Chris@49: // Copyright (C) 2010-2012 NICTA (www.nicta.com.au)
Chris@49: // Copyright (C) 2010-2012 Conrad Sanderson
Chris@49: // Copyright (C) 2010 Dimitrios Bouzas
Chris@49: // Copyright (C) 2011 Stanislav Funiak
Chris@49: // 
Chris@49: // This Source Code Form is subject to the terms of the Mozilla Public
Chris@49: // License, v. 2.0. If a copy of the MPL was not distributed with this
Chris@49: // file, You can obtain one at http://mozilla.org/MPL/2.0/.
Chris@49: 
Chris@49: 
Chris@49: //! \addtogroup op_princomp
Chris@49: //! @{
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! \brief
Chris@49: //! principal component analysis -- 4 arguments version
Chris@49: //! computation is done via singular value decomposition
Chris@49: //! coeff_out    -> principal component coefficients
Chris@49: //! score_out    -> projected samples
Chris@49: //! latent_out   -> eigenvalues of principal vectors
Chris@49: //! tsquared_out -> Hotelling's T^2 statistic
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: bool
Chris@49: op_princomp::direct_princomp
Chris@49:   (
Chris@49:          Mat<typename T1::elem_type>&     coeff_out,
Chris@49:          Mat<typename T1::elem_type>&     score_out,
Chris@49:          Col<typename T1::elem_type>&     latent_out,
Chris@49:          Col<typename T1::elem_type>&     tsquared_out,
Chris@49:   const Base<typename T1::elem_type, T1>& X,
Chris@49:   const typename arma_not_cx<typename T1::elem_type>::result* junk
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   arma_ignore(junk);
Chris@49:   
Chris@49:   typedef typename T1::elem_type eT;
Chris@49:   
Chris@49:   const unwrap_check<T1> Y( X.get_ref(), score_out );
Chris@49:   const Mat<eT>& in    = Y.M;
Chris@49: 
Chris@49:   const uword n_rows = in.n_rows;
Chris@49:   const uword n_cols = in.n_cols;
Chris@49:   
Chris@49:   if(n_rows > 1) // more than one sample
Chris@49:     {
Chris@49:     // subtract the mean - use score_out as temporary matrix
Chris@49:     score_out = in - repmat(mean(in), n_rows, 1);
Chris@49:  	  
Chris@49:     // singular value decomposition
Chris@49:     Mat<eT> U;
Chris@49:     Col<eT> s;
Chris@49:     
Chris@49:     const bool svd_ok = svd(U,s,coeff_out,score_out);
Chris@49:     
Chris@49:     if(svd_ok == false)
Chris@49:       {
Chris@49:       return false;
Chris@49:       }
Chris@49:     
Chris@49:     
Chris@49:     //U.reset();  // TODO: do we need this ?  U will get automatically deleted anyway
Chris@49:     
Chris@49:     // normalize the eigenvalues
Chris@49:     s /= std::sqrt( double(n_rows - 1) );
Chris@49:     
Chris@49:     // project the samples to the principals
Chris@49:     score_out *= coeff_out;
Chris@49:     
Chris@49:     if(n_rows <= n_cols) // number of samples is less than their dimensionality
Chris@49:       {
Chris@49:       score_out.cols(n_rows-1,n_cols-1).zeros();
Chris@49:       
Chris@49:       //Col<eT> s_tmp = zeros< Col<eT> >(n_cols);
Chris@49:       Col<eT> s_tmp(n_cols);
Chris@49:       s_tmp.zeros();
Chris@49:       
Chris@49:       s_tmp.rows(0,n_rows-2) = s.rows(0,n_rows-2);
Chris@49:       s = s_tmp;
Chris@49:           
Chris@49:       // compute the Hotelling's T-squared
Chris@49:       s_tmp.rows(0,n_rows-2) = eT(1) / s_tmp.rows(0,n_rows-2);
Chris@49:       
Chris@49:       const Mat<eT> S = score_out * diagmat(Col<eT>(s_tmp));   
Chris@49:       tsquared_out = sum(S%S,1); 
Chris@49:       }
Chris@49:     else
Chris@49:       {
Chris@49:       // compute the Hotelling's T-squared   
Chris@49:       const Mat<eT> S = score_out * diagmat(Col<eT>( eT(1) / s));
Chris@49:       tsquared_out = sum(S%S,1);
Chris@49:       }
Chris@49:             
