geneticweaklearner.cpp

WeakLearner,弱分类器

//LAB_LicenseBegin==============================================================
//  Copyright (c) 2005-2006, Hicham GHORAYEB < [e-mail] >
//  All rights reserved.
//
//	This software is a Library for Adaptive Boosting. It provides a generic 
//	framework for the study of the Boosting algorithms. The framework provides 
//	the different tasks for boosting: Learning, Validation, Test, Profiling and 
//	Performance Analysis Tasks.
//
//	This Library was developped during my PhD studies at:
//	Ecole des Mines de Paris - Centre de Robotique( CAOR )
//	http://caor.ensmp.fr
//	under the supervision of Pr. Claude Laurgeau and Bruno Steux
//
//  Redistribution and use in source and binary forms, with or without
//  modification, are permitted provided that the following conditions are met:
//
//      * Redistributions of source code must retain the above copyright
//        notice, this list of conditions and the following disclaimer.
//      * Redistributions in binary form must reproduce the above copyright
//        notice, this list of conditions and the following disclaimer
//        in the documentation and/or other materials provided with the distribution.
//      * Neither the name of the Ecole des Mines de Paris nor the names of
//        its contributors may be used to endorse or promote products
//        derived from this software without specific prior written permission.
//
//  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
//  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, 
//  THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 
//  PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR 
//  CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 
//  EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 
//  PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 
//  PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 
//  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 
//  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 
//  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
//================================================================LAB_LicenseEnd
#include "GeneticWeakLearner.h"
#include "samples/learningsets/ILearningSet.h"
#include "samples/Sample.h"
#include "classifiers/features/ImageFeature.h"
#include "classifiers/features/ControlPointFeature.h"
#include "classifiers/features/IRandomWeakClassifierGenerator.h"

#include "LibAdaBoost/modules/options/GeneticWeakLearnerOptions.h"
using namespace modules::options;

#include "wlinterfaces/GeneticWeakLearnerInterface.h"
using namespace learners::wlinterfaces;

#include <cstdlib>
#include <iostream>
#include <algorithm>
#include <iostream>
#include <iomanip>
#include <string>
#include <sstream>
using namespace std;

using namespace samples;
using namespace classifiers;
using namespace classifiers::features;
using namespace learners;

namespace learners{
	bool lessErr(const WeakClassifier *i1,const WeakClassifier *i2)
	{
		return i1->GetError() < i2->GetError();
	}
}

GeneticWeakLearner::GeneticWeakLearner()
:m_LearningSet	  ( NULL                      )
,m_RandomGenerator( NULL                      )
,m_nb_indiv		  ( GA_DEFAULT_NBR_INDIV      )
,m_winner_to_keep ( GA_DEFAULT_WINNER_TO_KEEP )
,m_looser_to_keep ( GA_DEFAULT_LOOSER_TO_KEEP )
,m_max_generation ( GA_DEFAULT_MAX_GENERATION )
,m_nb_mutation	  ( GA_DEFAULT_NBR_MUTATION   )
,m_interface      ( NULL                      )
,m_bestwc         ( t_indiv(NULL, 1.0)        )
{
	m_population.reserve( m_nb_indiv );
	m_StringLog.clear();
}

GeneticWeakLearner::GeneticWeakLearner(ILearningSet *learningSet)
:m_LearningSet    ( learningSet               )
,m_RandomGenerator( NULL                      )
,m_nb_indiv		  ( GA_DEFAULT_NBR_INDIV      )
,m_winner_to_keep ( GA_DEFAULT_WINNER_TO_KEEP )
,m_looser_to_keep ( GA_DEFAULT_LOOSER_TO_KEEP )
,m_max_generation ( GA_DEFAULT_MAX_GENERATION )
,m_nb_mutation	  ( GA_DEFAULT_NBR_MUTATION   )
,m_interface      ( NULL                      )
,m_bestwc         ( t_indiv(NULL, 1.0)        )
{
	m_population.reserve( m_nb_indiv );
	m_StringLog.clear();
}

GeneticWeakLearner::~GeneticWeakLearner()
{
	for(int i=0; i < m_population.size(); i++)
	{
		delete m_population[i];
		m_population[i] = NULL;
	}

	if( m_bestwc.first != NULL){
		delete m_bestwc.first;
	}

	m_population.clear();
}

// IWeakLearner
WeakClassifier *
GeneticWeakLearner::GetNewWeakClassifier(double &error)
{
	WeakClassifier *wc = NULL;

	// Initialize the Learner, random generation of the
	// initial population
	Reset();

	// Run the Genetic Algorithm 
	Learn();

	// Get the result of the genetic algorithm
	wc = (WeakClassifier *)m_bestwc.first->Clone();

