max@0: // Copyright (C) 2009-2011 NICTA (www.nicta.com.au)
max@0: // Copyright (C) 2009-2011 Conrad Sanderson
max@0: // 
max@0: // This file is part of the Armadillo C++ library.
max@0: // It is provided without any warranty of fitness
max@0: // for any purpose. You can redistribute this file
max@0: // and/or modify it under the terms of the GNU
max@0: // Lesser General Public License (LGPL) as published
max@0: // by the Free Software Foundation, either version 3
max@0: // of the License or (at your option) any later version.
max@0: // (see http://www.opensource.org/licenses for more info)
max@0: 
max@0: 
max@0: //! \addtogroup running_stat_vec
max@0: //! @{
max@0: 
max@0: 
max@0: 
max@0: template<typename eT>
max@0: running_stat_vec<eT>::~running_stat_vec()
max@0:   {
max@0:   arma_extra_debug_sigprint_this(this);
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: template<typename eT>
max@0: running_stat_vec<eT>::running_stat_vec(const bool in_calc_cov)
max@0:   : calc_cov(in_calc_cov)
max@0:   {
max@0:   arma_extra_debug_sigprint_this(this);
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: template<typename eT>
max@0: running_stat_vec<eT>::running_stat_vec(const running_stat_vec<eT>& in_rsv)
max@0:   : calc_cov    (in_rsv.calc_cov)
max@0:   , counter     (in_rsv.counter)
max@0:   , r_mean      (in_rsv.r_mean)
max@0:   , r_var       (in_rsv.r_var)
max@0:   , r_cov       (in_rsv.r_cov)
max@0:   , min_val     (in_rsv.min_val)
max@0:   , max_val     (in_rsv.max_val)
max@0:   , min_val_norm(in_rsv.min_val_norm)
max@0:   , max_val_norm(in_rsv.max_val_norm)
max@0:   {
max@0:   arma_extra_debug_sigprint_this(this);
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: template<typename eT>
max@0: const running_stat_vec<eT>&
max@0: running_stat_vec<eT>::operator=(const running_stat_vec<eT>& in_rsv)
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   access::rw(calc_cov) = in_rsv.calc_cov;
max@0:   
max@0:   counter      = in_rsv.counter;
max@0:   r_mean       = in_rsv.r_mean;
max@0:   r_var        = in_rsv.r_var;
max@0:   r_cov        = in_rsv.r_cov;
max@0:   min_val      = in_rsv.min_val;
max@0:   max_val      = in_rsv.max_val;
max@0:   min_val_norm = in_rsv.min_val_norm;
max@0:   max_val_norm = in_rsv.max_val_norm;
max@0:   
max@0:   return *this;
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! update statistics to reflect new sample
max@0: template<typename eT>
max@0: template<typename T1>
max@0: arma_hot
max@0: inline
max@0: void
max@0: running_stat_vec<eT>::operator() (const Base<typename get_pod_type<eT>::result, T1>& X)
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   //typedef typename get_pod_type<eT>::result T;
max@0:   
max@0:   const unwrap<T1>        tmp(X.get_ref());
max@0:   const Mat<eT>& sample = tmp.M;
max@0:   
max@0:   if( sample.is_empty() )
max@0:     {
max@0:     return;
max@0:     }
max@0:   
max@0:   if( sample.is_finite() == false )
max@0:     {
max@0:     arma_warn(true, "running_stat_vec: sample ignored as it has non-finite elements");
max@0:     return;
max@0:     }
max@0:   
max@0:   running_stat_vec_aux::update_stats(*this, sample);
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! update statistics to reflect new sample (version for complex numbers)
max@0: template<typename eT>
max@0: template<typename T1>
max@0: arma_hot
max@0: inline
max@0: void
max@0: running_stat_vec<eT>::operator() (const Base<std::complex<typename get_pod_type<eT>::result>, T1>& X)
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   //typedef typename std::complex<typename get_pod_type<eT>::result> eT;
max@0:   
max@0:   const unwrap<T1>        tmp(X.get_ref());
max@0:   const Mat<eT>& sample = tmp.M;
max@0:   
max@0:   if( sample.is_empty() )
max@0:     {
max@0:     return;
max@0:     }
max@0:   
max@0:   if( sample.is_finite() == false )
max@0:     {
max@0:     arma_warn(true, "running_stat_vec: sample ignored as it has non-finite elements");
max@0:     return;
max@0:     }
max@0:   
max@0:   running_stat_vec_aux::update_stats(*this, sample);
