Chris@49: // Copyright (C) 2009-2011 NICTA (www.nicta.com.au)
Chris@49: // Copyright (C) 2009-2011 Conrad Sanderson
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 running_stat_vec
Chris@49: //! @{
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: template<typename eT>
Chris@49: running_stat_vec<eT>::~running_stat_vec()
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint_this(this);
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: template<typename eT>
Chris@49: running_stat_vec<eT>::running_stat_vec(const bool in_calc_cov)
Chris@49:   : calc_cov(in_calc_cov)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint_this(this);
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: template<typename eT>
Chris@49: running_stat_vec<eT>::running_stat_vec(const running_stat_vec<eT>& in_rsv)
Chris@49:   : calc_cov    (in_rsv.calc_cov)
Chris@49:   , counter     (in_rsv.counter)
Chris@49:   , r_mean      (in_rsv.r_mean)
Chris@49:   , r_var       (in_rsv.r_var)
Chris@49:   , r_cov       (in_rsv.r_cov)
Chris@49:   , min_val     (in_rsv.min_val)
Chris@49:   , max_val     (in_rsv.max_val)
Chris@49:   , min_val_norm(in_rsv.min_val_norm)
Chris@49:   , max_val_norm(in_rsv.max_val_norm)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint_this(this);
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: template<typename eT>
Chris@49: const running_stat_vec<eT>&
Chris@49: running_stat_vec<eT>::operator=(const running_stat_vec<eT>& in_rsv)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   access::rw(calc_cov) = in_rsv.calc_cov;
Chris@49:   
Chris@49:   counter      = in_rsv.counter;
Chris@49:   r_mean       = in_rsv.r_mean;
Chris@49:   r_var        = in_rsv.r_var;
Chris@49:   r_cov        = in_rsv.r_cov;
Chris@49:   min_val      = in_rsv.min_val;
Chris@49:   max_val      = in_rsv.max_val;
Chris@49:   min_val_norm = in_rsv.min_val_norm;
Chris@49:   max_val_norm = in_rsv.max_val_norm;
Chris@49:   
Chris@49:   return *this;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! update statistics to reflect new sample
Chris@49: template<typename eT>
Chris@49: template<typename T1>
Chris@49: arma_hot
Chris@49: inline
Chris@49: void
Chris@49: running_stat_vec<eT>::operator() (const Base<typename get_pod_type<eT>::result, T1>& X)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   //typedef typename get_pod_type<eT>::result T;
Chris@49:   
Chris@49:   const unwrap<T1>        tmp(X.get_ref());
Chris@49:   const Mat<eT>& sample = tmp.M;
Chris@49:   
Chris@49:   if( sample.is_empty() )
Chris@49:     {
Chris@49:     return;
Chris@49:     }
Chris@49:   
Chris@49:   if( sample.is_finite() == false )
Chris@49:     {
Chris@49:     arma_warn(true, "running_stat_vec: sample ignored as it has non-finite elements");
Chris@49:     return;
Chris@49:     }
Chris@49:   
Chris@49:   running_stat_vec_aux::update_stats(*this, sample);
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! update statistics to reflect new sample (version for complex numbers)
Chris@49: template<typename eT>
Chris@49: template<typename T1>
Chris@49: arma_hot
Chris@49: inline
Chris@49: void
Chris@49: running_stat_vec<eT>::operator() (const Base<std::complex<typename get_pod_type<eT>::result>, T1>& X)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   //typedef typename std::complex<typename get_pod_type<eT>::result> eT;
Chris@49:   
Chris@49:   const unwrap<T1>        tmp(X.get_ref());