Chris@49:     // compute the eigenvalues of the principal vectors
Chris@49:     latent_out = s%s;
Chris@49:     }
Chris@49:   else // 0 or 1 samples
Chris@49:     {
Chris@49:     coeff_out.eye(n_cols, n_cols);
Chris@49:     
Chris@49:     score_out.copy_size(in);
Chris@49:     score_out.zeros();
Chris@49:     
Chris@49:     latent_out.set_size(n_cols);
Chris@49:     latent_out.zeros();
Chris@49:     
Chris@49:     tsquared_out.set_size(n_rows);
Chris@49:     tsquared_out.zeros();
Chris@49:     }
Chris@49:   
Chris@49:   return true;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! \brief
Chris@49: //! principal component analysis -- 3 arguments version
Chris@49: //! computation is done via singular value decomposition
Chris@49: //! coeff_out    -> principal component coefficients
Chris@49: //! score_out    -> projected samples
Chris@49: //! latent_out   -> eigenvalues of principal vectors
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: bool
Chris@49: op_princomp::direct_princomp
Chris@49:   (
Chris@49:          Mat<typename T1::elem_type>&     coeff_out,
Chris@49:          Mat<typename T1::elem_type>&     score_out,
Chris@49:          Col<typename T1::elem_type>&     latent_out,
Chris@49:   const Base<typename T1::elem_type, T1>& X,
Chris@49:   const typename arma_not_cx<typename T1::elem_type>::result* junk
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   arma_ignore(junk);
Chris@49:   
Chris@49:   typedef typename T1::elem_type eT;
Chris@49:   
Chris@49:   const unwrap_check<T1> Y( X.get_ref(), score_out );
Chris@49:   const Mat<eT>& in    = Y.M;
Chris@49:   
Chris@49:   const uword n_rows = in.n_rows;
Chris@49:   const uword n_cols = in.n_cols;
Chris@49:   
Chris@49:   if(n_rows > 1) // more than one sample
Chris@49:     {
Chris@49:     // subtract the mean - use score_out as temporary matrix
Chris@49:     score_out = in - repmat(mean(in), n_rows, 1);
Chris@49:  	  
Chris@49:     // singular value decomposition
Chris@49:     Mat<eT> U;
Chris@49:     Col<eT> s;
Chris@49:     
Chris@49:     const bool svd_ok = svd(U,s,coeff_out,score_out);
Chris@49:     
Chris@49:     if(svd_ok == false)
Chris@49:       {
Chris@49:       return false;
Chris@49:       }
Chris@49:     
Chris@49:     
Chris@49:     // U.reset();
Chris@49:     
Chris@49:     // normalize the eigenvalues
Chris@49:     s /= std::sqrt( double(n_rows - 1) );
Chris@49:     
Chris@49:     // project the samples to the principals
Chris@49:     score_out *= coeff_out;
Chris@49:     
Chris@49:     if(n_rows <= n_cols) // number of samples is less than their dimensionality
Chris@49:       {
Chris@49:       score_out.cols(n_rows-1,n_cols-1).zeros();
Chris@49:       
Chris@49:       Col<eT> s_tmp = zeros< Col<eT> >(n_cols);
Chris@49:       s_tmp.rows(0,n_rows-2) = s.rows(0,n_rows-2);
Chris@49:       s = s_tmp;
Chris@49:       }
Chris@49:       
Chris@49:     // compute the eigenvalues of the principal vectors
Chris@49:     latent_out = s%s;
Chris@49:     
Chris@49:     }
Chris@49:   else // 0 or 1 samples
Chris@49:     {
Chris@49:     coeff_out.eye(n_cols, n_cols);
Chris@49:     
Chris@49:     score_out.copy_size(in);
Chris@49:     score_out.zeros();
Chris@49:     
Chris@49:     latent_out.set_size(n_cols);
Chris@49:     latent_out.zeros(); 
Chris@49:     }
Chris@49:   
Chris@49:   return true;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! \brief
Chris@49: //! principal component analysis -- 2 arguments version
Chris@49: //! computation is done via singular value decomposition
Chris@49: //! coeff_out    -> principal component coefficients
Chris@49: //! score_out    -> projected samples
