void kmedoids_allocator::allocate(const std::size_t p_amount, const dataset & p_data, cluster_sequence & p_clusters) {
medoid_sequence initial_medoids;
kmeans_plus_plus(p_amount).initialize(p_data, initial_medoids);
kmedoids_data result;
kmedoids(initial_medoids).process(p_data, result);
p_clusters = std::move(result.clusters());
}
void AgglomerativeClustering::startClustering(const ClusterMethodParameters* pParameters, Document *pDocument, ClusteringResult *pClusteringResult)
{
mpDocument = pDocument;
mpClusteringResult = pClusteringResult;
// preprocess glyphs or compute features, depending on distance type:
if (pParameters->dataType == FEATURES_BASED) {
pDocument->computeFeatures();
}
else {
pDocument->preprocessAllGlyphs();
}
// this->mpDocument->setDistanceType(pParameters->dataType);
// retrieve parameters:
const AgglomerativeParameters *pParams = (const AgglomerativeParameters*)(pParameters);
// print some debug info:
std::cout << "Starting agglomerative clustering algorithm..." << std::endl;
std::cout << "nr of clusters: " << pParams->nClusters << std::endl;
std::cout << "feature dist type: " << pParams->featureDistType << std::endl;
std::cout << "cluster dist type: " << pParams->clusterDistType << std::endl;
mpClusteringResult->deleteClustering(); // delete probably old clustering result
// setClusterDistanceTypeFunctionPointer(pParams->clusterDistType);
const int nSamples = this->mpDocument->nParsedImages();
if (nSamples < 2) {
throw NoDataException("No data found for clustering!");
}
if (pParameters->dataType == FEATURES_BASED && this->mpDocument->nFeatures() < 2) {
throw NoDataException("No features found for clustering!");
}
StopWatch watch;
// initialize distance matrix:
std::cout << "computing distance matrix..." << std::endl;
watch.start();
initDistanceMatrix();
watch.stop();
std::cout << "successfully computed distance matrix" << std::endl;
// TEST: USING CLUSTER LIBRARY:
double **tmpdistmatrix;
// Allocate memory for distance matrix
tmpdistmatrix = new double*[nSamples];
for (int i = 0; i < nSamples; ++i)
tmpdistmatrix[i] = new double[nSamples];
// copy distance matrix
for (int i=0;i<nSamples;++i){
for (int j=0; j<=i; ++j) {
tmpdistmatrix[i][j] = mDistMat(i,j);
tmpdistmatrix[j][i] = mDistMat(i,j);
}
}
std::vector<int> labels(nSamples);
#if 0
int npass = 100;
double error;
int ifound;
watch.start();
kmedoids (pParams->nClusters, nSamples, tmpdistmatrix,
npass, &labels[0], &error, &ifound);
watch.stop();
std::cout << "finished kmedioids, error = " << error << ", ifound = " << ifound << std::endl;
#else
// method = 's' (single-linkage), 'm' (complete-linkage), 'a' (average-linkage) or 'c' (centroid-linkage):
char methodChar='a';
switch (pParams->clusterDistType) {
case AVG_DIST:
methodChar='a';
break;
case MIN_DIST:
methodChar='s';
break;
case MAX_DIST:
methodChar='m';
break;
default:
throw Exception("Unknown distance type in AgglomerativeClustering::startClustering()");
break;
} // end switch
Node* tree = treecluster(nSamples, 1, NULL, NULL, NULL, 0, '_', methodChar, tmpdistmatrix);
if (tree==NULL)
std::cerr << "FATAL ERROR - NULL POINTER IN CLUSTER RESULT!" << std::endl;
// cut hierarchical cluster tree at specified nr of clusters:
cuttree(nSamples, tree, pParams->nClusters, &labels[0]);
delete [] tree;
#endif
// De-Allocate memory for temporary distance matrix:
for (int i = 0; i < nSamples; ++i)
delete [] tmpdistmatrix[i];
delete [] tmpdistmatrix;
std::cout << "finished agglo clustering with cluster libarary!!" << std::endl;
pClusteringResult->createClusteringResultFromLabelVector(labels, pDocument);
/////////////////// END TEST
#if 0
// get pointer to image chars:
std::vector<ImageChar*> *imageCharVecPointer = this->mpDocument->getImageCharsVecPointer();
// create cluster for each instance:
CharCluster *pCluster=NULL;
for (int i=0; i<nSamples; ++i) {
pCluster = mpClusteringResult->addEmptyCluster(this->mpDocument);
pCluster->addChar( (*imageCharVecPointer)[i] );
}
std::cout << "Starting merging process..." << std::endl;
// watch.start();
std::vector<float> minValVec;
std::vector<int> nClustsVec;
// while nr of clusters not reached -> merge two nearest clusters
// while (mClusterVec.size() > pParams->nClusters) {
while (mpClusteringResult->nClusters() > pParams->nClusters) {
// find min element of distance matrix:
minValVec.push_back(mMinDist);
// nClustsVec.push_back(mClusterVec.size());
nClustsVec.push_back(mpClusteringResult->nClusters());
// std::cout << "nr of clusters is " << mClusterVec.size() << std::endl;
// std::cout << "min dist is " << mMinDist << " on index " << mMinIndex << std::endl;
watch.start();
updateClusterLabels();
std::cout << "updated cluster labels, time = " << watch.stop(false) << std::endl;
watch.start();
findMinDist();
std::cout << "found min dist, time = " << watch.stop(false) << std::endl;
std::cout << "new nr of clusters is " << mpClusteringResult->nClusters() << std::endl;
} // end while
// writeTxtFile(nClustsVec, "C:/projekte/impact/matlab_sebastian/c_prog_out/n_clusts_vec.txt");
// writeTxtFile(minValVec, "C:/projekte/impact/matlab_sebastian/c_prog_out/min_val_vec.txt");
// watch.stop();
#endif
pDocument->clearAllPreprocessing();
pDocument->clearFeatures();
pClusteringResult->computePrototypeFeatures();
pClusteringResult->updatePrototypes(true);
return;
} // end startClustering()