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! set all statistics to zero
max@0: template<typename eT>
max@0: inline
max@0: void
max@0: running_stat_vec<eT>::reset()
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   counter.reset();
max@0:   
max@0:   r_mean.reset();
max@0:   r_var.reset();
max@0:   r_cov.reset();
max@0:   
max@0:   min_val.reset();
max@0:   max_val.reset();
max@0:   
max@0:   min_val_norm.reset();
max@0:   max_val_norm.reset();
max@0:   
max@0:   r_var_dummy.reset();
max@0:   r_cov_dummy.reset();
max@0:   
max@0:   tmp1.reset();
max@0:   tmp2.reset();
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! mean or average value
max@0: template<typename eT>
max@0: inline
max@0: const Mat<eT>&
max@0: running_stat_vec<eT>::mean() const
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   return r_mean;
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! variance
max@0: template<typename eT>
max@0: inline
max@0: const Mat<typename get_pod_type<eT>::result>&
max@0: running_stat_vec<eT>::var(const uword norm_type)
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   const T N = counter.value();
max@0:   
max@0:   if(N > T(1))
max@0:     {
max@0:     if(norm_type == 0)
max@0:       {
max@0:       return r_var;
max@0:       }
max@0:     else
max@0:       {
max@0:       const T N_minus_1 = counter.value_minus_1();
max@0:       
max@0:       r_var_dummy = (N_minus_1/N) * r_var;
max@0:       
max@0:       return r_var_dummy;
max@0:       }
max@0:     }
max@0:   else
max@0:     {
max@0:     r_var_dummy.zeros(r_mean.n_rows, r_mean.n_cols);
max@0:     
max@0:     return r_var_dummy;
max@0:     }
max@0:   
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! standard deviation
max@0: template<typename eT>
max@0: inline
max@0: Mat<typename get_pod_type<eT>::result>
max@0: running_stat_vec<eT>::stddev(const uword norm_type) const
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   const T N = counter.value();
max@0:   
max@0:   if(N > T(1))
max@0:     {
max@0:     if(norm_type == 0)
max@0:       {
max@0:       return sqrt(r_var);
max@0:       }
max@0:     else
max@0:       {
max@0:       const T N_minus_1 = counter.value_minus_1();
max@0:       
max@0:       return sqrt( (N_minus_1/N) * r_var );
max@0:       }
max@0:     }
max@0:   else
max@0:     {
max@0:     return Mat<T>();
max@0:     }
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! covariance
max@0: template<typename eT>
max@0: inline
max@0: const Mat<eT>&
max@0: running_stat_vec<eT>::cov(const uword norm_type)
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   if(calc_cov == true)
max@0:     {
max@0:     const T N = counter.value();
max@0:     
max@0:     if(N > T(1))
max@0:       {
max@0:       if(norm_type == 0)
max@0:         {
max@0:         return r_cov;
max@0:         }
max@0:       else
max@0:         {
max@0:         const T N_minus_1 = counter.value_minus_1();
max@0:         
max@0:         r_cov_dummy = (N_minus_1/N) * r_cov;
max@0:         
max@0:         return r_cov_dummy;
max@0:         }
max@0:       }
max@0:     else
max@0:       {
max@0:       r_cov_dummy.zeros(r_mean.n_rows, r_mean.n_cols);
max@0:       
max@0:       return r_cov_dummy;
max@0:       }
max@0:     }
max@0:   else
max@0:     {
max@0:     r_cov_dummy.reset();
max@0:     
max@0:     return r_cov_dummy;
max@0:     }
max@0:   
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! vector with minimum values
max@0: template<typename eT>
max@0: inline
max@0: const Mat<eT>&
max@0: running_stat_vec<eT>::min() const
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   return min_val;
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! vector with maximum values
max@0: template<typename eT>
max@0: inline
max@0: const Mat<eT>&
max@0: running_stat_vec<eT>::max() const
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   return max_val;
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! number of samples so far
max@0: template<typename eT>
max@0: inline
max@0: typename get_pod_type<eT>::result
max@0: running_stat_vec<eT>::count() const