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: bool
Chris@49: op_princomp::direct_princomp
Chris@49:   (
Chris@49:          Mat<typename T1::elem_type>&     coeff_out,
Chris@49:          Mat<typename T1::elem_type>&     score_out,
Chris@49:   const Base<typename T1::elem_type, T1>& X,
Chris@49:   const typename arma_not_cx<typename T1::elem_type>::result* junk
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   arma_ignore(junk);
Chris@49:   
Chris@49:   typedef typename T1::elem_type eT;
Chris@49:   
Chris@49:   const unwrap_check<T1> Y( X.get_ref(), score_out );
Chris@49:   const Mat<eT>& in    = Y.M;
Chris@49:   
Chris@49:   const uword n_rows = in.n_rows;
Chris@49:   const uword n_cols = in.n_cols;
Chris@49:   
Chris@49:   if(n_rows > 1) // more than one sample
Chris@49:     {
Chris@49:     // subtract the mean - use score_out as temporary matrix
Chris@49:     score_out = in - repmat(mean(in), n_rows, 1);
Chris@49:  	  
Chris@49:     // singular value decomposition
Chris@49:     Mat<eT> U;
Chris@49:     Col<eT> s;
Chris@49:     
Chris@49:     const bool svd_ok = svd(U,s,coeff_out,score_out);
Chris@49:     
Chris@49:     if(svd_ok == false)
Chris@49:       {
Chris@49:       return false;
Chris@49:       }
Chris@49:     
Chris@49:     // U.reset();
Chris@49:     
Chris@49:     // normalize the eigenvalues
Chris@49:     s /= std::sqrt( double(n_rows - 1) );
Chris@49:     
Chris@49:     // project the samples to the principals
Chris@49:     score_out *= coeff_out;
Chris@49:     
Chris@49:     if(n_rows <= n_cols) // number of samples is less than their dimensionality
Chris@49:       {
Chris@49:       score_out.cols(n_rows-1,n_cols-1).zeros();
Chris@49:       
Chris@49:       Col<eT> s_tmp = zeros< Col<eT> >(n_cols);
Chris@49:       s_tmp.rows(0,n_rows-2) = s.rows(0,n_rows-2);
Chris@49:       s = s_tmp;
Chris@49:       }
Chris@49:     }
Chris@49:   else // 0 or 1 samples
Chris@49:     {
Chris@49:     coeff_out.eye(n_cols, n_cols);
Chris@49:     score_out.copy_size(in);
Chris@49:     score_out.zeros();
Chris@49:     }
Chris@49:   
Chris@49:   return true;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! \brief
Chris@49: //! principal component analysis -- 1 argument version
Chris@49: //! computation is done via singular value decomposition
Chris@49: //! coeff_out    -> principal component coefficients
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: bool
Chris@49: op_princomp::direct_princomp
Chris@49:   (
Chris@49:          Mat<typename T1::elem_type>&     coeff_out,
Chris@49:   const Base<typename T1::elem_type, T1>& X,
Chris@49:   const typename arma_not_cx<typename T1::elem_type>::result* junk
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   arma_ignore(junk);
Chris@49:   
Chris@49:   typedef typename T1::elem_type eT;
Chris@49:   
Chris@49:   const unwrap<T1>    Y( X.get_ref() );
Chris@49:   const Mat<eT>& in = Y.M;
Chris@49:   
Chris@49:   if(in.n_elem != 0)
Chris@49:     {
Chris@49:     // singular value decomposition
Chris@49:     Mat<eT> U;
Chris@49:     Col<eT> s;
Chris@49:     
Chris@49:     const Mat<eT> tmp = in - repmat(mean(in), in.n_rows, 1);
Chris@49:     
Chris@49:     const bool svd_ok = svd(U,s,coeff_out, tmp);
Chris@49:     
Chris@49:     if(svd_ok == false)
Chris@49:       {
Chris@49:       return false;
Chris@49:       }
Chris@49:     }
Chris@49:   else
Chris@49:     {
Chris@49:     coeff_out.eye(in.n_cols, in.n_cols);
Chris@49:     }
Chris@49:   
Chris@49:   return true;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! \brief
Chris@49: //! principal component analysis -- 4 arguments complex version
Chris@49: //! computation is done via singular value decomposition
