/**
* Print out Chain after each iteration
*/
void MCMCObjective::DoExecute() {
if (!mcmc_)
LOG_CODE_ERROR() << "if (!mcmc_)";
if (first_write_ && !model_->global_configuration().resume()) {
cache_ << "starting_covariance_matrix {m}\n";
auto covariance = mcmc_->covariance_matrix();
for (unsigned i = 0; i < covariance.size1(); ++i) {
for (unsigned j = 0; j < covariance.size2() - 1; ++j)
cache_ << covariance(i,j) << " ";
cache_ << covariance(i, covariance.size2() - 1) << "\n";
}
cache_ << "samples {d} \n";
cache_ << "sample objective_score prior likelihood penalties additional_priors jacobians step_size acceptance_rate acceptance_rate_since_adapt\n";
}
auto chain = mcmc_->chain();
unsigned element = chain.size() - 1;
cache_ << chain[element].iteration_ << " "
<< chain[element].score_ << " "
<< chain[element].prior_ << " "
<< chain[element].likelihood_ << " "
<< chain[element].penalty_ << " "
<< chain[element].additional_priors_ << " "
<< chain[element].jacobians_ << " "
<< chain[element].step_size_ << " "
<< chain[element].acceptance_rate_ << " "
<< chain[element].acceptance_rate_since_adapt_ << "\n";
ready_for_writing_ = true;
}
void CovarianceMatrix::DoExecute() {
/*
* This reports the Covariance, Correlation and Hessian matrix
*/
LOG_TRACE();
auto minimiser_ = model_->managers().minimiser()->active_minimiser();
covariance_matrix_ = minimiser_->covariance_matrix();
cache_ << "*" << label_ << " " << "(" << type_ << ")" << "\n";
cache_ << "Covariance_Matrix " << REPORT_R_MATRIX << "\n";
for (unsigned i = 0; i < covariance_matrix_.size1(); ++i) {
for (unsigned j = 0; j < covariance_matrix_.size2(); ++j)
cache_ << covariance_matrix_(i, j) << " ";
cache_ << "\n";
}
if (model_->run_mode() == RunMode::kMCMC) {
auto mcmc_ = model_->managers().mcmc()->active_mcmc();
if (mcmc_->recalculate_covariance()) {
cache_ << REPORT_END << "\n\n";
LOG_FINE() << "During the mcmc run you recalculated the the covariance matrix, so we will print the modified matrix at the end of the chain";
cache_ << "Modified_covariance_matrix " << REPORT_R_MATRIX << "\n";
auto covariance = mcmc_->covariance_matrix();
for (unsigned i = 0; i < covariance.size1(); ++i) {
for (unsigned j = 0; j < covariance.size2() - 1; ++j)
cache_ << covariance(i, j) << " ";
cache_ << covariance(i, covariance.size2() - 1) << "\n";
}
}
}
ready_for_writing_ = true;
}
/*
void InitProc()
{
static tbb::task_scheduler_init init;
progress_mutex = new mutex();
}
void UnInitProc()
{
delete progress_mutex;
}
class tbb_ResampleTo256
{
VData *data,*res;
float sc;
public:
tbb_ResampleTo256(VData* _data,VData* _res) : data(_data),res(_res){ sc = data->GetSize().z/256.0f;};
public:
void operator()(const blocked_range <size_t>& r) const
{
ivec3 size = data->GetSize();
for (size_t i=r.begin(); i!=r.end(); i++)
{
for(int j=0;j<256;j++)
for(int k=0;k<256;k++)
{
float sum=0,ww=0;
int k1=int(k*sc)-1,k2=int((k+1)*sc)+1;
if(k1<0)k1=0;
if(k2>=size.z)k2=size.z-1;
for(int kk=k1;kk<=k2;kk++)
for(int ii=0;ii<2;ii++)
for(int jj=0;jj<2;jj++)
{
float www = 1.0/(abs(k*sc-kk)+1);
sum += data->GetValue(i*2+ii,j*2+jj,kk)*www;
ww += www;
}
//res[i+256*(j+256*k)] = short(sum/ww);
res->SetValue(short(sum/ww),i,j,k);