Chris@49:   const Mat<eT>& sample = tmp.M;
Chris@49:   
Chris@49:   if( sample.is_empty() )
Chris@49:     {
Chris@49:     return;
Chris@49:     }
Chris@49:   
Chris@49:   if( sample.is_finite() == false )
Chris@49:     {
Chris@49:     arma_warn(true, "running_stat_vec: sample ignored as it has non-finite elements");
Chris@49:     return;
Chris@49:     }
Chris@49:   
Chris@49:   running_stat_vec_aux::update_stats(*this, sample);
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! set all statistics to zero
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: void
Chris@49: running_stat_vec<eT>::reset()
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   counter.reset();
Chris@49:   
Chris@49:   r_mean.reset();
Chris@49:   r_var.reset();
Chris@49:   r_cov.reset();
Chris@49:   
Chris@49:   min_val.reset();
Chris@49:   max_val.reset();
Chris@49:   
Chris@49:   min_val_norm.reset();
Chris@49:   max_val_norm.reset();
Chris@49:   
Chris@49:   r_var_dummy.reset();
Chris@49:   r_cov_dummy.reset();
Chris@49:   
Chris@49:   tmp1.reset();
Chris@49:   tmp2.reset();
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! mean or average value
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: const Mat<eT>&
Chris@49: running_stat_vec<eT>::mean() const
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   return r_mean;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! variance
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: const Mat<typename get_pod_type<eT>::result>&
Chris@49: running_stat_vec<eT>::var(const uword norm_type)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   const T N = counter.value();
Chris@49:   
Chris@49:   if(N > T(1))
Chris@49:     {
Chris@49:     if(norm_type == 0)
Chris@49:       {
Chris@49:       return r_var;
Chris@49:       }
Chris@49:     else
Chris@49:       {
Chris@49:       const T N_minus_1 = counter.value_minus_1();
Chris@49:       
Chris@49:       r_var_dummy = (N_minus_1/N) * r_var;
Chris@49:       
Chris@49:       return r_var_dummy;
Chris@49:       }
Chris@49:     }
Chris@49:   else
Chris@49:     {
Chris@49:     r_var_dummy.zeros(r_mean.n_rows, r_mean.n_cols);
Chris@49:     
Chris@49:     return r_var_dummy;
Chris@49:     }
Chris@49:   
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! standard deviation
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: Mat<typename get_pod_type<eT>::result>
Chris@49: running_stat_vec<eT>::stddev(const uword norm_type) const
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   const T N = counter.value();
Chris@49:   
Chris@49:   if(N > T(1))
Chris@49:     {
Chris@49:     if(norm_type == 0)
Chris@49:       {
Chris@49:       return sqrt(r_var);
Chris@49:       }
Chris@49:     else
Chris@49:       {
Chris@49:       const T N_minus_1 = counter.value_minus_1();
Chris@49:       
Chris@49:       return sqrt( (N_minus_1/N) * r_var );
Chris@49:       }
Chris@49:     }
Chris@49:   else
Chris@49:     {
Chris@49:     return Mat<T>();
Chris@49:     }
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! covariance
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: const Mat<eT>&
Chris@49: running_stat_vec<eT>::cov(const uword norm_type)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   if(calc_cov == true)
Chris@49:     {
Chris@49:     const T N = counter.value();
Chris@49:     
Chris@49:     if(N > T(1))
Chris@49:       {
Chris@49:       if(norm_type == 0)
Chris@49:         {