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   return counter.value();
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //
max@0: 
max@0: 
max@0: 
max@0: //! update statistics to reflect new sample
max@0: template<typename eT>
max@0: inline
max@0: void
max@0: running_stat_vec_aux::update_stats(running_stat_vec<eT>& x, const Mat<eT>& sample)
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   typedef typename running_stat_vec<eT>::T T;
max@0:   
max@0:   const T N = x.counter.value();
max@0:   
max@0:   if(N > T(0))
max@0:     {
max@0:     arma_debug_assert_same_size(x.r_mean, sample, "running_stat_vec(): dimensionality mismatch");
max@0:     
max@0:     const uword n_elem      = sample.n_elem;
max@0:     const eT* sample_mem  = sample.memptr();
max@0:           eT* r_mean_mem  = x.r_mean.memptr();
max@0:            T* r_var_mem   = x.r_var.memptr();
max@0:           eT* min_val_mem = x.min_val.memptr();
max@0:           eT* max_val_mem = x.max_val.memptr();
max@0:     
max@0:     const T  N_plus_1   = x.counter.value_plus_1();
max@0:     const T  N_minus_1  = x.counter.value_minus_1();
max@0:     
max@0:     if(x.calc_cov == true)
max@0:       {
max@0:       Mat<eT>& tmp1 = x.tmp1;
max@0:       Mat<eT>& tmp2 = x.tmp2;
max@0:       
max@0:       tmp1 = sample - x.r_mean;
max@0:       
max@0:       if(sample.n_cols == 1)
max@0:         {
max@0:         tmp2 = tmp1*trans(tmp1);
max@0:         }
max@0:       else
max@0:         {
max@0:         tmp2 = trans(tmp1)*tmp1;
max@0:         }
max@0:       
max@0:       x.r_cov *= (N_minus_1/N);
max@0:       x.r_cov += tmp2 / N_plus_1;
max@0:       }
max@0:     
max@0:     
max@0:     for(uword i=0; i<n_elem; ++i)
max@0:       {
max@0:       const eT val = sample_mem[i];
max@0:       
max@0:       if(val < min_val_mem[i])
max@0:         {
max@0:         min_val_mem[i] = val;
max@0:         }
max@0:       
max@0:       if(val > max_val_mem[i])
max@0:         {
max@0:         max_val_mem[i] = val;
max@0:         }
max@0:         
max@0:       const eT r_mean_val = r_mean_mem[i];
max@0:       const eT tmp        = val - r_mean_val;
max@0:     
max@0:       r_var_mem[i] = N_minus_1/N * r_var_mem[i] + (tmp*tmp)/N_plus_1;
max@0:       
max@0:       r_mean_mem[i] = r_mean_val + (val - r_mean_val)/N_plus_1;
max@0:       }
max@0:     }
max@0:   else
max@0:     {
max@0:     arma_debug_check( (sample.is_vec() == false), "running_stat_vec(): given sample is not a vector");
max@0:     
max@0:     x.r_mean.set_size(sample.n_rows, sample.n_cols);
max@0:     
max@0:     x.r_var.zeros(sample.n_rows, sample.n_cols);
max@0:     
max@0:     if(x.calc_cov == true)
max@0:       {
max@0:       x.r_cov.zeros(sample.n_elem, sample.n_elem);
max@0:       }
max@0:     
max@0:     x.min_val.set_size(sample.n_rows, sample.n_cols);
max@0:     x.max_val.set_size(sample.n_rows, sample.n_cols);
max@0:     
max@0:     
max@0:     const uword n_elem      = sample.n_elem;
max@0:     const eT* sample_mem  = sample.memptr();
max@0:           eT* r_mean_mem  = x.r_mean.memptr();
max@0:           eT* min_val_mem = x.min_val.memptr();
max@0:           eT* max_val_mem = x.max_val.memptr();
max@0:           
max@0:     
max@0:     for(uword i=0; i<n_elem; ++i)
max@0:       {
max@0:       const eT val = sample_mem[i];
max@0:       
max@0:       r_mean_mem[i]  = val;
max@0:       min_val_mem[i] = val;
max@0:       max_val_mem[i] = val;
max@0:       }
max@0:     }
max@0:   
max@0:   x.counter++;
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! update statistics to reflect new sample (version for complex numbers)
max@0: template<typename T>
max@0: inline
max@0: void
max@0: running_stat_vec_aux::update_stats(running_stat_vec< std::complex<T> >& x, const Mat<T>& sample)
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   const Mat< std::complex<T> > tmp = conv_to< Mat< std::complex<T> > >::from(sample);
max@0:   
max@0:   running_stat_vec_aux::update_stats(x, tmp);
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! alter statistics to reflect new sample (version for complex numbers)
max@0: template<typename T>
max@0: inline
max@0: void
