criterion_normal_bhattacharyya.hpp

Go to the documentation of this file.

#ifndef FSTCRITERIONNORMALBHATTACHARYYA_H
#define FSTCRITERIONNORMALBHATTACHARYYA_H
 /* 
=========================================================================
Copyright and license:
  * FST4 software (with the exception of any bundled or externally linked
    third-party components) is Copyright (C) 2000-present Institute of
    Information Theory and Automation of the Czech Academy of Sciences (UTIA).
  * FST4 is free software: you can redistribute it and/or modify it under
    the terms of the GNU Affero General Public License as published by the
    Free Software Foundation, version 3. The full text of the license is in
    the file LICENSE in the distribution root and at
    https://www.gnu.org/licenses/agpl-3.0.html.
  * FST4 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 Affero General Public
    License for more details.
  * Closed-source use without the AGPL's copyleft obligations requires a
    commercial license from UTIA; in case of interest write to
    utia@utia.cas.cz. LICENSING.md in the distribution root explains both
    options, the contributor license agreement that every contribution
    requires, and the contacts (technical questions and pull requests:
    somol@utia.cas.cz, with "FST" in the subject line).
========================================================================== */
 
#include <memory>
#include <cassert>
#include <iostream>
#include <cmath>
#include "error.hpp"
#include "global.hpp"
#include "criterion_normal.hpp"
#include "indexed_vector.hpp"
#include "indexed_matrix.hpp"
 
/*============== Template parameter type naming conventions ==============
--------- Numeric types: -------------------------------------------------
DATATYPE - data sample values - usually real numbers (but may be integers
          in text processing etc.)
REALTYPE - must be real numbers - for representing intermediate results of 
          calculations like mean, covariance etc.
IDXTYPE - index values for enumeration of data samples - (nonnegative) integers, 
          extent depends on numbers of samples in data
DIMTYPE - index values for enumeration of features (dimensions), or classes (not 
          class sizes) - (nonnegative) integers, usually lower extent than IDXTYPE, 
          but be aware of expressions like _classes*_features*_features ! 
          in linearized representations of feature matrices for all classes
BINTYPE - feature selection marker type - represents ca. <10 different feature 
          states (selected, deselected, sel./desel. temporarily 1st nested loop, 2nd...)
RETURNTYPE - criterion value: real value, but may be extended in future to support 
          multiple values 
--------- Class types: ---------------------------------------------------
SUBSET       - class of class type Subset 
CLASSIFIER   - class implementing interface defined in abstract class Classifier 
EVALUATOR    - class implementing interface defined in abstract class Sequential_Step 
DISTANCE     - class implementing interface defined in abstract class Distance 
DATAACCESSOR - class implementing interface defined in abstract class Data_Accessor 
INTERVALCONTAINER - class of class type Data_Intervaller 
CONTAINER    - STL container of class type Data_Interval  
========================================================================== */
 
namespace FST {

template<class RETURNTYPE, typename DATATYPE, typename REALTYPE, typename IDXTYPE, typename DIMTYPE, class SUBSET, class DATAACCESSOR>
class Criterion_Normal_Bhattacharyya : public Clonable_Impl<Criterion_Normal_Bhattacharyya<RETURNTYPE,DATATYPE,REALTYPE,IDXTYPE,DIMTYPE,SUBSET,DATAACCESSOR>,Criterion_Normal<RETURNTYPE,DATATYPE,REALTYPE,IDXTYPE,DIMTYPE,SUBSET,DATAACCESSOR> > {
public:
        typedef Criterion_Normal<RETURNTYPE,DATATYPE,REALTYPE,IDXTYPE,DIMTYPE,SUBSET,DATAACCESSOR> parent; 
        typedef Clonable_Impl<Criterion_Normal_Bhattacharyya,parent> parent_impl; 
        typedef std::shared_ptr<DATAACCESSOR> PDataAccessor;  
        typedef std::shared_ptr<SUBSET> PSubset;              
        Criterion_Normal_Bhattacharyya() {notify("Criterion_Normal_Bhattacharyya constructor.");}
        virtual ~Criterion_Normal_Bhattacharyya() {notify("Criterion_Normal_Bhattacharyya destructor.");}
protected:
        // Template-method hooks: initialize()/evaluate() skeletons live in Criterion_Normal
        virtual const char* criterion_name() const override {return "Criterion_Normal_Bhattacharyya";}
        virtual void allocate_buffers() override {
                parent::alloc_matrix(_cov12);
                parent::alloc_matrix(_cov12_div2);
                parent::alloc_matrix(_invbuf);
                parent::alloc_matrix(_LUtemp,true);
        }
        virtual void update_derived_model() override {
                _cov12->copy_raw(parent::_model->get_cov()[0]);
                _cov12->add_raw(parent::_model->get_cov()[1]);
                _cov12_div2->copy_constmul_raw(*_cov12,0.5);
        }
        virtual bool evaluate_narrowed(RETURNTYPE &result, const PSubset sub) override;
private:
        friend parent_impl; 
        Criterion_Normal_Bhattacharyya(const Criterion_Normal_Bhattacharyya& cnb);
private:
        std::unique_ptr<Indexed_Matrix<DATATYPE,DIMTYPE,SUBSET> > _cov12; 
        std::unique_ptr<Indexed_Matrix<DATATYPE,DIMTYPE,SUBSET> > _cov12_div2; 
        std::unique_ptr<Indexed_Matrix<REALTYPE,DIMTYPE,SUBSET> > _invbuf; 
        std::unique_ptr<Indexed_Matrix<REALTYPE,DIMTYPE,SUBSET> > _LUtemp; 
};
 
