search_seq_sfs.hpp

Go to the documentation of this file.

#ifndef FSTSEARCHSEQSFS_H
#define FSTSEARCHSEQSFS_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 <sstream>
#include "error.hpp"
#include "global.hpp"
#include "stopwatch.hpp"
#include "search_seq.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 CRITERION, class EVALUATOR>
class Search_SFS : public Search_Sequential<RETURNTYPE,DIMTYPE,SUBSET,CRITERION,EVALUATOR>{ 
public:
        typedef Search_Sequential<RETURNTYPE,DIMTYPE,SUBSET,CRITERION,EVALUATOR> parent; 
        typedef typename parent::parent grandparent; 
        typedef std::shared_ptr<EVALUATOR> PEvaluator; 
        typedef std::shared_ptr<CRITERION> PCriterion; 
        typedef std::shared_ptr<SUBSET> PSubset; 
        Search_SFS(const PEvaluator evaluator) : Search_Sequential<RETURNTYPE,DIMTYPE,SUBSET,CRITERION,EVALUATOR>(evaluator) {_direction=FORWARD; notify("Search_SFS constructor.");} 
        virtual ~Search_SFS() {notify("Search_SFS destructor.");}
        
        // NOTE: crit is expected to be already initialize()d before use here
        virtual bool search(const DIMTYPE target_d, RETURNTYPE &result, const PSubset sub, const PCriterion crit, std::ostream& os=std::cout) override;
 
        void set_search_direction(const SearchDirection direction) {assert(direction==FORWARD || direction==BACKWARD); _direction=direction;} 
        SearchDirection get_search_direction() const {return _direction;} 
        
        bool get_result(const DIMTYPE d, RETURNTYPE &result, PSubset &sub);
 
        virtual std::ostream& print(std::ostream& os) const override;
protected:
        SearchDirection _direction; 
        struct OneSubset {
                RETURNTYPE critvalue; 
                std::unique_ptr<SUBSET> sub; 
        };
        std::unique_ptr<OneSubset[]> bsubs; 
        DIMTYPE _n; 
};
 
template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class CRITERION, class EVALUATOR>
std::ostream& Search_SFS<RETURNTYPE,DIMTYPE,SUBSET,CRITERION,EVALUATOR>::print(std::ostream& os) const 
{
        if(parent::get_generalization_level()>1) os << "(G)";
        if(_direction==FORWARD) os << "SFS"; else os << "SBS";
        if(parent::get_generalization_level()>1) os << ", G=" << parent::get_generalization_level();
        os << "  [Search_SFS()";
        if(parent::_evaluator) os << " with " << *parent::_evaluator; 
        if(grandparent::result_tracker_active()) os << " with " << *grandparent::_tracker; 
        os << "]";
        return os;
}
 
template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class CRITERION, class EVALUATOR>
bool Search_SFS<RETURNTYPE,DIMTYPE,SUBSET,CRITERION,EVALUATOR>::search(const DIMTYPE target_d, RETURNTYPE &result, const PSubset sub, const PCriterion crit, std::ostream& os) // returns found subset + criterion value
{
        // NOTE: certain parameter combinations (initial sub->get_d() versus target_d versus parent::_generalization_level) 
        //       make it impossible to yield a valid solution. In such case search() just returns false
        StopWatch swatch;
        if(parent::_evaluator) parent::_evaluator->set_output_detail(grandparent::get_output_detail()); else throw FST::fst_error("Search_Sequential: Missing evaluator.");
        bool track=grandparent::result_tracker_active(); if(track) grandparent::_tracker->set_output_detail(grandparent::get_output_detail());
        assert(sub);
        assert(crit);
        assert(sub->get_n()>0);
        assert(target_d<=sub->get_n());
        assert(parent::get_generalization_level()>=1 && parent::get_generalization_level()<=sub->get_n());
 
        const bool _forward=(_direction==FORWARD);
        sub->set_forward_mode(true); // select_all()/deselect_all() write mode-dependent ids - a Subset left in backward mode by a previous search() would otherwise re-initialize empty
        if(!parent::initial_subset_enabled()) {
                if(_forward) sub->deselect_all(); else sub->select_all();
        }
        DIMTYPE _d=sub->get_d();
        _n=sub->get_n();
        bsubs.reset(new OneSubset[_n]);
 