}
}
}
};
VData ResampleTo256(VData& data)
{
InitProc();
VData data_small(ivec3(256));
ivec3 size=data.GetSize();
if(!(size.x==512 && size.y==512))throw "123123";
{
for(int i=0;i<4;i++)
{
parallel_for(blocked_range<size_t>(i*64, (i+1)*64, 16), tbb_ResampleTo256(&data,&data_small) );
Progress::inst->SetPercent((i+1)*0.25);
}
}
UnInitProc();
return data_small;
}*/
VData GetMipData(VData& data,int dv)
{
ivec3 size = data.GetSize();
ivec3 size2(size.x/dv,size.y/dv,size.z/dv);
VData data_small(size2,data.GetValueFormat());
data_small.spacing = data.spacing*dv;
for(int i=0;i<size2.x;i++)
{
for(int j=0;j<size2.y;j++)
for(int k=0;k<size2.z;k++)
{
int sum = 0;
for(int ii=0;ii<dv;ii++)
for(int jj=0;jj<dv;jj++)
for(int kk=0;kk<dv;kk++)
sum += data.GetValue(i*dv+ii,j*dv+jj,k*dv+kk);
data_small.SetValue(sum/float(dv*dv*dv), i,j,k);
}
Progress::inst->SetPercent(i/float(size2.x));
}
return data_small;
}
/// Constructor
/// @param M :: A matrix to copy.
GSLMatrix::GSLMatrix(const Kernel::Matrix<double> &M) {
m_matrix = gsl_matrix_alloc(M.numRows(), M.numCols());
for (size_t i = 0; i < size1(); ++i)
for (size_t j = 0; j < size2(); ++j) {
set(i, j, M[i][j]);
}
}
/// Copy a column into a GSLVector
/// @param i :: A column index.
GSLVector GSLMatrix::copyColumn(size_t i) const {
if (i >= size2()) {
throw std::out_of_range("GSLMatrix column index is out of range.");
}
auto columnView = gsl_matrix_const_column(gsl(), i);
return GSLVector(&columnView.vector);
}
bool fullMatrix<double>::invert(fullMatrix<double> &result) const
{
int M = size1(), N = size2(), lda = size1(), info;
int *ipiv = new int[std::min(M, N)];
if (result.size2() != M || result.size1() != N) {
if (result._own_data || !result._data)
result.resize(M,N,false);
else
Msg::Fatal("FullMatrix: Bad dimension, I cannot write in proxy");
}
result.setAll(*this);
F77NAME(dgetrf)(&M, &N, result._data, &lda, ipiv, &info);
if(info == 0){
int lwork = M * 4;
double *work = new double[lwork];
F77NAME(dgetri)(&M, result._data, &lda, ipiv, work, &lwork, &info);
delete [] work;
}
delete [] ipiv;
if(info == 0) return true;
else if(info > 0)
Msg::Error("U(%d,%d)=0 in matrix inversion", info, info);
else
Msg::Error("Wrong %d-th argument in matrix inversion", -info);
return false;
}
void run()
{
// Code from
// http://llvm.org/svn/llvm-project/libcxx/trunk/test/utilities/intseq/intseq.general/integer_seq.pass.cpp
// Make a couple of sequences
using int3 = std::make_integer_sequence<int, 3>; // generates int: 0,1,2
using size7 = std::make_integer_sequence<size_t, 7>; // generates size_t: 0,1,2,3,4,5,6
using size4 = std::make_index_sequence<4>; // generates size_t: 0,1,2,3
using size2 = std::index_sequence_for<int, size_t>; // generates size_t: 0,1
using intmix = std::integer_sequence<int, 9, 8, 7, 2>; // generates int: 9,8,7,2
using sizemix = std::index_sequence<1, 1, 2, 3, 5>; // generates size_t: 1,1,2,3,5
// Make sure they're what we expect
static_assert ( std::is_same <int3::value_type, int>::value, "int3 type wrong" );
static_assert ( int3::static_size == 3, "int3 size wrong" );
static_assert ( std::is_same <size7::value_type, size_t>::value, "size7 type wrong" );