Chris@49: //! coeff_out    -> principal component coefficients
Chris@49: //! score_out    -> projected samples
Chris@49: //! latent_out   -> eigenvalues of principal vectors
Chris@49: //! tsquared_out -> Hotelling's T^2 statistic
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: bool
Chris@49: op_princomp::direct_princomp
Chris@49:   (
Chris@49:          Mat< std::complex<typename T1::pod_type> >&     coeff_out,
Chris@49:          Mat< std::complex<typename T1::pod_type> >&     score_out,
Chris@49:          Col<              typename T1::pod_type  >&     latent_out,
Chris@49:          Col< std::complex<typename T1::pod_type> >&     tsquared_out,
Chris@49:   const Base< std::complex<typename T1::pod_type>, T1 >& X,
Chris@49:   const typename arma_cx_only<typename T1::elem_type>::result* junk
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   arma_ignore(junk);
Chris@49:   
Chris@49:   typedef typename T1::pod_type     T;
Chris@49:   typedef          std::complex<T> eT;
Chris@49:   
Chris@49:   const unwrap_check<T1> Y( X.get_ref(), score_out );
Chris@49:   const Mat<eT>& in    = Y.M;
Chris@49:   
Chris@49:   const uword n_rows = in.n_rows;
Chris@49:   const uword n_cols = in.n_cols;
Chris@49:   
Chris@49:   if(n_rows > 1) // more than one sample
Chris@49:     {
Chris@49:     // subtract the mean - use score_out as temporary matrix
Chris@49:     score_out = in - repmat(mean(in), n_rows, 1);
Chris@49:  	  
Chris@49:     // singular value decomposition
Chris@49:     Mat<eT> U;
Chris@49:     Col<T> s;
Chris@49:     
Chris@49:     const bool svd_ok = svd(U,s,coeff_out,score_out); 
Chris@49:     
Chris@49:     if(svd_ok == false)
Chris@49:       {
Chris@49:       return false;
Chris@49:       }
Chris@49:     
Chris@49:     
Chris@49:     //U.reset();
Chris@49:     
Chris@49:     // normalize the eigenvalues
Chris@49:     s /= std::sqrt( double(n_rows - 1) );
Chris@49:     
Chris@49:     // project the samples to the principals
Chris@49:     score_out *= coeff_out;
Chris@49:     
Chris@49:     if(n_rows <= n_cols) // number of samples is less than their dimensionality
Chris@49:       {
Chris@49:       score_out.cols(n_rows-1,n_cols-1).zeros();
Chris@49:       
Chris@49:       Col<T> s_tmp = zeros< Col<T> >(n_cols);
Chris@49:       s_tmp.rows(0,n_rows-2) = s.rows(0,n_rows-2);
Chris@49:       s = s_tmp;
Chris@49:           
Chris@49:       // compute the Hotelling's T-squared   
Chris@49:       s_tmp.rows(0,n_rows-2) = 1.0 / s_tmp.rows(0,n_rows-2);
Chris@49:       const Mat<eT> S = score_out * diagmat(Col<T>(s_tmp));                     
Chris@49:       tsquared_out = sum(S%S,1); 
Chris@49:       }
Chris@49:     else
Chris@49:       {
Chris@49:       // compute the Hotelling's T-squared   
Chris@49:       const Mat<eT> S = score_out * diagmat(Col<T>(T(1) / s));                     
Chris@49:       tsquared_out = sum(S%S,1);
Chris@49:       }
Chris@49:     
Chris@49:     // compute the eigenvalues of the principal vectors
Chris@49:     latent_out = s%s;
Chris@49:     
Chris@49:     }
Chris@49:   else // 0 or 1 samples
Chris@49:     {
Chris@49:     coeff_out.eye(n_cols, n_cols);
Chris@49:     
Chris@49:     score_out.copy_size(in);
Chris@49:     score_out.zeros();
Chris@49:       
Chris@49:     latent_out.set_size(n_cols);
Chris@49:     latent_out.zeros();
Chris@49:       
Chris@49:     tsquared_out.set_size(n_rows);
Chris@49:     tsquared_out.zeros();
Chris@49:     }
Chris@49:   
Chris@49:   return true;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! \brief
Chris@49: //! principal component analysis -- 3 arguments complex version
Chris@49: //! computation is done via singular value decomposition
