bool Node::checkAndRunLastIteration() {
if(clusteringType() == FLAT ||
numberOfClusters == -1)
return false;
if(clusteringType() == BOTTOM_UP &&
numberOfClusters != -1 && clusteringSize() < finalStepTrigger) {
for(int i = clusteringSize(); i > numberOfClusters; i--)
runLastIteration();
return true;
}
if(clusteringType() == TOP_DOWN &&
numberOfClusters != -1 && clusteringSize() > finalStepTrigger) {
for(int i = clusteringSize(); i < numberOfClusters; i++)
runLastIteration();
return true;
}
return false;
}
void PartitionByOnePivot_Trisection::partitionByOnePivot(shared_ptr<ClusteringTask>& task,CMetricDistance* metric,int& numpartition,shared_ptr<CMetricData>& pivot,int indictor,int& cluster_Start,int& cluster_end)
{
int pivotSize = task->getNewPivots()->size();
int taskSize = task->getData()->size();
task->getMetricData_Distance()->clear();
for(int i=0; i < taskSize; ++i)
{
task->getMetricData_Distance()->push_back(metric->getDistance(pivot.get(),((task->getData())->at(i)).get()));
}
numpartition = numpartition > taskSize ? taskSize : numpartition;
shared_ptr<vector<double> > metricData_distance_temp(new vector<double>());
for(int i=0; i < task->getMetricData_Distance()->size(); ++i)
{
metricData_distance_temp->push_back(task->getMetricData_Distance()->at(i));
}
vector<double>* kth_quantiles = new vector<double>();
vector<vector<int> > equal_k_label(numpartition-1);
vector<int>clusteringSize(numpartition,0);
double partitionRadius = this->trisectionRadius;
double mediaValue = 0;
for(int i=1; i < numpartition; i++)
{
kth_quantiles->push_back(utilFunction::select_min_k_iteration(metricData_distance_temp,taskSize,i*taskSize/numpartition + 1));
}
if(kth_quantiles->size() == 2 && (kth_quantiles->at(1) - kth_quantiles->at(0)) < 2 * partitionRadius)
{
mediaValue = utilFunction::select_min_k_iteration(metricData_distance_temp,taskSize,1.0 * taskSize/2 + 1);
kth_quantiles->at(0) = mediaValue-partitionRadius;
kth_quantiles->at(1) = mediaValue+partitionRadius;
}
//double minValue = utilFunction::find_min(*metricData_distance_temp.get());
//double maxValue = utilFunction::find_max(*metricData_distance_temp.get());
//if((maxValue-minValue) > (kth_quantiles->at(1)-kth_quantiles->at(0)))
if(kth_quantiles->at(0) > 0)
{
for(int i=0; i < taskSize; ++i)
{
double distanceValue = task->getMetricData_Distance()->at(i);
for(int j=0; j< kth_quantiles->size(); ++j)
{
if(distanceValue < kth_quantiles->at(j) && task->getClusterId()->at(i) == -1)
{
task->getClusterId()->at(i) = j + indictor*numpartition + 1;
clusteringSize[j]++;
break;
}
if(j== kth_quantiles->size()-1 && distanceValue > kth_quantiles->at(j))
{
task->getClusterId()->at(i) = j + indictor*numpartition + 2;
clusteringSize[j+1]++;
}
if (distanceValue == kth_quantiles->at(j))
{
equal_k_label[j].push_back(i);
break;
}
}
}
for(int i=0; i < equal_k_label.size(); ++i)
{
int leftTaskSize = clusteringSize[i] + equal_k_label[i].size() - clusteringSize[i+1];
int rightTaskSize = clusteringSize[i+1] + equal_k_label[i].size() - clusteringSize[i];
if(abs(leftTaskSize) <= abs(rightTaskSize))
{
for(int j=0; j < equal_k_label[i].size(); ++j)
{
task->getClusterId()->at(equal_k_label[i][j]) = i + indictor*numpartition + 1;
}
}
else
{
for(int j=0; j < equal_k_label[i].size(); ++j)
{
task->getClusterId()->at(equal_k_label[i][j]) = i + indictor*numpartition + 2;
}
}
}
}
else
{
for(int i=0; i < taskSize; ++i)
{
task->getClusterId()->at(i) = indictor* numpartition + 1;
}
}
cluster_Start = indictor* numpartition + 1;
cluster_end = (indictor+1) * numpartition;
}
void Node::runLastIteration() {
cout << "Last iteration: starting with " <<
clusteringSize() << " clusters" << endl;
iteration(true);
print(clusteringSize(), true);
}
void Node::runIteration(int iterationNumber) {
cout << "Iteration " << iterationNumber << " (" << name << ")" << endl;
iteration(false);
cout << "Clustering size: " << clusteringSize() << endl;
print(iterationNumber, false);
}