLT failed\n";
//
// VERIFY_IS_APPROX(refMat2.template selfadjointView<Upper>() * x, b);
// VERIFY(refX.isApprox(x,test_precision<Scalar>()) && "LDLT: default");
}