static_assert ( size7::static_size == 7, "size7 size wrong" );
static_assert ( std::is_same <size4::value_type, size_t>::value, "size4 type wrong" );
static_assert ( size4::static_size == 4, "size4 size wrong" );
static_assert ( std::is_same <size2::value_type, size_t>::value, "size2 type wrong" );
static_assert ( size2::static_size == 2, "size2 size wrong" );
static_assert ( std::is_same <intmix::value_type, int>::value, "intmix type wrong" );
static_assert ( intmix::static_size == 4, "intmix size wrong" );
static_assert ( std::is_same <sizemix::value_type, size_t>::value, "sizemix type wrong" );
static_assert ( sizemix::static_size == 5, "sizemix size wrong" );
auto tup = std::make_tuple ( 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 );
// Use them
auto t3 = extract ( tup, int3() );
static_assert ( std::tuple_size<decltype(t3)>::value == int3::static_size, "t3 size wrong");
expect ( t3 == std::make_tuple ( 10, 11, 12 ));
auto t7 = extract ( tup, size7 ());
static_assert ( std::tuple_size<decltype(t7)>::value == size7::static_size, "t7 size wrong");
expect ( t7 == std::make_tuple ( 10, 11, 12, 13, 14, 15, 16 ));
auto t4 = extract ( tup, size4 ());
static_assert ( std::tuple_size<decltype(t4)>::value == size4::static_size, "t4 size wrong");
expect ( t4 == std::make_tuple ( 10, 11, 12, 13 ));
auto t2 = extract ( tup, size2 ());
static_assert ( std::tuple_size<decltype(t2)>::value == size2::static_size, "t2 size wrong");
expect ( t2 == std::make_tuple ( 10, 11 ));
auto tintmix = extract ( tup, intmix ());
static_assert ( std::tuple_size<decltype(tintmix)>::value == intmix::static_size, "tintmix size wrong");
expect ( tintmix == std::make_tuple ( 19, 18, 17, 12 ));
auto tsizemix = extract ( tup, sizemix ());
static_assert ( std::tuple_size<decltype(tsizemix)>::value == sizemix::static_size, "tsizemix size wrong");
expect ( tsizemix == std::make_tuple ( 11, 11, 12, 13, 15 ));
pass();
}
int main()
{
#if _LIBCPP_STD_VER > 11
// Make a couple of sequences
using int3 = std::make_integer_sequence<int, 3>; // generates int: 0,1,2
using size7 = std::make_integer_sequence<size_t, 7>; // generates size_t: 0,1,2,3,4,5,6
using size4 = std::make_index_sequence<4>; // generates size_t: 0,1,2,3
using size2 = std::index_sequence_for<int, size_t>; // generates size_t: 0,1
using intmix = std::integer_sequence<int, 9, 8, 7, 2>; // generates int: 9,8,7,2
using sizemix = std::index_sequence<1, 1, 2, 3, 5>; // generates size_t: 1,1,2,3,5
// Make sure they're what we expect
static_assert ( std::is_same<int3::value_type, int>::value, "int3 type wrong" );
static_assert ( int3::size () == 3, "int3 size wrong" );
static_assert ( std::is_same<size7::value_type, size_t>::value, "size7 type wrong" );
static_assert ( size7::size () == 7, "size7 size wrong" );
static_assert ( std::is_same<size4::value_type, size_t>::value, "size4 type wrong" );
static_assert ( size4::size () == 4, "size4 size wrong" );
static_assert ( std::is_same<size2::value_type, size_t>::value, "size2 type wrong" );
static_assert ( size2::size () == 2, "size2 size wrong" );
static_assert ( std::is_same<intmix::value_type, int>::value, "intmix type wrong" );