template<class RETURNTYPE, typename DATATYPE, typename REALTYPE, typename IDXTYPE, typename DIMTYPE, class SUBSET, class DATAACCESSOR>
Criterion_Normal_Bhattacharyya<RETURNTYPE,DATATYPE,REALTYPE,IDXTYPE,DIMTYPE,SUBSET,DATAACCESSOR>::Criterion_Normal_Bhattacharyya(const Criterion_Normal_Bhattacharyya& cnb) :
        parent_impl(cnb)
{
        notify("Criterion_Normal_Bhattacharyya copy-constructor.");
        if(cnb._cov12) _cov12.reset(new Indexed_Matrix<DATATYPE,DIMTYPE,SUBSET>(*cnb._cov12));
        if(cnb._cov12_div2) _cov12_div2.reset(new Indexed_Matrix<DATATYPE,DIMTYPE,SUBSET>(*cnb._cov12_div2));
        if(cnb._invbuf) _invbuf.reset(new Indexed_Matrix<REALTYPE,DIMTYPE,SUBSET>(*cnb._invbuf));
        if(cnb._LUtemp) _LUtemp.reset(new Indexed_Matrix<REALTYPE,DIMTYPE,SUBSET>(*cnb._LUtemp));
}
 
template<class RETURNTYPE, typename DATATYPE, typename REALTYPE, typename IDXTYPE, typename DIMTYPE, class SUBSET, class DATAACCESSOR>
bool Criterion_Normal_Bhattacharyya<RETURNTYPE,DATATYPE,REALTYPE,IDXTYPE,DIMTYPE,SUBSET,DATAACCESSOR>::evaluate_narrowed(RETURNTYPE &result, const PSubset sub)
{ // (model and _meandif have been narrowed to sub by Criterion_Normal::evaluate)
        notify("Criterion_Normal_Bhattacharyya::evaluate_narrowed().");
        assert(_cov12);
        assert(_cov12_div2);
        assert(_invbuf);
        assert(_LUtemp);
        assert(parent::get_n()==_cov12->get_n());
        assert(parent::get_n()==_cov12_div2->get_n());
        assert(parent::get_n()<=_invbuf->get_n_max());
        assert(parent::get_n()<=_LUtemp->get_n_max());
 
        RETURNTYPE logdet1,logdet2,logdet3;
 
        _cov12->narrow_to(sub);
        _cov12_div2->narrow_to(sub);
 
        _invbuf->redim(parent::get_d());
        _LUtemp->redim(parent::get_d());
 
        // The covariance term ln( |S| / sqrt(|S0| |S1|) ) is formed in the LOG domain from
        // log_determinant(): the three determinants themselves are products of d LU pivots and
        // leave the double range (overflow to inf, or underflow to 0 and a bogus "singular"
        // report) long before the matrices become ill-conditioned -- a 100-feature subset of
        // features with variance 1e8 has |S| ~ 1e800. The log form depends only on the geometry,
        // not on the feature scale, and is exact where the plain product would still be finite.
        _cov12->LUdecompose(*_LUtemp);
        _cov12->invert(*_invbuf,*_LUtemp);
        parent::_model->get_cov()[0].LUdecompose(*_LUtemp);
        logdet1=parent::_model->get_cov()[0].log_determinant(*_LUtemp);
        parent::_model->get_cov()[1].LUdecompose(*_LUtemp);
        logdet2=parent::_model->get_cov()[1].log_determinant(*_LUtemp);
        _cov12_div2->LUdecompose(*_LUtemp);
        logdet3=_cov12_div2->log_determinant(*_LUtemp);
 
        // LIMITATION: a singular class covariance on the evaluated subset makes the criterion
        // undefined and is reported by throwing, not by returning false. It happens when the
        // subset contains constant, duplicate or linearly dependent features, or when d exceeds
        // the number of training samples of a class. Remedies: reduce the target subset size,
        // drop the constant/collinear features, use more training data, or pick a criterion that
        // does not invert per-class covariances (e.g. a wrapper criterion). A zero LU pivot gives
        // ln 0 = -inf here (LUdecompose() itself throws on an all-zero row).
        if(!std::isfinite(logdet1) || !std::isfinite(logdet2))
                throw FST::fst_error("Criterion_Normal_Bhattacharyya::evaluate: singular class covariance matrix on the evaluated subset (zero determinant)");
        RETURNTYPE right=0.5*(logdet3-0.5*(logdet1+logdet2));
        RETURNTYPE left=0.0;
        RETURNTYPE pom;
        
        for(DIMTYPE j=0;j<parent::get_d();j++)
        {
                pom=0.0;
                for(DIMTYPE i=0;i<parent::get_d();i++)
                {
                        pom+=parent::_meandif->at(i)*_invbuf->at_raw(i,j);
                }
                left+=pom*parent::_meandif->at(j);
        }
        result=0.25*left+right;
        return true;
}
 
} // namespace
#endif // FSTCRITERIONNORMALBHATTACHARYYA_H ///:~