        // no solution is reachable under this setting when, with g = generalization level:
        //   forward, but there is no room for another g-tuple (d+g>n),
        //   backward, but removing a g-tuple would not leave a non-empty subset (d<=g),
        //   a target size is requested that the starting subset has already reached or passed
        //     (forward: d>=target_d, backward: d<=target_d),
        //   or the distance between d and target_d is not a whole number of g-sized steps
        if( (_forward && _d+parent::get_generalization_level()>_n) || 
                (!_forward && _d<=parent::get_generalization_level()) || 
                (target_d>0 && _forward && _d>=target_d) || 
                (target_d>0 && !_forward && _d<=target_d) ||
                (target_d>0 && ((target_d>_d)?(target_d-_d):(_d-target_d))%parent::get_generalization_level()>0) ) return false; // no solution reachable under this setting
        
        if(grandparent::output_normal()) {
                msgbuf sos; 
                sos << "---------------------------------------" << std::endl;
                sos << "Starting " << *this << std::endl;
                sos << "with Criterion: " << *crit << std::endl;
                sos << "with initial subset: " << *sub << std::endl;
                if(target_d==0) sos << "Subset size to be optimized." << std::endl; else sos << "Target subset size set to: " << target_d << std::endl;
                sos << std::endl << std::flush;
                this->report(os,sos);
        }
 
        sub->set_forward_mode(_forward); // FORWARD is the prevailing mode
 
        if(_d>0) { // evaluate initial subset if any
                if(!crit->evaluate(result,sub)) return false;
                if(track) grandparent::_tracker->add(result,sub);
                bsubs[_d-1].sub.reset(new SUBSET(_n));
                bsubs[_d-1].sub->stateless_copy(*sub);
                bsubs[_d-1].critvalue=result;
                if(grandparent::output_normal()) this->report_solution(os, result, *sub, msgbuf() << "subresult=" << result << " " << *sub << std::endl << std::endl);
        }
 
        DIMTYPE steps;
        if(target_d>0 && target_d<=_n) { // search for subset of size target_d
                steps=((target_d>_d)?(target_d-_d):(_d-target_d))/parent::get_generalization_level();
        } else { // find best dimensionality
                if(_forward) steps=(_n-_d)/parent::get_generalization_level();
                else steps=(_d-1)/parent::get_generalization_level();
        }
        
        // main search loop
        for(DIMTYPE i=0;i<steps;i++) {
                if(!parent::Step(_forward, result, sub, crit, os)) return false;
                if(track) grandparent::_tracker->add(result,sub);
                if(grandparent::output_normal()) this->report_solution(os, result, *sub, msgbuf() << "subresult=" << result << " " << *sub << std::endl << std::endl << std::flush);
                _d=sub->get_d();
                bsubs[_d-1].sub.reset(new SUBSET(_n));
                bsubs[_d-1].sub->stateless_copy(*sub);
                bsubs[_d-1].critvalue=result;
        }
        
        DIMTYPE finalidx;
        if(target_d>0 && target_d<=_n) finalidx=target_d-1;
        else {
                for(finalidx=0;finalidx<_n && !bsubs[finalidx].sub;finalidx++);
                for(DIMTYPE i=finalidx+1;i<_n;i++) if(bsubs[i].sub && bsubs[i].critvalue>bsubs[finalidx].critvalue) finalidx=i;
        }
 
        if(finalidx>=_n || !bsubs[finalidx].sub) return false;
 
        result=bsubs[finalidx].critvalue;
        sub->stateless_copy(*(bsubs[finalidx].sub));
 
        if(grandparent::output_normal()) {
                msgbuf sos; 
                sos << "......................................." << std::endl;
                sos << "Search_SFS() search finished. " << swatch << std::endl;
                sos << "Search result: "<< std::endl << *sub << std::endl << "Criterion value: " << result << std::endl << std::endl << std::flush;
                this->report(os,sos);
        }
        return true;
}
 
template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class CRITERION, class EVALUATOR>
bool Search_SFS<RETURNTYPE,DIMTYPE,SUBSET,CRITERION,EVALUATOR>::get_result(const DIMTYPE d, RETURNTYPE &result, PSubset &sub) // retrieves temporary solution of cardinality d that had been updated in the course of the last search
{
        assert(sub);
        assert(d>0 && d<=_n);
        if(bsubs && d>0 && d<=_n && bsubs[d-1].sub) {
                result=bsubs[d-1].critvalue;
                sub->stateless_copy(*(bsubs[d-1].sub));
                return true;
        }
        return false;
}
 
 
} // namespace
#endif // FSTSEARCHSEQSFS_H ///:~