static_assert ( intmix::size () == 4, "intmix size wrong" );
static_assert ( std::is_same<sizemix::value_type, size_t>::value, "sizemix type wrong" );
static_assert ( sizemix::size () == 5, "sizemix size wrong" );
auto tup = std::make_tuple ( 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 );
// Use them
auto t3 = extract ( tup, int3() );
static_assert ( std::tuple_size<decltype(t3)>::value == int3::size (), "t3 size wrong");
assert ( t3 == std::make_tuple ( 10, 11, 12 ));
auto t7 = extract ( tup, size7 ());
static_assert ( std::tuple_size<decltype(t7)>::value == size7::size (), "t7 size wrong");
assert ( t7 == std::make_tuple ( 10, 11, 12, 13, 14, 15, 16 ));
auto t4 = extract ( tup, size4 ());
static_assert ( std::tuple_size<decltype(t4)>::value == size4::size (), "t4 size wrong");
assert ( t4 == std::make_tuple ( 10, 11, 12, 13 ));
auto t2 = extract ( tup, size2 ());
static_assert ( std::tuple_size<decltype(t2)>::value == size2::size (), "t2 size wrong");
assert ( t2 == std::make_tuple ( 10, 11 ));
auto tintmix = extract ( tup, intmix ());
static_assert ( std::tuple_size<decltype(tintmix)>::value == intmix::size (), "tintmix size wrong");
assert ( tintmix == std::make_tuple ( 19, 18, 17, 12 ));
auto tsizemix = extract ( tup, sizemix ());
static_assert ( std::tuple_size<decltype(tsizemix)>::value == sizemix::size (), "tsizemix size wrong");
assert ( tsizemix == std::make_tuple ( 11, 11, 12, 13, 15 ));
#endif // _LIBCPP_STD_VER > 11
}
inline
Eigen::Block<const Eigen::Matrix<Scalar, Eigen::Dynamic, Eigen::Dynamic> >
block(int start_row, int start_col, int rows, int cols) const {
assert(is_allocated());
assert(start_row+rows<=size1());
assert(start_col+cols<=size2());
return values_.block(start_row, start_col, rows, cols);
}
void CRSSparsity::spy(std::ostream &stream) const {
for (int i=0;i<size1();++i) {
for (int j=0;j<size2();++j) {
stream << (getNZ(i,j)==-1? "." : "*");
}
stream << std::endl;
}
}
double& el (int i, int j) {
int dead1, dead2;
if (j < size2 () && i-j < 1)
return data ().el (i, j);
barrier ();
return zero_;
}
bool fullMatrix<double>::luFactor(fullVector<int> &ipiv)
{
int M = size1(), N = size2(), lda = size1(), info;
ipiv.resize(std::min(M, N));
F77NAME(dgetrf)(&M, &N, _data, &lda, &ipiv(0), &info);
if(info == 0) return true;
return false;
}
void fullMatrix<double>::gemm(const fullMatrix<double> &a, const fullMatrix<double> &b,
double alpha, double beta, bool transposeA, bool transposeB)
{
int M = size1(), N = size2(), K = transposeA ? a.size1() : a.size2();
int LDA = a.size1(), LDB = b.size1(), LDC = size1();
F77NAME(dgemm)(transposeA ? "T" : "N", transposeB ? "T" :"N", &M, &N, &K, &alpha, a._data, &LDA, b._data, &LDB,
&beta, _data, &LDC);
}
vcl_size_t size(vector_expression<LHS, const int, op_matrix_diag> const & proxy)
{
int k = proxy.rhs();
int A_size1 = static_cast<int>(size1(proxy.lhs()));
int A_size2 = static_cast<int>(size2(proxy.lhs()));
int row_depth = std::min(A_size1, A_size1 + k);
int col_depth = std::min(A_size2, A_size2 - k);
return std::min(row_depth, col_depth);
}
/// Calculate the eigensystem of a symmetric matrix
/// @param eigenValues :: Output variable that receives the eigenvalues of this
/// matrix.
