source: trunk/yat/classifier/IGP.h @ 3562

#ifndef _theplu_yat_classifier_igp_
#define _theplu_yat_classifier_igp_

// $Id$

/*
  Copyright (C) 2006 Jari Häkkinen, Markus Ringnér
  Copyright (C) 2007, 2008 Jari Häkkinen, Peter Johansson, Markus Ringnér
  Copyright (C) 2009, 2010, 2014, 2017 Peter Johansson

  This file is part of the yat library, http://dev.thep.lu.se/yat

  The yat library is free software; you can redistribute it and/or
  modify it under the terms of the GNU General Public License as
  published by the Free Software Foundation; either version 3 of the
  License, or (at your option) any later version.

  The yat library is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  General Public License for more details.

  You should have received a copy of the GNU General Public License
  along with yat. If not, see <http://www.gnu.org/licenses/>.
*/

#include "MatrixLookup.h"
#include "Target.h"
#include "yat/utility/concept_check.h"
#include "yat/utility/Matrix.h"
#include "yat/utility/Vector.h"
#include "yat/utility/yat_assert.h"

#include <boost/concept_check.hpp>

#include <cmath>
#include <limits>
#include <stdexcept>

namespace theplu {
namespace yat {
namespace classifier {

  class Target;
  class MatrixLookup;

  /**
     \brief Class for In Group Proportions (IGP)

     IGP is defined to be the proportion of samples in a group whose
     nearest neighbours are also in the same group.

     \see <a HREF="
     http://biostatistics.oxfordjournals.org/cgi/content/abstract/kxj029v1">
     Kapp and Tibshirani, Biostatistics (2006)</a>.

     Distance should model concept \ref concept_distance (but support
     for weighted iterators is not required).

     \note Distance must be symmetric, i.e., Distance(x,y) == Distance(y,x).
  */
  template <typename Distance>
  class IGP
  {

  public:
    ///
    /// Constructor taking the training data and the target vector and
    /// as input.
    ///
    IGP(const MatrixLookup&, const Target&);


    ///
    /// Constructor taking the training data, the target vector and
    /// the distance measure as input.
    ///
    IGP(const MatrixLookup&, const Target&, const Distance&);

    ///
    /// Destrucutor
    ///
    virtual ~IGP();

    ///
    /// @return the IGP score for each class as elements in a vector.
    ///
    const utility::Vector& score(void) const;


  private:
    void calculate();

    utility::Vector igp_;
    Distance distance_;

    const MatrixLookup& matrix_;
    const Target& target_;
  };


  // templates

  template <typename Distance>
  IGP<Distance>::IGP(const MatrixLookup& data, const Target& target)
    : matrix_(data), target_(target)
  {
    BOOST_CONCEPT_ASSERT((utility::DistanceConcept<Distance>));
    calculate();
  }

  template <typename Distance>
  IGP<Distance>::IGP(const MatrixLookup& data, const Target& target,
                     const Distance& dist)
    : matrix_(data), target_(target), distance_(dist)
  {
    BOOST_CONCEPT_ASSERT((utility::DistanceConcept<Distance>));
    calculate();
  }


  template <typename Distance>
  IGP<Distance>::~IGP()
  {
  }


  template <typename Distance>
  void IGP<Distance>::calculate()
  {
    YAT_ASSERT(target_.size()==matrix_.columns());

    // Calculate IGP for each class
    igp_.resize(target_.nof_classes(), 0.0);

    // calculate distances
    utility::Matrix dist(matrix_.columns(), matrix_.columns());
    for (size_t i=0; i<dist.rows(); ++i)
      for (size_t j=i+1; j<dist.rows(); ++j) {
        dist(i,j) = dist(j,i) = distance_(matrix_.begin_column(i),
                                          matrix_.end_column(i),
                                          matrix_.begin_column(j));
      }

    // find nearest neigbour for each sample
    for(size_t i=0; i<target_.size(); i++) {
      size_t neighbor=i;
      double mindist=std::numeric_limits<double>::max();
      for(size_t j=0; j<target_.size(); j++) {
        if (i==j) // avoid self-self comparison
          continue;
        if(dist(i,j)<mindist) {
          mindist=dist(i,j);
          neighbor=j;
        }
      }
      if(target_(i)==target_(neighbor))
        igp_(target_(i))++;

    }
    for(size_t i=0; i<target_.nof_classes(); i++) {
      igp_(i)/=static_cast<double>(target_.size(i));
    }
  }


  template <typename Distance>
  const utility::Vector& IGP<Distance>::score(void) const
  {
    return igp_;
  }

}}} // of namespace classifier, yat, and theplu

#endif