classifier.hpp

Go to the documentation of this file.

#ifndef FSTCLASSIFIER_H
#define FSTCLASSIFIER_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 <vector>
#include <utility>
#include <cassert>
#include "clonable.hpp"
#include "error.hpp"
#include "global.hpp"
#include "wrapper_metrics.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 DIMTYPE, class SUBSET, class DATAACCESSOR>
class Classifier : public Clonable { // abstract class
public:
        typedef std::shared_ptr<DATAACCESSOR> PDataAccessor;      
        typedef std::shared_ptr<SUBSET> const PSubset;            
        typedef typename DATAACCESSOR::PPattern PPattern;         
        Classifier() {}
        virtual ~Classifier() {}

        virtual bool classify(DIMTYPE &cls, const PPattern &pattern)  =0;
        virtual bool train(const PDataAccessor da, const PSubset sub) =0; 
        virtual bool test(RETURNTYPE &result, const PDataAccessor da);

        virtual bool classify_scores(std::vector<RETURNTYPE> &scores, const PPattern &pattern) {(void)scores; (void)pattern; return false;}

        virtual bool test_confusion(std::vector<RETURNTYPE> &confusion, DIMTYPE &classes, const PDataAccessor da);

        virtual bool test_scores(std::vector<std::pair<RETURNTYPE,DIMTYPE>> &scores, const PDataAccessor da);

        virtual bool test_scores_matrix(std::vector<RETURNTYPE> &scores, std::vector<DIMTYPE> &labels, DIMTYPE &classes, const PDataAccessor da);

        virtual bool test_metric(RETURNTYPE &result, const PDataAccessor da, const Metric_Spec<RETURNTYPE> &spec);
 
        Classifier* clone() const override =0;
        Classifier* sharing_clone() const override {throw fst_error("Classifier::sharing_clone() not supported, use Classifier::clone() instead.");}
        Classifier* stateless_clone() const override {throw fst_error("Classifier::stateless_clone() not supported, use Classifier::clone() instead.");}
};
 
template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class DATAACCESSOR>
bool Classifier<RETURNTYPE,DIMTYPE,SUBSET,DATAACCESSOR>::test(RETURNTYPE &result, const PDataAccessor da)
{
        // the canonical accuracy loop (see test_confusion below for the setClass re-issue note)
        notify("Classifier::test().");
        assert(da);
        typename DATAACCESSOR::PPattern p;
        unsigned long long count=0, correct=0;
        DIMTYPE clstmp;
        const DIMTYPE stride=da->getNoOfFeatures();
        for(DIMTYPE c_test=0;c_test<da->getNoOfClasses();c_test++)
        {
                da->setClass(c_test);
                for(const auto &blk : da->blocks(TEST)) { p=blk.data; const auto s=blk.size;
                        for(auto i=decltype(s){0};i<s;i++)
                        {
                                if(!classify(clstmp,&p[(std::size_t)i*stride])) return false;
                                if(clstmp==c_test) correct++;
                                count++;
                        }
                        da->setClass(c_test); // necessary for the case of da pointing to the same object as the classifier's training accessor
                }
        }
        if(count==0) return false;
        result=(RETURNTYPE)correct/(RETURNTYPE)count;
        return true;
}
 
template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class DATAACCESSOR>
bool Classifier<RETURNTYPE,DIMTYPE,SUBSET,DATAACCESSOR>::test_confusion(std::vector<RETURNTYPE> &confusion, DIMTYPE &classes, const PDataAccessor da)
{
        // the canonical test() loop, tallying a confusion matrix instead of correct/count.
        // NOTE: the re-issued setClass() after each block is necessary for classifiers whose
        //       classify() iterates a data accessor that may be the same instance (kNN)
        assert(da);
        classes=da->getNoOfClasses();
        if(classes<2) return false;
        confusion.assign((size_t)classes*(size_t)classes,(RETURNTYPE)0.0);
        typename DATAACCESSOR::PPattern p;
        unsigned long long count=0;
        DIMTYPE clstmp;
        const DIMTYPE stride=da->getNoOfFeatures();
        for(DIMTYPE c_test=0;c_test<classes;c_test++)
        {
                da->setClass(c_test);
                for(const auto &blk : da->blocks(TEST)) { p=blk.data; const auto s=blk.size;
                        for(auto i=decltype(s){0};i<s;i++)
                        {
                                if(!classify(clstmp,&p[(std::size_t)i*stride])) return false;
                                if(clstmp>=classes) clstmp=(c_test+1)%classes; // "wrong class" convention (kNN, no neighbours): deterministic error column, keeps the true-class row total intact
                                confusion[(size_t)c_test*(size_t)classes+(size_t)clstmp]+=1.0;
                                count++;
                        }
                        da->setClass(c_test); // necessary for the case of da pointing to the same object as the classifier's training accessor
                }
        }
        return count>0;
}
 