/// @param eigenVectors :: Output variable that receives the eigenvectors of
/// this matrix.
void GSLMatrix::eigenSystem(GSLVector &eigenValues, GSLMatrix &eigenVectors) {
size_t n = size1();
if (n != size2()) {
throw std::runtime_error("Matrix eigenSystem: the matrix must be square.");
}
eigenValues.resize(n);
eigenVectors.resize(n, n);
auto workspace = gsl_eigen_symmv_alloc(n);
gsl_eigen_symmv(gsl(), eigenValues.gsl(), eigenVectors.gsl(), workspace);
gsl_eigen_symmv_free(workspace);
}
int CRSSparsity::getNZ(int i, int j){
casadi_assert_message(i<size1() && j<size2(),"Indices out of bounds");
if (i<0) i += size1();
if (j<0) j += size2();
// Quick return if matrix is dense
if(numel()==size())
return j+i*size2();
// Quick return if we are adding an element to the end
if(rowind(i)==size() || (rowind(i+1)==size() && col().back()<j)){
vector<int>& colv = colRef();
vector<int>& rowindv = rowindRef();
colv.push_back(j);
for(int ii=i; ii<size1(); ++ii){
rowindv[ii+1]++;
}
return colv.size()-1;
}
// go to the place where the element should be
int ind;
for(ind=rowind(i); ind<rowind(i+1); ++ind){ // better: loop from the back to the front
if(col(ind) == j){
return ind; // element exists
} else if(col(ind) > j)
break; // break at the place where the element should be added
}
// Make sure that there no other objects are affected
makeUnique();
// insert the element
colRef().insert(colRef().begin()+ind,j);
for(int row=i+1; row<size1()+1; ++row)
rowindRef()[row]++;
// Return the location of the new element
return ind;
}
/// Invert this matrix
void GSLMatrix::invert() {
if (size1() != size2()) {
throw std::runtime_error("Matrix inverse: the matrix must be square.");
}
size_t n = size1();
int s;
GSLMatrix LU(*this);
gsl_permutation *p = gsl_permutation_alloc(n);
gsl_linalg_LU_decomp(LU.gsl(), p, &s);
gsl_linalg_LU_invert(LU.gsl(), p, this->gsl());
gsl_permutation_free(p);
}
/**
* @brief Read-write access to a single element of the matrix
*
* @param row_index Row index of accessed element
* @param col_index Column index of accessed element
* @return Proxy for matrix entry
*/
entry_proxy<SCALARTYPE> operator()(std::size_t row_index, std::size_t col_index)
{
assert(row_index < size1() && col_index < size2() && bool("Invalid access"));
int index = static_cast<int>(col_index) - static_cast<int>(row_index);
if (index < 0)
index = -index;
else if
(index > 0) index = 2 * size1() - index;
return elements_[index];
}
double spectral_norm_diag(const M &mat) {
typedef Eigen::Matrix<SCALAR, Eigen::Dynamic, Eigen::Dynamic> matrix_t;
const double cutoff = 1E-15;
const int rows = size1(mat);
const int cols = size2(mat);
if (rows == 0 || cols == 0) {
return 0.0;
}
matrix_t mat_tmp(rows, cols);
double max_abs = -1;
for (int j = 0; j < cols; ++j) {
for (int i = 0; i < rows; ++i) {
mat_tmp(i, j) = mat(i, j);
max_abs = std::max(max_abs, std::abs(mat_tmp(i, j)));
}
}
if (std::abs(max_abs) < 1E-300) {
