source: trunk/lib/statistics/AveragerPairWeighted.h @ 490

// $Id: AveragerPairWeighted.h 490 2006-01-04 20:32:05Z peter $

#ifndef _theplu_statistics_averager_pair_weighted_
#define _theplu_statistics_averager_pair_weighted_

#include <c++_tools/statistics/AveragerWeighted.h>
#include <c++_tools/gslapi/vector.h>

#include <cmath>
//#include <utility>


namespace theplu{
namespace statistics{
  ///
  /// Class for taking care of mean and covariance of two variables in
  /// a weighted manner.
  ///
  /// <a href="../Statistics/index.html">Weighted Statistics document</a>
  ///
  /// If nothing else stated, each function fulfills the
  /// following:<br> <ul><li>Setting a weight to zero corresponds to
  /// removing the data point from the dataset.</li><li> Setting all
  /// weights to unity, the yields the same result as from
  /// corresponding function in AveragerPair.</li><li> Rescaling weights
  /// does not change the performance of the object.</li></ul>
  ///
  /// @see Averager AveragerWeighted AveragerPair
  ///
  class AveragerPairWeighted
  {
  public:

    AveragerPairWeighted()
      : wxy_(0), w_(0)
    {
    }

    ///
    /// Adding a pair of data points with value \a x and \a y, and
    /// their weights. If either of the weights are zero the addition
    /// is ignored
    ///
    void  add(const double x, const double y, 
              const double wx, const double wy);

    ///
    /// Adding pair of data and corresponding weight in vectors
    ///
    void add(const gslapi::vector& x, 
             const gslapi::vector& y, 
             const gslapi::vector& wx, 
             const gslapi::vector& wy);

    ///
    /// @brief Pearson correlation coefficient.
    ///
    /// @return \f$ \frac{\sum w_xw_y (x-m_x)(y-m_y)}{\sqrt{\sum
    /// w_xw_y (y-m_y)^2\sum w_xw_y (y-m_y)^2}} \f$ where m is
    /// calculated as \f$ m_x = \frac {\sum w_xw_yx}{\sum w} \f$
    ///
    inline double correlation(void) const 
    { return covariance() / ( x_.std()*y_.std() ); }
  
    ///
    /// \f$ \frac{\sum w_xw_y (x-m_x)(y-m_y)}{\sum w_xw_y} \f$ where m
    /// is calculated as \f$ m_x = \frac {\sum w_xw_yx}{\sum w} \f$
    ///
    inline double covariance(void) const { return sum_xy_centered()/sum_w(); }

    ///
    /// reset everything to zero
    ///
    inline void reset(void) { x_.reset(); y_.reset(); wxy_=0; w_=0; }

    ///
    /// @return \f$ \sum w_xw_y \f$
    ///
    inline double sum_w(void) const { return w_; }

    ///
    /// @return \f$ \sum w_xw_yxy \f$
    ///
    inline double sum_xy(void) const { return wxy_; }

    ///
    /// @return \f$ \sum w_xw_y (x-m_x)(y-m_y) \f$ where m is calculated as
    /// \f$ m_x = \frac {\sum w_xw_yx}{\sum w} \f$
    ///
    inline double sum_xy_centered(void) const 
    { return sum_xy() - x_.sum_wx()*y_.mean(); }

    ///
    /// @note the weights are calculated as \f$ w = w_x * w_y \f$
    ///
    /// @return AveragerWeighted for x
    ///
    inline const AveragerWeighted& x_averager(void) const { return x_; }

    ///
    /// @note the weights are calculated as \f$ w = w_x * w_y \f$
    ///
    /// @return AveragerWeighted for y
    ///
    inline const AveragerWeighted& y_averager(void) const { return y_; }

  private:
    AveragerWeighted x_; // weighted averager with w = w_x*w_y
    AveragerWeighted y_; // weighted averager with w = w_x*w_y
    double wxy_;
    double w_;

  };

}} // of namespace statistics and namespace theplu

#endif