Chris@49: //! coeff_out    -> principal component coefficients
Chris@49: //! score_out    -> projected samples
Chris@49: //! latent_out   -> eigenvalues of principal vectors
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: bool
Chris@49: op_princomp::direct_princomp
Chris@49:   (
Chris@49:          Mat< std::complex<typename T1::pod_type> >&     coeff_out,
Chris@49:          Mat< std::complex<typename T1::pod_type> >&     score_out,
Chris@49:          Col<              typename T1::pod_type  >&     latent_out,
Chris@49:   const Base< std::complex<typename T1::pod_type>, T1 >& X,
Chris@49:   const typename arma_cx_only<typename T1::elem_type>::result* junk
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   arma_ignore(junk);
Chris@49:   
Chris@49:   typedef typename T1::pod_type     T;
Chris@49:   typedef          std::complex<T> eT;
Chris@49:   
Chris@49:   const unwrap_check<T1> Y( X.get_ref(), score_out );
Chris@49:   const Mat<eT>& in    = Y.M;
Chris@49:   
Chris@49:   const uword n_rows = in.n_rows;
Chris@49:   const uword n_cols = in.n_cols;
Chris@49:   
Chris@49:   if(n_rows > 1) // more than one sample
Chris@49:     {
Chris@49:     // subtract the mean - use score_out as temporary matrix
Chris@49:     score_out = in - repmat(mean(in), n_rows, 1);
Chris@49:  	  
Chris@49:     // singular value decomposition
Chris@49:     Mat<eT> U;
Chris@49:     Col< T> s;
Chris@49:     
Chris@49:     const bool svd_ok = svd(U,s,coeff_out,score_out);
Chris@49:     
Chris@49:     if(svd_ok == false)
Chris@49:       {
Chris@49:       return false;
Chris@49:       }
Chris@49:     
Chris@49:     
Chris@49:     // U.reset();
Chris@49:     
Chris@49:     // normalize the eigenvalues
Chris@49:     s /= std::sqrt( double(n_rows - 1) );
Chris@49:     
Chris@49:     // project the samples to the principals
Chris@49:     score_out *= coeff_out;
Chris@49:     
Chris@49:     if(n_rows <= n_cols) // number of samples is less than their dimensionality
Chris@49:       {
Chris@49:       score_out.cols(n_rows-1,n_cols-1).zeros();
Chris@49:       
Chris@49:       Col<T> s_tmp = zeros< Col<T> >(n_cols);
Chris@49:       s_tmp.rows(0,n_rows-2) = s.rows(0,n_rows-2);
Chris@49:       s = s_tmp;
Chris@49:       }
Chris@49:       
Chris@49:     // compute the eigenvalues of the principal vectors
Chris@49:     latent_out = s%s;
Chris@49: 
Chris@49:     }
Chris@49:   else // 0 or 1 samples
Chris@49:     {
Chris@49:     coeff_out.eye(n_cols, n_cols);
Chris@49: 
Chris@49:     score_out.copy_size(in);
Chris@49:     score_out.zeros();
Chris@49: 
Chris@49:     latent_out.set_size(n_cols);
Chris@49:     latent_out.zeros();
Chris@49:     }
Chris@49:   
Chris@49:   return true;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! \brief
Chris@49: //! principal component analysis -- 2 arguments complex version
Chris@49: //! computation is done via singular value decomposition
Chris@49: //! coeff_out    -> principal component coefficients
Chris@49: //! score_out    -> projected samples
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: bool
Chris@49: op_princomp::direct_princomp
Chris@49:   (
Chris@49:          Mat< std::complex<typename T1::pod_type> >&     coeff_out,
Chris@49:          Mat< std::complex<typename T1::pod_type> >&     score_out,
Chris@49:   const Base< std::complex<typename T1::pod_type>, T1 >& X,
Chris@49:   const typename arma_cx_only<typename T1::elem_type>::result* junk
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   arma_ignore(junk);
Chris@49:   
Chris@49:   typedef typename T1::pod_type     T;
Chris@49:   typedef          std::complex<T> eT;
Chris@49:   
Chris@49:   const unwrap_check<T1> Y( X.get_ref(), score_out );