return 0.0;
} else {
const double coeff = 1.0 / max_abs;
for (int j = 0; j < cols; ++j) {
for (int i = 0; i < rows; ++i) {
mat_tmp(i, j) *= coeff;
if (std::abs(mat_tmp(i, j)) < cutoff) {
mat_tmp(i, j) = 0.0;
}
}
}
if (mat_tmp.rows() > mat_tmp.cols()) {
mat_tmp = mat_tmp.adjoint() * mat_tmp;
} else {
mat_tmp = mat_tmp * mat_tmp.adjoint();
}
const double max_abs2 = mat_tmp.cwiseAbs().maxCoeff();
for (int j = 0; j < mat_tmp.cols(); ++j) {
for (int i = 0; i < mat_tmp.rows(); ++i) {
if (std::abs(mat_tmp(i, j)) < cutoff) {
mat_tmp(i, j) = 0.0;
}
}
}
Eigen::SelfAdjointEigenSolver<matrix_t> esolv(mat_tmp, false);
const double norm = std::sqrt(esolv.eigenvalues().cwiseAbs().maxCoeff()) / coeff;
if (std::isnan(norm)) {
std::cout << "Warning: spectral_norm_diag is NaN. max_abs = " << max_abs << " max_abs2 = " << max_abs2
<< std::endl;
return 0.0;
}
//assert(!isnan(norm));
return norm;
}
}
/// Calculate the determinant
double GSLMatrix::det() {
if (size1() != size2()) {
throw std::runtime_error("Matrix inverse: the matrix must be square.");
}
size_t n = size1();
int s;
GSLMatrix LU(*this);
gsl_permutation *p = gsl_permutation_alloc(n);
gsl_linalg_LU_decomp(LU.gsl(), p, &s);
double res = gsl_linalg_LU_det(LU.gsl(), s);
gsl_permutation_free(p);
return res;
}
/// Construct from an initialisation list
/// @param ilist :: Initialisation list as a list of rows:
/// {{M00, M01, M02, ...},
/// {M10, M11, M12, ...},
/// ...
/// {Mn0, Mn1, Mn2, ...}}
GSLMatrix::GSLMatrix(std::initializer_list<std::initializer_list<double>> ilist)
: GSLMatrix(ilist.size(), ilist.begin()->size()) {
for (auto row = ilist.begin(); row != ilist.end(); ++row) {
if (row->size() != size2()) {
throw std::runtime_error(
"All rows in initializer list must have the same size.");
}
auto i = static_cast<size_t>(std::distance(ilist.begin(), row));
for (auto cell = row->begin(); cell != row->end(); ++cell) {
auto j = static_cast<size_t>(std::distance(row->begin(), cell));
set(i, j, *cell);
}
}
}
bool Brush::operator==(const Brush& brush) const
{
return size1() == brush.size1() && size2() == brush.size2() &&
qAbs(hardness1() - brush.hardness1()) <= 1.0/256.0 &&
qAbs(hardness2() - brush.hardness2()) <= 1.0/256.0 &&
qAbs(opacity1() - brush.opacity1()) <= 1.0/256.0 &&
qAbs(opacity2() - brush.opacity2()) <= 1.0/256.0 &&
color1() == brush.color1() &&
color2() == brush.color2() &&
spacing() == brush.spacing() &&
subpixel() == brush.subpixel() &&
incremental() == brush.incremental() &&
blendingMode() == brush.blendingMode();
}
void Reshape::evaluateMX(const MXPtrV& input, MXPtrV& output, const MXPtrVV& fwdSeed, MXPtrVV& fwdSens, const MXPtrVV& adjSeed, MXPtrVV& adjSens, bool output_given){
// Quick return if inplace
if(input[0]==output[0]) return;
if(!output_given){
*output[0] = reshape(*input[0],size1(),size2());
}
// Forward sensitivities
int nfwd = fwdSens.size();
for(int d = 0; d<nfwd; ++d){