Chris@49:         return r_cov;
Chris@49:         }
Chris@49:       else
Chris@49:         {
Chris@49:         const T N_minus_1 = counter.value_minus_1();
Chris@49:         
Chris@49:         r_cov_dummy = (N_minus_1/N) * r_cov;
Chris@49:         
Chris@49:         return r_cov_dummy;
Chris@49:         }
Chris@49:       }
Chris@49:     else
Chris@49:       {
Chris@49:       r_cov_dummy.zeros(r_mean.n_rows, r_mean.n_cols);
Chris@49:       
Chris@49:       return r_cov_dummy;
Chris@49:       }
Chris@49:     }
Chris@49:   else
Chris@49:     {
Chris@49:     r_cov_dummy.reset();
Chris@49:     
Chris@49:     return r_cov_dummy;
Chris@49:     }
Chris@49:   
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! vector with minimum values
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: const Mat<eT>&
Chris@49: running_stat_vec<eT>::min() const
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   return min_val;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! vector with maximum values
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: const Mat<eT>&
Chris@49: running_stat_vec<eT>::max() const
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   return max_val;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! number of samples so far
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: typename get_pod_type<eT>::result
Chris@49: running_stat_vec<eT>::count() const
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   return counter.value();
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! update statistics to reflect new sample
Chris@49: template<typename eT>
Chris@49: inline
Chris@49: void
Chris@49: running_stat_vec_aux::update_stats(running_stat_vec<eT>& x, const Mat<eT>& sample)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   typedef typename running_stat_vec<eT>::T T;
Chris@49:   
Chris@49:   const T N = x.counter.value();
Chris@49:   
Chris@49:   if(N > T(0))
Chris@49:     {
Chris@49:     arma_debug_assert_same_size(x.r_mean, sample, "running_stat_vec(): dimensionality mismatch");
Chris@49:     
Chris@49:     const uword n_elem      = sample.n_elem;
Chris@49:     const eT*   sample_mem  = sample.memptr();
Chris@49:           eT*   r_mean_mem  = x.r_mean.memptr();
Chris@49:            T*   r_var_mem   = x.r_var.memptr();
Chris@49:           eT*   min_val_mem = x.min_val.memptr();
Chris@49:           eT*   max_val_mem = x.max_val.memptr();
Chris@49:     
Chris@49:     const T  N_plus_1   = x.counter.value_plus_1();
Chris@49:     const T  N_minus_1  = x.counter.value_minus_1();
Chris@49:     
Chris@49:     if(x.calc_cov == true)
Chris@49:       {
Chris@49:       Mat<eT>& tmp1 = x.tmp1;
Chris@49:       Mat<eT>& tmp2 = x.tmp2;
Chris@49:       
Chris@49:       tmp1 = sample - x.r_mean;
Chris@49:       
Chris@49:       if(sample.n_cols == 1)
Chris@49:         {
Chris@49:         tmp2 = tmp1*trans(tmp1);
Chris@49:         }
Chris@49:       else
Chris@49:         {
Chris@49:         tmp2 = trans(tmp1)*tmp1;
Chris@49:         }
Chris@49:       
Chris@49:       x.r_cov *= (N_minus_1/N);
Chris@49:       x.r_cov += tmp2 / N_plus_1;
Chris@49:       }
Chris@49:     
Chris@49:     
Chris@49:     for(uword i=0; i<n_elem; ++i)
Chris@49:       {
Chris@49:       const eT val = sample_mem[i];
Chris@49:       
Chris@49:       if(val < min_val_mem[i])
Chris@49:         {
Chris@49:         min_val_mem[i] = val;
Chris@49:         }
Chris@49:       
