util::Clustering::Clustering(const DataVector& _data, unsigned _max) : m_maxClusters(_max) {
m_data.insert(m_data.begin(), _data.begin(), _data.end());
ClusterMap clusterQualityScores;
for (unsigned clusterCount = 1; clusterCount <= m_maxClusters; clusterCount++) {
clusterQualityScores[clusterCount] = _kmeans(clusterCount);
}
std::vector< CountClusterPair > sortedScores;
std::copy(clusterQualityScores.begin(), clusterQualityScores.end(), std::back_inserter(sortedScores));
ScoreComparator comparator;
std::sort(sortedScores.begin(), sortedScores.end(), comparator);
report("Scores:");
for (int i = 0; i < sortedScores.size(); i++) {
report(sortedScores[i].first << " clusters: " << sortedScores[i].second.getScore());
}
report("Clustering with highest score: ");
report(util::Indents(2) << "cluster count: " << sortedScores[0].first);
report(util::Indents(2) << "aggregate score: " << sortedScores[0].second.getScore());
report(util::Indents(2) << "detected clusters:");
for (auto it = sortedScores[0].second.getClusters().begin(); it != sortedScores[0].second.getClusters().end(); ++it) {
const Cluster& cluster = *it;
report(util::Indents(4) << "position = " << cluster.position << ", elements = " << cluster.data.size());
}
result = sortedScores[0].second;
}
void CodeAtlas::RandomWalkClusterer::randomWalk()
{
srand(m_seed);
int nVtx = m_vtx.size();
typedef QHash<int, double> ClusterMap;
typedef QHash<int, float> EdgeMap;
for (int ithIter = 0 ; ithIter < m_maxIter;)
{
// choose a vtx to modify randomly
int curVtxID = rand()%nVtx;//randInt(nVtx);
//printf("vtx %d\n", curVtxID);
Vertex& curVtx= m_vtx[curVtxID];
// collect near cluster
ClusterMap nearClusterWeight;
for (EdgeMap::Iterator pE = curVtx.m_outEdge.begin(); pE != curVtx.m_outEdge.end(); ++pE)
{
int tarID = pE.key();
double w = pE.value();
if (w == 0)
continue;
Vertex& nearVtx = m_vtx[tarID];
if (nearClusterWeight.contains(nearVtx.m_clusterID))
{
nearClusterWeight[nearVtx.m_clusterID] += w;
}
else
{
nearClusterWeight[nearVtx.m_clusterID] = w;
}
}
if (nearClusterWeight.size() == 0)
continue;
// compute probability
double maxWeight = 0;
for (ClusterMap::Iterator pC = nearClusterWeight.begin(); pC != nearClusterWeight.end(); ++pC)
{
//printf("%lf ", pC.value());
maxWeight = max(maxWeight, pC.value());
}
//printf("\n");
double weightSum = 0;
double t = m_t;
for (ClusterMap::Iterator pC = nearClusterWeight.begin(); pC != nearClusterWeight.end(); ++pC)
{
double w = exp((pC.value() - maxWeight)/ t);
pC.value() = w;
weightSum += w;
}
for (ClusterMap::Iterator pC = nearClusterWeight.begin(); pC != nearClusterWeight.end(); ++pC)
{
pC.value() /= weightSum;
//printf("%lf ", pC.value());
}
//printf("\n");
// choose new cluster
double prob = randFloat();
double accProb = 0;
int newCluster = -1;
for (ClusterMap::Iterator pC = nearClusterWeight.begin(); pC != nearClusterWeight.end(); ++pC)
{
accProb += pC.value();
if (accProb > prob)
{
newCluster = pC.key();
}
}
// prevent round-off error
if (newCluster == -1)
{
newCluster = (nearClusterWeight.end() - 1).key();
}
//printf("old: %d new: %d\n\n", curVtx.m_clusterID, newCluster);
// update graph
curVtx.m_clusterID = newCluster;
++ithIter;
}
}