Chris@49:   const Mat<eT>& in    = Y.M;
Chris@49:   
Chris@49:   const uword n_rows = in.n_rows;
Chris@49:   const uword n_cols = in.n_cols;
Chris@49:   
Chris@49:   if(n_rows > 1) // more than one sample
Chris@49:     {
Chris@49:     // subtract the mean - use score_out as temporary matrix
Chris@49:     score_out = in - repmat(mean(in), n_rows, 1);
Chris@49:  	  
Chris@49:     // singular value decomposition
Chris@49:     Mat<eT> U;
Chris@49:     Col< T> s;
Chris@49:     
Chris@49:     const bool svd_ok = svd(U,s,coeff_out,score_out);
Chris@49:     
Chris@49:     if(svd_ok == false)
Chris@49:       {
Chris@49:       return false;
Chris@49:       }
Chris@49:     
Chris@49:     // U.reset();
Chris@49:     
Chris@49:     // normalize the eigenvalues
Chris@49:     s /= std::sqrt( double(n_rows - 1) );
Chris@49: 
Chris@49:     // project the samples to the principals
Chris@49:     score_out *= coeff_out;
Chris@49: 
Chris@49:     if(n_rows <= n_cols) // number of samples is less than their dimensionality
Chris@49:       {
Chris@49:       score_out.cols(n_rows-1,n_cols-1).zeros();
Chris@49:       }
Chris@49: 
Chris@49:     }
Chris@49:   else // 0 or 1 samples
Chris@49:     {
Chris@49:     coeff_out.eye(n_cols, n_cols);
Chris@49:     
Chris@49:     score_out.copy_size(in);
Chris@49:     score_out.zeros();
Chris@49:     }
Chris@49:   
Chris@49:   return true;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! \brief
Chris@49: //! principal component analysis -- 1 argument complex version
Chris@49: //! computation is done via singular value decomposition
Chris@49: //! coeff_out    -> principal component coefficients
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: bool
Chris@49: op_princomp::direct_princomp
Chris@49:   (
Chris@49:          Mat< std::complex<typename T1::pod_type> >&     coeff_out,
Chris@49:   const Base< std::complex<typename T1::pod_type>, T1 >& X,
Chris@49:   const typename arma_cx_only<typename T1::elem_type>::result* junk
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   arma_ignore(junk);
Chris@49:   
Chris@49:   typedef typename T1::pod_type     T;
Chris@49:   typedef          std::complex<T> eT;
Chris@49:   
Chris@49:   const unwrap<T1>    Y( X.get_ref() );
Chris@49:   const Mat<eT>& in = Y.M;
Chris@49:   
Chris@49:   if(in.n_elem != 0)
Chris@49:     {
Chris@49:  	  // singular value decomposition
Chris@49:  	  Mat<eT> U;
Chris@49:     Col< T> s;
Chris@49:     
Chris@49:     const Mat<eT> tmp = in - repmat(mean(in), in.n_rows, 1);
Chris@49:     
Chris@49:     const bool svd_ok = svd(U,s,coeff_out, tmp);
Chris@49:     
Chris@49:     if(svd_ok == false)
Chris@49:       {
Chris@49:       return false;
Chris@49:       }
Chris@49:     }
Chris@49:   else
Chris@49:     {
Chris@49:     coeff_out.eye(in.n_cols, in.n_cols);
Chris@49:     }
Chris@49:   
Chris@49:   return true;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: template<typename T1>
Chris@49: inline
Chris@49: void
Chris@49: op_princomp::apply
Chris@49:   (
Chris@49:         Mat<typename T1::elem_type>& out,
Chris@49:   const Op<T1,op_princomp>&          in
Chris@49:   )
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   typedef typename T1::elem_type eT;
Chris@49:   
Chris@49:   const unwrap_check<T1> tmp(in.m, out);
Chris@49:   const Mat<eT>& A     = tmp.M;
Chris@49:   
Chris@49:   const bool status = op_princomp::direct_princomp(out, A);
Chris@49:   
Chris@49:   if(status == false)
Chris@49:     {
Chris@49:     out.reset();
Chris@49:     
Chris@49:     arma_bad("princomp(): failed to converge");
Chris@49:     }
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! @}