template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class DATAACCESSOR>
bool Classifier<RETURNTYPE,DIMTYPE,SUBSET,DATAACCESSOR>::test_scores(std::vector<std::pair<RETURNTYPE,DIMTYPE>> &scores, const PDataAccessor da)
{
        // same canonical loop, collecting (positive-class score, true class) pairs (two-class data)
        assert(da);
        const DIMTYPE classes=da->getNoOfClasses();
        if(classes!=2) return false;
        scores.clear();
        std::vector<RETURNTYPE> sc;
        typename DATAACCESSOR::PPattern p;
        const DIMTYPE stride=da->getNoOfFeatures();
        for(DIMTYPE c_test=0;c_test<classes;c_test++)
        {
                da->setClass(c_test);
                for(const auto &blk : da->blocks(TEST)) { p=blk.data; const auto s=blk.size;
                        for(auto i=decltype(s){0};i<s;i++)
                        {
                                if(!classify_scores(sc,&p[(std::size_t)i*stride]) || sc.size()<2) return false;
                                scores.push_back(std::make_pair(sc[1]-sc[0],c_test));
                        }
                        da->setClass(c_test); // necessary for the case of da pointing to the same object as the classifier's training accessor
                }
        }
        return !scores.empty();
}
 
template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class DATAACCESSOR>
bool Classifier<RETURNTYPE,DIMTYPE,SUBSET,DATAACCESSOR>::test_scores_matrix(std::vector<RETURNTYPE> &scores, std::vector<DIMTYPE> &labels, DIMTYPE &classes, const PDataAccessor da)
{
        // same canonical loop as test_scores, keeping every class's score column
        assert(da);
        classes=da->getNoOfClasses();
        if(classes<2) return false;
        scores.clear(); labels.clear();
        std::vector<RETURNTYPE> sc;
        typename DATAACCESSOR::PPattern p;
        const DIMTYPE stride=da->getNoOfFeatures();
        for(DIMTYPE c_test=0;c_test<classes;c_test++)
        {
                da->setClass(c_test);
                for(const auto &blk : da->blocks(TEST)) { p=blk.data; const auto s=blk.size;
                        for(auto i=decltype(s){0};i<s;i++)
                        {
                                if(!classify_scores(sc,&p[(std::size_t)i*stride]) || sc.size()<(size_t)classes) return false;
                                for(DIMTYPE c=0;c<classes;c++) scores.push_back(sc[c]);
                                labels.push_back(c_test);
                        }
                        da->setClass(c_test); // necessary for the case of da pointing to the same object as the classifier's training accessor
                }
        }
        return !labels.empty();
}
 
template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class DATAACCESSOR>
bool Classifier<RETURNTYPE,DIMTYPE,SUBSET,DATAACCESSOR>::test_metric(RETURNTYPE &result, const PDataAccessor da, const Metric_Spec<RETURNTYPE> &spec)
{
        if(spec.is_default()) return test(result,da); // plain accuracy, served by test() itself
        if(spec.is_regression_metric()) return false; // regression metrics apply to Regressor stacks only
        if(spec.metric==Wrapper_Metric::auc) {
                if(da->getNoOfClasses()==2) { // two classes: the exact binary AUC
                        std::vector<std::pair<RETURNTYPE,DIMTYPE>> scores;
                        if(!test_scores(scores,da)) return false;
                        return auc_from_scores<RETURNTYPE,DIMTYPE>(result,scores);
                }
                std::vector<RETURNTYPE> scores; // K>2: one-vs-rest macro AUC
                std::vector<DIMTYPE> labels;
                DIMTYPE classes;
                if(!test_scores_matrix(scores,labels,classes,da)) return false;
                return auc_macro_from_scores<RETURNTYPE,DIMTYPE>(result,scores,labels,classes);
        }
        std::vector<RETURNTYPE> confusion;
        DIMTYPE classes;
        if(!test_confusion(confusion,classes,da)) return false;
        return metric_from_confusion<RETURNTYPE,DIMTYPE>(result,spec.metric,confusion,classes,spec.class_weights,(DIMTYPE)spec.f1_positive);
}
 
} // namespace
#endif // FSTCLASSIFIER_H ///:~