classification/KMeanCluster.cpp

Go to the documentation of this file.

#include "KMeanCluster.h"
#include <iostream>

KMeanCluster::KMeanCluster(int nCentres)
{
    for (int i = 0; i < nCentres; ++i) {
        m_centres.push_back(Vector3d(0));
    }
}

void KMeanCluster::train(VectorList &v)
{
    std::cout << "enter KMeanCluster::train" << std::endl;
    VectorList::iterator itCentres = m_centres.begin();
    VectorList::iterator itTraining = v.begin();

    // initialize centres
    std::cout << "\tinit centres" << std::endl;
    for (; itCentres != m_centres.end() && itTraining != v.end();
            ++itCentres, ++itTraining) {
        *itCentres = *itTraining;
    }
    if (itTraining == v.end())
        return;

    // assign groups
    std::cout << "\tassign groups" << std::endl;
    VectorList oldCentres;
    VectorList groups[m_centres.size()];

    do {
        oldCentres = m_centres;
        for (itTraining = v.begin(); itTraining != v.end(); ++itTraining) {
            groups[clusterize(*itTraining)].push_back(*itTraining);
        }

        // reassign centres
        std::cout << "\t\treassign centres" << std::endl;
        Vector3d tmp(0);
        int n = 0;
        for (itCentres = m_centres.begin(); 
                itCentres != m_centres.end(); 
                ++itCentres, ++n) {
            for (itTraining = groups[n].begin(); itTraining != groups[n].end(); 
                                                                ++itTraining){
                tmp += *itTraining;
            }
            tmp /= groups[n].size();
        }
    } while (isSameSet(oldCentres, m_centres));
    std::cout << "leave KMeanCluster::train" << std::endl;
}

int KMeanCluster::clusterize(Vector3d &v)
{
    float dist, bestDist = 1e100; // very far away...
    int n = 0, nBest = 0;
    Vector3d tmp;
    for(VectorList::iterator it = m_centres.begin(); it != m_centres.end(); ++n, ++it) {
        tmp = *it - v;
        dist = tmp.get(0)*tmp.get(0) + tmp.get(1)*tmp.get(1) + tmp.get(2)*tmp.get(2);
        if (dist < bestDist) {
            nBest = n;
            bestDist = dist;
        }
    }
    return nBest;
}

bool KMeanCluster::isSameSet(VectorList &a, VectorList &b)
{
    // assert(a.size() == b.size());
    VectorList::iterator itA, itB;
    for (itA = a.begin(), itB = b.begin(); itA != a.end() && itB != b.end(); 
                                                        ++itA, ++itB) {
        if (*itA != *itB)
            return false;
    }
    return true;
}

---