*fwdSens[d][0] = reshape(*fwdSeed[d][0],size1(),size2());
}
// Adjoint sensitivities
int nadj = adjSeed.size();
for(int d=0; d<nadj; ++d){
MX& aseed = *adjSeed[d][0];
MX& asens = *adjSens[d][0];
asens += reshape(aseed,dep().size1(),dep().size2());
aseed = MX();
}
}
void CWndNeuz::OnSize(UINT nType, int cx, int cy)
{
if(IsWndRoot())
return;
if( m_bTile ) //m_strTexture.IsEmpty() == FALSE )
{
CRect rectWnd = GetWndRect();
CSize size2( rectWnd.Width(), rectWnd.Height() );
CSize sizeDiv = size2;
sizeDiv.cx %= 16;
sizeDiv.cy %= 16;
size2.cx /= 16; size2.cx *= 16;
size2.cy /= 16; size2.cy *= 16;
if( sizeDiv.cx ) size2.cx += 16;
if( sizeDiv.cy ) size2.cy += 16;
rectWnd.bottom = rectWnd.top + size2.cy;
rectWnd.right = rectWnd.left + size2.cx;
SetWndRect( rectWnd, FALSE );
}
AdjustWndBase();
m_wndTitleBar.Replace();
CWndBase::OnSize( nType, cx, cy );
// if( rectOld.Width() != m_rectClient.Width() || rectOld.Height() != m_rectClient.Height() )
// 차일드 윈도우들의 사이즈를 조절
/*
for(int i = 0; i < m_wndArray.GetSize(); i++)
{
CWndBase* pWnd = (CWndBase*)m_wndArray.GetAt(i);
if(pWnd->IsWndStyle(WBS_DOCKED))// && pWnd->IsWndStyle(WBS_CHILD))
{
CRect rect = pWnd->GetWindowRect(TRUE);
rect.SetRect(0,0,cx,cy);
pWnd->SetWndRect(rect);
}
}
*/
/*
if(IsWndStyle(WBS_DOCKED))// && IsWndStyle(WBS_CHILD))
{
m_wndTitleBar.m_wndMinimize.SetVisible(FALSE);
m_wndTitleBar.m_wndMaximize.SetVisible(FALSE);
}
*/
}
int main()
{
int s1, s2;
printf("size1: %d\n", s1=size1());
printf("size2: %d\n", s2=size2());
printf("(correct output is 8, then 12)\n");
if (s1==8 && s2==12) {
return 0;
}
else {
return 2;
}
}
double spectral_norm_SVD(const M &mat) {
typedef Eigen::Matrix<SCALAR, Eigen::Dynamic, Eigen::Dynamic> matrix_t;
const double cutoff = 1E-15;
const int rows = size1(mat);
const int cols = size2(mat);
if (rows == 0 || cols == 0) {
return 0.0;
}
matrix_t mat_tmp(rows, cols);
double max_abs = -1;
for (int j = 0; j < cols; ++j) {
for (int i = 0; i < rows; ++i) {
mat_tmp(i, j) = mat(i, j);
max_abs = std::max(max_abs, std::abs(mat_tmp(i, j)));
}
}
if (max_abs == 0.0) {
return 0.0;
}
const double coeff = 1.0 / max_abs;
for (int j = 0; j < cols; ++j) {
for (int i = 0; i < rows; ++i) {
mat_tmp(i, j) *= coeff;
if (std::abs(mat_tmp(i, j)) < cutoff) {
mat_tmp(i, j) = 0.0;
}
}
}
//const double tmp = mat_tmp.squaredNorm();
Eigen::JacobiSVD<matrix_t> svd(mat_tmp);
#ifndef NDEBUG
const int size_SVD = svd.singularValues().size();
for (int i = 0; i < size_SVD - 1; ++i) {
assert(std::abs(svd.singularValues()[i]) >= std::abs(svd.singularValues()[i + 1]));
}
if (isnan(std::abs(svd.singularValues()[0]) / coeff)) {
std::cout << "Norm is Nan" << std::endl;
std::cout << "max_abs is " << max_abs << std::endl;
std::cout << "coeff is " << coeff << std::endl;
std::cout << "mat is " << std::endl << mat << std::endl;
std::cout << "mat_tmp is " << std::endl << mat_tmp << std::endl;
exit(-1);
}
#endif
