NICE_VisLearning/math/cluster/KMeansOnline.cpp

/**
* @file KMeansOnline.cpp
* @brief online kmeans clustering
* @author Erik Rodner
* @date 02/13/2008

*/

#include <iostream>
#include <list>

#include "vislearning/math/cluster/KMeansOnline.h"

using namespace OBJREC;

using namespace std;
// refactor-nice.pl: check this substitution
// old: using namespace ice;

using namespace NICE;



KMeansOnline::KMeansOnline(size_t _numClusters, double _gamma)
{
	gamma = _gamma;
	numClusters = _numClusters;

	init();
}

KMeansOnline::~KMeansOnline()
{
}

void KMeansOnline::init()
{
	ClusterOnline::init();

	weights.clear();
	weights.resize(numClusters, 0.0);
}

int KMeansOnline::updateClusters(const NICE::Vector & x)
{
	if (clusters.size() != numClusters)
	{
		clusters.push_back(x);
		fprintf(stderr, "clusters.size() %d weights.size() %d\n",
				(int)clusters.size(), (int)weights.size());
		weights[ clusters.size() - 1 ]++;
		return clusters.size() - 1;
	}

	double mindist = std::numeric_limits<double>::max();

	int k = 0;
	int my_cluster = 0;

	const double *xp = x.getDataPointer();
	int len = x.size();

	for (VVector::const_iterator i = clusters.begin();
						i != clusters.end();
						i++, k++)
	{
		const NICE::Vector & cluster = *i;
		//double distance = (x - cluster).Length();
		double distance = 0.0;
		const double *clusterp = cluster.getDataPointer();

		for (int i = 0 ; i < len ; i++)
		{
			double h = clusterp[i] - xp[i];
			distance += h * h;
		}

		if (distance < mindist)
		{
			my_cluster = k;
			mindist = distance;
		}
	}

	weights[my_cluster]++;

	// refactor-nice.pl: check this substitution
	// old: Vector & cluster = clusters[my_cluster];
	NICE::Vector & cluster = clusters[my_cluster];
	double f = gamma / weights[my_cluster];

	for (size_t i = 0 ; i < cluster.size() ; i++)
		cluster[i] += f * (x[i] - cluster[i]);

	return my_cluster;

}

void KMeansOnline::cluster(const Examples &ex)
{
	for (int i = 0; i < (int)ex.size(); i++)
	{
		updateClusters(*(ex[i].second.vec));
	}
}

void KMeansOnline::getDist(const NICE::Vector &vin, SparseVector &dist, int size, bool hard)
{
	int len = vin.size();

	list<double> alldist;

	int k = 0;

	double mindist = std::numeric_limits<double>::max();

	// get for each cluster the distance to the given feature

	for (vector<Vector>::const_iterator i = clusters.begin();
						i != clusters.end();
						i++, k++)
	{
		const NICE::Vector & cluster = *i;

		double distance = 0.0;
		const double *clusterp = cluster.getDataPointer();

		for (int i = 0 ; i < len ; i++)
		{
			double h = clusterp[i] - vin[i];
			distance += h * h;
		}

		mindist = std::min(mindist, distance);

		alldist.push_back(distance);
	}

	list<double> sortedalldist = alldist;

	sortedalldist.sort();

	dist.clear();
	dist.setDim((int)numClusters);
	size = std::min((int)numClusters, size);


	for (k = 0; k < size; k++)
	{
		sortedalldist.pop_front();
	}

	double maxval = *(sortedalldist.begin());

	list<double>::const_iterator i = alldist.begin();

	for (k = 0; i != alldist.end(); i++, k++)
	{
		double val = mindist / (*i);

		if ((*i) < maxval && val > 1e-07)
		{
			if (hard)
			{
				dist[k] = 1.0;
			}
			else
			{
				dist[k] = val;
			}
		}
	}
}