NICE_VisLearning/classifier/kernelclassifier/KCOneVsAll.cpp

/** 
* @file KCOneVsAll.cpp
* @brief One vs. All interface for kernel classifiers
* @author Erik Rodner
* @date 12/10/2009

*/
#include <iostream>

#include "KCOneVsAll.h"

#include "core/vector/Algorithms.h"

#include "core/algebra/CholeskyRobust.h"
#include "core/algebra/CholeskyRobustAuto.h"
#include "vislearning/regression/regressionbase/TeachWithInverseKernelMatrix.h"

using namespace std;
using namespace NICE;
using namespace OBJREC;



KCOneVsAll::KCOneVsAll( const Config *conf, const KernelClassifier *prototype, const string & section )
	: KernelClassifier ( conf, prototype->getKernelFunction() )
{
	this->prototype = prototype;
	this->maxClassNo = 0;
	this->verbose = conf->gB( section, "verbose", false );
}

KCOneVsAll::~KCOneVsAll()
{
}

void KCOneVsAll::teach ( KernelData *kernelData, const NICE::Vector & y )
{
	maxClassNo = (int)y.Max();
	
	classifiers.clear();

    for ( int i = 0 ; i <= maxClassNo ; i++ )
    {
		NICE::Vector ySub ( y );
		for ( size_t j = 0 ; j < ySub.size() ; j++ )
			ySub[j] = ((int)y[j] == i) ? 1 : 0;

		KernelClassifier *classifier;
		classifier = prototype->clone();

		fprintf (stderr, "KCOneVsAll: training classifier class %d <-> remainder\n", i );

		KernelData *kernelDataCopy = kernelData->clone();
		classifier->teach ( kernelDataCopy, ySub );

		// FIXME: This might be trickier for kernel classifiers which need the kernel data
		// explicitly, but otherwise we get a huge! memory leak
		delete kernelDataCopy;

		classifiers.push_back ( pair<int, KernelClassifier*> (i, classifier) );
    }


}

ClassificationResult KCOneVsAll::classifyKernel ( const NICE::Vector & kernelVector, double kernelSelf ) const
{
	if ( classifiers.size() <= 0 ) 
		fthrow(Exception, "The classifier was not trained with training data (use teach(...))");

	FullVector scores ( maxClassNo+1 );
    scores.set(0);
    for ( vector< pair<int, KernelClassifier *> >::const_iterator i = 
	    classifiers.begin(); i != classifiers.end(); i++ )
    {
		int classno = i->first;
		KernelClassifier *classifier = i->second;
		ClassificationResult r = classifier->classifyKernel(kernelVector, kernelSelf);

		scores[classno] += r.scores[1];
    }

	return ClassificationResult( scores.maxElement(), scores );
}