	// Get the learning result
	error = m_bestwc.second;
	return wc;
}

// GA Protected Methods
void
GeneticWeakLearner::TestWeakClassifier(WeakClassifier *feature, const ILearningSet *learningSet)
{
	double err = 0.0;
	int nbTruePositive  = 0;
	int nbTrueNegative  = 0;
	int nbFalsePositive = 0;
	int nbFalseNegative = 0;

	//std::cout << "[CHECK POSITIVE SAMPLES]" <<std::endl;
	// CHECK POSITIVE SAMPLES
	for(unsigned i=0; i < learningSet->NbrOfPositiveSamples(); i++)
	{
		const Sample *sample = learningSet->GetPositiveSampleConstAt( i );

		int curClass = feature->Classify( sample );

		if( curClass == POSITIVE_CLASS_ID)
		{
			nbTruePositive++;
		}else{
			nbFalseNegative++;
			// INCREMENT ERROR
			err += sample->GetWeight();
		}
	}

	//std::cout << "[CHECK NEGATIVE SAMPLES]" <<std::endl;
	// CHECK NEGATIVE SAMPLES
	for(unsigned i=0; i < learningSet->NbrOfNegativeSamples(); i++)
	{
		const Sample *sample = learningSet->GetNegativeSampleConstAt( i );

		int curClass = feature->Classify( sample );

		if( curClass == NEGATIVE_CLASS_ID)
		{
			nbTrueNegative++;
		}else{
			nbFalsePositive++;
			// INCREMENT ERROR
			err += sample->GetWeight();
		}
	}

	//std::cout << "[SET THE ERROR]" <<std::endl;
	// SET THE ERROR
	feature->SetError( err );

	// LOG CLASSIFICATION RESULT FOR FURTHER PROCESSING
	feature->SetTruePositive( nbTruePositive );
	feature->SetTrueNegative( nbTrueNegative );
	feature->SetFalsePositive( nbFalsePositive );
	feature->SetFalseNegative( nbFalseNegative );
}

// IWeakLearner
void
GeneticWeakLearner::Learn(void)
{
	assert( m_LearningSet != NULL );

	const GeneticWeakLearnerInterface *myInterface = (const GeneticWeakLearnerInterface *)GetWeakLearnerInterface();

	ostringstream oStrLog;

	unsigned generation = 0;
	unsigned nbr_generation_no_improvement = 0;

	oStrLog.clear();

	/* TODO:: SAVE THE BEST FEATURE IN THE EVOLUTION POPULATION
	* from one iteration to another
	*/
	generation = 0;
	while( true ){
		// ITERATES OVER ALL THE INDIV IN THE POPULATION
		// APPLY EACH WEAK CLASSIFIER ON THE LEARNING SET
		// AND GET THE RESULT ERROR.
		GeneticWeakLearner::t_population::iterator it;
		for(it = m_population.begin(); it != m_population.end(); it++){
			// EVALUATE THE CURENT WEAK CLASSIFIER ON THE LEARNING SET
			WeakClassifier *wc = (*it);
			TestWeakClassifier(wc, m_LearningSet);
		}

		// SORT THE WEAK CLASSIFIERS IN THE POPULATION RESPECT TO ERROR
		std::sort(m_population.begin(), m_population.end(), lessErr );

		for(unsigned t=0; t < (m_population.size() - 1); t++)
		{
			//std::cerr << t << "-->"<< m_population[t]->GetError() << ", " << m_population[t+1]->GetError() << std::endl;
			assert( m_population[t]->GetError() <= m_population[t+1]->GetError());
		}

		// Clone The Best WeakClassifier
		WeakClassifier *cur_bestwc = (WeakClassifier *)m_population.front()->Clone();
		double cur_error = m_population.front()->GetError();
		
		// DO SOME LOG ON THAT STEP: Report results of this generation
		std::cout
			//<< "Current Error = "
			<< setprecision( 5 )
			<< setw( 7 )
			<< left
			<< cur_error << " c, "
			<< setprecision( 5 )
			<< setw( 7 )
			<< left
			<< m_bestwc.second << " l, "
			<< setfill(' ')
			<< setw( 4 )
			<< nbr_generation_no_improvement
			<< "th time"
			<< "("
			<< setw( 4 )
			<< cur_bestwc->GetTruePositive()
			<< "/"
			<< setw( 4 )
			<< m_LearningSet->NbrOfPositiveSamples() 
			<< " pos"
			<< ", "
			<< setw( 4 )
			<< cur_bestwc->GetTrueNegative() 
			<< "/" 
			<< setw( 4 )
			<< m_LearningSet->NbrOfNegativeSamples() 
			<< " neg"
			<< ", "
			<< setw( 4 )
			<< cur_bestwc->GetTrueNegative() + cur_bestwc->GetTruePositive()
			<< "/"
			<< setw( 4 )