STK_CategoricalBase.h

Go to the documentation of this file.

/*--------------------------------------------------------------------*/
/*     Copyright (C) 2004-2016 Serge Iovleff
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU Lesser General Public License as
    published by the Free Software Foundation; either version 2 of the
    License, or (at your option) any later version.
 
    This program 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 Lesser General Public License for more details.
 
    You should have received a copy of the GNU Lesser General Public
    License along with this program; if not, write to the
    Free Software Foundation, Inc.,
    59 Temple Place,
    Suite 330,
    Boston, MA 02111-1307
    USA
 
    Contact : S..._DOT_I..._AT_stkpp.org (see copyright for ...)
*/
 
/*
 * Project:  stkpp::Clustering
 * created on: Dec 4, 2013
 * Authors: Serge Iovleff
 **/
 
#ifndef STK_CATEGORICALBASE_H
#define STK_CATEGORICALBASE_H
 
#include "../STK_IMixtureDensity.h"
#include "../CategoricalModels/STK_CategoricalParameters.h"
#include <Arrays/include/STK_Array2DPoint.h>
 
namespace STK
{
 
template<class Derived>
class CategoricalBase: public IMixtureDensity<Derived >
{
  protected:
    typedef IMixtureDensity<Derived> Base;
    using Base::param_;
    using Base::p_data;
 
    CategoricalBase( int nbCluster): Base(nbCluster) {}
    CategoricalBase( CategoricalBase const& model)
                         : Base(model), modalities_(model.modalities_)
    {}
    ~CategoricalBase() {}
 
  public:
    inline PointXi const& nbModalities() const { return nbModalities_;}
    inline Range const& modalities() const { return modalities_;}
    inline Real proba(int k, int j, int l) const { return param_.proba(k,j,l);}
    inline CVectorX proba(int k, int j) const { return param_.proba(k,j);}
    void initializeModelImpl()
    {
      // compute the maximal number of modalities
      nbModalities_.resize(p_data()->cols());
      int amin = Arithmetic<int>::max(), amax = Arithmetic<int>::min();
      for (int j= p_data()->beginCols(); j < p_data()->endCols(); ++j)
      {
        int min = p_data()->col(j).minElt(), max = p_data()->col(j).maxElt();
        amin = std::min(amin, min); amax = std::max(amax, max);
        nbModalities_[j] = max-min+1;
      }
      // set range of the modalities
      modalities_ = _R(amin, amax);
      // resize vectors of probabilities
      param_.resize(modalities_,p_data()->cols());
#ifdef STK_MIXTURE_VERBOSE
      stk_cout << _T("In CatagoricalBase::initializeModelImpl. modalities_ = ")
               << modalities_ << _T("\n");
#endif
    }
    Real lnComponentProbability(int i, int k) const;
    template<class Weights>
    int impute(int i, int j, Weights const& tk) const;
    inline int rand(int i, int j, int k) const
    { return Law::Categorical::rand(proba(k,j));}
    template<class Array>
    void getParameters(Array& params) const;
    void writeParameters(CArrayXX const* p_tik, ostream& os) const;
 
  protected:
    PointXi nbModalities_;
    Range modalities_;
};
 
/* @return the value of the probability of the i-th sample in the k-th component.
 *  @param i,k indexes of the sample and of the component
 **/
template<class Derived>
inline Real CategoricalBase<Derived>::lnComponentProbability(int i, int k) const
{
  Real sum =0.;
  for (int j=p_data()->beginCols(); j<p_data()->endCols(); ++j)
  { // what to do if the probability is zero but a sample get this modality
    // for now, just ignore it (it's possible if tik_(i,k) == 0)
    Real prob= proba(k, j, p_data()->elt(i,j));
    if (prob) { sum += std::log(prob);}
   }
  return sum;
}
 
/* Implementation  */
template<class Derived>
template<class Weights>
int CategoricalBase<Derived>::impute(int i, int j, Weights const& tk) const
{
  int lmax = modalities_.begin();
  Real pmax = -Arithmetic<Real>::max();
  // compute for each modality the pondered probability of occurrence
  for (int l=modalities_.begin(); l< modalities_.end(); ++l)
  {
    Real p = 0.;
    for (int k= tk.begin(); k < tk.end(); ++k)
    { p += tk[k] * proba(k, j, l);}
 
    if (pmax < p) { pmax = p; lmax = l;}
  }
  return lmax;
}
 
/* This function can be used to write summary of parameters to the output stream.
 *  @param p_tik a constant pointer on the posterior probabilities
 *  @param os Stream where you want to write the summary of parameters.
 */
template<class Derived>
void CategoricalBase<Derived>::writeParameters(CArrayXX const* p_tik, ostream& os) const
{
  ArrayXX p(modalities(), p_data()->cols());
  for (int k= p_tik->beginCols(); k < p_tik->endCols(); ++k)
  {
    // store proba values in an array for a nice output
    for (int j= p.beginCols();  j < p.endCols(); ++j)
    {
      for (int l= modalities().begin(); l < modalities().end(); ++l)
      { p(l, j) = proba(k,j,l);}
    }
    os << _T("---> Component ") << k << _T("\n");
    os << _T("probabilities =\n") << p  << _T("\n");
  }
 
}
 
template<class Derived>
template<class Array>
void CategoricalBase<Derived>::getParameters(Array& params) const
{
    int nbCluster    = this->nbCluster();
    int nbModalities = modalities().size();
 
    params.resize(nbModalities * nbCluster, p_data()->cols());
    for (int k = 0; k < nbCluster; ++k)
    {
      for (int j = p_data()->beginCols(); j < p_data()->endCols(); ++j)
      {
        for (int l = 0; l < nbModalities; ++l)
        { params(baseIdx+k * nbModalities + l, j) = proba(baseIdx+k, j, modalities().begin() + l);}
      }
    }
}
 
 
} // namespace STK
 
#endif /* STK_CategoricalBASE_H */