Chris@49:       if(val > max_val_mem[i])
Chris@49:         {
Chris@49:         max_val_mem[i] = val;
Chris@49:         }
Chris@49:         
Chris@49:       const eT r_mean_val = r_mean_mem[i];
Chris@49:       const eT tmp        = val - r_mean_val;
Chris@49:     
Chris@49:       r_var_mem[i] = N_minus_1/N * r_var_mem[i] + (tmp*tmp)/N_plus_1;
Chris@49:       
Chris@49:       r_mean_mem[i] = r_mean_val + (val - r_mean_val)/N_plus_1;
Chris@49:       }
Chris@49:     }
Chris@49:   else
Chris@49:     {
Chris@49:     arma_debug_check( (sample.is_vec() == false), "running_stat_vec(): given sample is not a vector");
Chris@49:     
Chris@49:     x.r_mean.set_size(sample.n_rows, sample.n_cols);
Chris@49:     
Chris@49:     x.r_var.zeros(sample.n_rows, sample.n_cols);
Chris@49:     
Chris@49:     if(x.calc_cov == true)
Chris@49:       {
Chris@49:       x.r_cov.zeros(sample.n_elem, sample.n_elem);
Chris@49:       }
Chris@49:     
Chris@49:     x.min_val.set_size(sample.n_rows, sample.n_cols);
Chris@49:     x.max_val.set_size(sample.n_rows, sample.n_cols);
Chris@49:     
Chris@49:     
Chris@49:     const uword n_elem      = sample.n_elem;
Chris@49:     const eT*   sample_mem  = sample.memptr();
Chris@49:           eT*   r_mean_mem  = x.r_mean.memptr();
Chris@49:           eT*   min_val_mem = x.min_val.memptr();
Chris@49:           eT*   max_val_mem = x.max_val.memptr();
Chris@49:           
Chris@49:     
Chris@49:     for(uword i=0; i<n_elem; ++i)
Chris@49:       {
Chris@49:       const eT val = sample_mem[i];
Chris@49:       
Chris@49:       r_mean_mem[i]  = val;
Chris@49:       min_val_mem[i] = val;
Chris@49:       max_val_mem[i] = val;
Chris@49:       }
Chris@49:     }
Chris@49:   
Chris@49:   x.counter++;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! update statistics to reflect new sample (version for complex numbers)
Chris@49: template<typename T>
Chris@49: inline
Chris@49: void
Chris@49: running_stat_vec_aux::update_stats(running_stat_vec< std::complex<T> >& x, const Mat<T>& sample)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   const Mat< std::complex<T> > tmp = conv_to< Mat< std::complex<T> > >::from(sample);
Chris@49:   
Chris@49:   running_stat_vec_aux::update_stats(x, tmp);
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! alter statistics to reflect new sample (version for complex numbers)
Chris@49: template<typename T>
Chris@49: inline
Chris@49: void
Chris@49: running_stat_vec_aux::update_stats(running_stat_vec< std::complex<T> >& x, const Mat< std::complex<T> >& sample)
Chris@49:   {
Chris@49:   arma_extra_debug_sigprint();
Chris@49:   
Chris@49:   typedef typename std::complex<T> eT;
Chris@49:   
Chris@49:   const T N = x.counter.value();
Chris@49:   
Chris@49:   if(N > T(0))
Chris@49:     {
Chris@49:     arma_debug_assert_same_size(x.r_mean, sample, "running_stat_vec(): dimensionality mismatch");
Chris@49:     
Chris@49:     const uword n_elem           = sample.n_elem;
Chris@49:     const eT*   sample_mem       = sample.memptr();
Chris@49:           eT*   r_mean_mem       = x.r_mean.memptr();
Chris@49:            T*   r_var_mem        = x.r_var.memptr();
Chris@49:           eT*   min_val_mem      = x.min_val.memptr();
Chris@49:           eT*   max_val_mem      = x.max_val.memptr();
Chris@49:            T*   min_val_norm_mem = x.min_val_norm.memptr();
Chris@49:            T*   max_val_norm_mem = x.max_val_norm.memptr();
Chris@49:     