const double norm = std::abs(svd.singularValues()[0]) / coeff;
assert(!isnan(norm));
return norm;
}
bool fullMatrix<double>::svd(fullMatrix<double> &V, fullVector<double> &S)
{
fullMatrix<double> VT(V.size2(), V.size1());
int M = size1(), N = size2(), LDA = size1(), LDVT = VT.size1(), info;
int lwork = std::max(3 * std::min(M, N) + std::max(M, N), 5 * std::min(M, N));
fullVector<double> WORK(lwork);
F77NAME(dgesvd)("O", "A", &M, &N, _data, &LDA, S._data, _data, &LDA,
VT._data, &LDVT, WORK._data, &lwork, &info);
V = VT.transpose();
if(info == 0) return true;
if(info > 0)
Msg::Error("SVD did not converge");
else
Msg::Error("Wrong %d-th argument in SVD decomposition", -info);
return false;
}
/// Matrix by vector multiplication
/// @param v :: A vector to multiply by. Must have the same size as size2().
/// @returns A vector - the result of the multiplication. Size of the returned
/// vector equals size1().
/// @throws std::invalid_argument if the input vector has a wrong size.
/// @throws std::runtime_error if the underlying GSL routine fails.
GSLVector GSLMatrix::operator*(const GSLVector &v) const {
if (v.size() != size2()) {
throw std::invalid_argument(
"Matrix by vector multiplication: wrong size of vector.");
}
GSLVector res(size1());
auto status =
gsl_blas_dgemv(CblasNoTrans, 1.0, gsl(), v.gsl(), 0.0, res.gsl());
if (status != GSL_SUCCESS) {
std::string message = "Failed to multiply matrix by a vector.\n"
"Error message returned by the GSL:\n" +
std::string(gsl_strerror(status));
throw std::runtime_error(message);
}
return res;
}
void FBOTexture::config(fzTextureFormat format, fzSize size, const fzPoint& anchorPoint, fzFloat quality)
{
FZ_ASSERT(getTexture() == NULL, "FBOTexture is already configured.");
fzFloat factor = Director::Instance().getResourcesFactor() * quality;
Texture2D *texture = new Texture2D(format, size * factor);
texture->setFactor(factor);
texture->setAliasTexParameters();
m_grabber.grab(texture);
// generate transform
fzSize size2(size.width * anchorPoint.x, size.height * anchorPoint.y);
// Inverted projection
fzMath_mat4OrthoProjection(-size2.width, size.width-size2.width, size.height-size2.height, -size2.height, -1500, 1500, m_transform);
}
/// Solve system of linear equations M*x == rhs, M is this matrix
/// This matrix is destroyed.
/// @param rhs :: The right-hand-side vector
/// @param x :: The solution vector
void GSLMatrix::solve(const GSLVector &rhs, GSLVector &x) {
if (size1() != size2()) {
throw std::runtime_error(
"System of linear equations: the matrix must be square.");
}
size_t n = size1();
if (rhs.size() != n) {
throw std::runtime_error(
"System of linear equations: right-hand side vector has wrong size.");
}
x.resize(n);
int s;
gsl_permutation *p = gsl_permutation_alloc(n);
gsl_linalg_LU_decomp(gsl(), p, &s); // matrix is modified at this moment
gsl_linalg_LU_solve(gsl(), p, rhs.gsl(), x.gsl());
gsl_permutation_free(p);
}