max@0: running_stat_vec_aux::update_stats(running_stat_vec< std::complex<T> >& x, const Mat< std::complex<T> >& sample)
max@0:   {
max@0:   arma_extra_debug_sigprint();
max@0:   
max@0:   typedef typename std::complex<T> eT;
max@0:   
max@0:   const T N = x.counter.value();
max@0:   
max@0:   if(N > T(0))
max@0:     {
max@0:     arma_debug_assert_same_size(x.r_mean, sample, "running_stat_vec(): dimensionality mismatch");
max@0:     
max@0:     const uword n_elem           = sample.n_elem;
max@0:     const eT* sample_mem       = sample.memptr();
max@0:           eT* r_mean_mem       = x.r_mean.memptr();
max@0:            T* r_var_mem        = x.r_var.memptr();
max@0:           eT* min_val_mem      = x.min_val.memptr();
max@0:           eT* max_val_mem      = x.max_val.memptr();
max@0:            T* min_val_norm_mem = x.min_val_norm.memptr();
max@0:            T* max_val_norm_mem = x.max_val_norm.memptr();
max@0:     
max@0:     const T  N_plus_1   = x.counter.value_plus_1();
max@0:     const T  N_minus_1  = x.counter.value_minus_1();
max@0:     
max@0:     if(x.calc_cov == true)
max@0:       {
max@0:       Mat<eT>& tmp1 = x.tmp1;
max@0:       Mat<eT>& tmp2 = x.tmp2;
max@0:       
max@0:       tmp1 = sample - x.r_mean;
max@0:       
max@0:       if(sample.n_cols == 1)
max@0:         {
max@0:         tmp2 = arma::conj(tmp1)*strans(tmp1);
max@0:         }
max@0:       else
max@0:         {
max@0:         tmp2 = trans(tmp1)*tmp1;  //tmp2 = strans(conj(tmp1))*tmp1;
max@0:         }
max@0:       
max@0:       x.r_cov *= (N_minus_1/N);
max@0:       x.r_cov += tmp2 / N_plus_1;
max@0:       }
max@0:     
max@0:     
max@0:     for(uword i=0; i<n_elem; ++i)
max@0:       {
max@0:       const eT& val      = sample_mem[i];
max@0:       const  T  val_norm = std::norm(val);
max@0:       
max@0:       if(val_norm < min_val_norm_mem[i])
max@0:         {
max@0:         min_val_norm_mem[i] = val_norm;
max@0:         min_val_mem[i]      = val;
max@0:         }
max@0:       
max@0:       if(val_norm > max_val_norm_mem[i])
max@0:         {
max@0:         max_val_norm_mem[i] = val_norm;
max@0:         max_val_mem[i]      = val;
max@0:         }
max@0:       
max@0:       const eT& r_mean_val = r_mean_mem[i];
max@0:       
max@0:       r_var_mem[i] = N_minus_1/N * r_var_mem[i] + std::norm(val - r_mean_val)/N_plus_1;
max@0:       
max@0:       r_mean_mem[i] = r_mean_val + (val - r_mean_val)/N_plus_1;
max@0:       }
max@0:     
max@0:     }
max@0:   else
max@0:     {
max@0:     arma_debug_check( (sample.is_vec() == false), "running_stat_vec(): given sample is not a vector");
max@0:     
max@0:     x.r_mean.set_size(sample.n_rows, sample.n_cols);
max@0:     
max@0:     x.r_var.zeros(sample.n_rows, sample.n_cols);
max@0:     
max@0:     if(x.calc_cov == true)
max@0:       {
max@0:       x.r_cov.zeros(sample.n_elem, sample.n_elem);
max@0:       }
max@0:     
max@0:     x.min_val.set_size(sample.n_rows, sample.n_cols);
max@0:     x.max_val.set_size(sample.n_rows, sample.n_cols);
max@0:     
max@0:     x.min_val_norm.set_size(sample.n_rows, sample.n_cols);
max@0:     x.max_val_norm.set_size(sample.n_rows, sample.n_cols);
max@0:     
max@0:     
max@0:     const uword n_elem           = sample.n_elem;
max@0:     const eT* sample_mem       = sample.memptr();
max@0:           eT* r_mean_mem       = x.r_mean.memptr();
max@0:           eT* min_val_mem      = x.min_val.memptr();
max@0:           eT* max_val_mem      = x.max_val.memptr();
max@0:            T* min_val_norm_mem = x.min_val_norm.memptr();
max@0:            T* max_val_norm_mem = x.max_val_norm.memptr();
max@0:     
max@0:     for(uword i=0; i<n_elem; ++i)
max@0:       {
max@0:       const eT& val      = sample_mem[i];
max@0:       const  T  val_norm = std::norm(val);
max@0:       
max@0:       r_mean_mem[i]  = val;
max@0:       min_val_mem[i] = val;
max@0:       max_val_mem[i] = val;
max@0:       
max@0:       min_val_norm_mem[i] = val_norm;
max@0:       max_val_norm_mem[i] = val_norm;
max@0:       }
max@0:     }
max@0:   
max@0:   x.counter++;
max@0:   }
max@0: 
max@0: 
max@0: 
max@0: //! @}