Chris@49:     const T  N_plus_1   = x.counter.value_plus_1();
Chris@49:     const T  N_minus_1  = x.counter.value_minus_1();
Chris@49:     
Chris@49:     if(x.calc_cov == true)
Chris@49:       {
Chris@49:       Mat<eT>& tmp1 = x.tmp1;
Chris@49:       Mat<eT>& tmp2 = x.tmp2;
Chris@49:       
Chris@49:       tmp1 = sample - x.r_mean;
Chris@49:       
Chris@49:       if(sample.n_cols == 1)
Chris@49:         {
Chris@49:         tmp2 = arma::conj(tmp1)*strans(tmp1);
Chris@49:         }
Chris@49:       else
Chris@49:         {
Chris@49:         tmp2 = trans(tmp1)*tmp1;  //tmp2 = strans(conj(tmp1))*tmp1;
Chris@49:         }
Chris@49:       
Chris@49:       x.r_cov *= (N_minus_1/N);
Chris@49:       x.r_cov += tmp2 / N_plus_1;
Chris@49:       }
Chris@49:     
Chris@49:     
Chris@49:     for(uword i=0; i<n_elem; ++i)
Chris@49:       {
Chris@49:       const eT& val      = sample_mem[i];
Chris@49:       const  T  val_norm = std::norm(val);
Chris@49:       
Chris@49:       if(val_norm < min_val_norm_mem[i])
Chris@49:         {
Chris@49:         min_val_norm_mem[i] = val_norm;
Chris@49:         min_val_mem[i]      = val;
Chris@49:         }
Chris@49:       
Chris@49:       if(val_norm > max_val_norm_mem[i])
Chris@49:         {
Chris@49:         max_val_norm_mem[i] = val_norm;
Chris@49:         max_val_mem[i]      = val;
Chris@49:         }
Chris@49:       
Chris@49:       const eT& r_mean_val = r_mean_mem[i];
Chris@49:       
Chris@49:       r_var_mem[i] = N_minus_1/N * r_var_mem[i] + std::norm(val - r_mean_val)/N_plus_1;
Chris@49:       
Chris@49:       r_mean_mem[i] = r_mean_val + (val - r_mean_val)/N_plus_1;
Chris@49:       }
Chris@49:     
Chris@49:     }
Chris@49:   else
Chris@49:     {
Chris@49:     arma_debug_check( (sample.is_vec() == false), "running_stat_vec(): given sample is not a vector");
Chris@49:     
Chris@49:     x.r_mean.set_size(sample.n_rows, sample.n_cols);
Chris@49:     
Chris@49:     x.r_var.zeros(sample.n_rows, sample.n_cols);
Chris@49:     
Chris@49:     if(x.calc_cov == true)
Chris@49:       {
Chris@49:       x.r_cov.zeros(sample.n_elem, sample.n_elem);
Chris@49:       }
Chris@49:     
Chris@49:     x.min_val.set_size(sample.n_rows, sample.n_cols);
Chris@49:     x.max_val.set_size(sample.n_rows, sample.n_cols);
Chris@49:     
Chris@49:     x.min_val_norm.set_size(sample.n_rows, sample.n_cols);
Chris@49:     x.max_val_norm.set_size(sample.n_rows, sample.n_cols);
Chris@49:     
Chris@49:     
Chris@49:     const uword n_elem           = sample.n_elem;
Chris@49:     const eT*   sample_mem       = sample.memptr();
Chris@49:           eT*   r_mean_mem       = x.r_mean.memptr();
Chris@49:           eT*   min_val_mem      = x.min_val.memptr();
Chris@49:           eT*   max_val_mem      = x.max_val.memptr();
Chris@49:            T*   min_val_norm_mem = x.min_val_norm.memptr();
Chris@49:            T*   max_val_norm_mem = x.max_val_norm.memptr();
Chris@49:     
Chris@49:     for(uword i=0; i<n_elem; ++i)
Chris@49:       {
Chris@49:       const eT& val      = sample_mem[i];
Chris@49:       const  T  val_norm = std::norm(val);
Chris@49:       
Chris@49:       r_mean_mem[i]  = val;
Chris@49:       min_val_mem[i] = val;
Chris@49:       max_val_mem[i] = val;
Chris@49:       
Chris@49:       min_val_norm_mem[i] = val_norm;
Chris@49:       max_val_norm_mem[i] = val_norm;
Chris@49:       }
Chris@49:     }
Chris@49:   
Chris@49:   x.counter++;
Chris@49:   }
Chris@49: 
Chris@49: 
Chris@49: 
Chris@49: //! @}