private void BuildHierarchicalClustering(int indexNewCluster, ClusterDistance.Strategy strategy, int k)
{
ClusterPair closestClusterPair = this._GetClosestClusterPairInDissimilarityMatrix();
// create a new cluster by merge the closest cluster pair
Cluster newCluster = new Cluster();
newCluster.AddSubCluster(closestClusterPair.Cluster1);
newCluster.AddSubCluster(closestClusterPair.Cluster2);
newCluster.Id = indexNewCluster;
newCluster.UpdateTotalQuantityOfPatterns(); //update the total quantity of patterns of the new cluster (this quantity is used by UPGMA clustering strategy)
//remove the closest cluster pair from the clustering data structure (clusters)
_clusters.RemoveClusterPair(closestClusterPair);
_UpdateDissimilarityMatrix(newCluster, strategy);
//add the new cluster to clustering
_clusters.AddCluster(newCluster);
closestClusterPair = null;
// recursive call of this method while there is more than 1 cluster (k>2) in the clustering
if (_clusters.Count() > k)
{
this.BuildHierarchicalClustering(indexNewCluster + 1, strategy, k);
}
}
//this functionality is usefull to see the distance matrix computed and to use as in input to other clustering algorithms implementation (R or Python)
public void CreateCSVMatrixFile(string path)
{
File.Delete(path);
this._BuildSingletonCluster();
StringBuilder matrix = new StringBuilder();
string headerLine = "AggloCluster";
foreach (Cluster cluster in _clusters)
{
headerLine = headerLine + ", Cluster" + cluster.Id;
}
bool writeBlank = false;
matrix.Append(headerLine);
double distanceBetweenTwoClusters;
ClusterPair clusterPair;
for (int i = 0; i < _clusters.Count(); i++)
{
matrix.Append("\r\n");
matrix.Append("Cluster" + _clusters.GetCluster(i).Id);
writeBlank = false;
for (int j = 0; j < _clusters.Count(); j++)
{
clusterPair = new ClusterPair();
clusterPair.Cluster1 = _clusters.GetCluster(i);
clusterPair.Cluster2 = _clusters.GetCluster(j);
distanceBetweenTwoClusters = ClusterDistance.ComputeDistance(clusterPair.Cluster1, clusterPair.Cluster2);
if (distanceBetweenTwoClusters == 0)
{
writeBlank = true;
matrix.Append(",0");
}
else
{
if (writeBlank)
{
matrix.Append("," + string.Empty);
}
else
{
matrix.Append("," + distanceBetweenTwoClusters);
}
}
}
}
File.AppendAllText(path, matrix.ToString());
}
public void RemoveClusterPair(ClusterPair clusterPair)
{
double outvalue;
if (_distanceMatrix.ContainsKey(clusterPair))
{
_distanceMatrix.TryRemove(clusterPair, out outvalue);
}
else
{
_distanceMatrix.TryRemove(new ClusterPair(clusterPair.Cluster2, clusterPair.Cluster1), out outvalue);
}
}
// get the distance value from a cluster pair. THIS METHOD DEPENDS ON THE EqualityComparer IMPLEMENTATION IN ClusterPair CLASS
public double ReturnClusterPairDistance(ClusterPair clusterPair)
{
double clusterPairDistance = Double.MaxValue;
// look in distance matrix if there is an input of cluster1 and cluster2 (remember that ClusterPair has two childs cluster1 and cluster2)
if (_distanceMatrix.ContainsKey(clusterPair))
{
clusterPairDistance = _distanceMatrix[clusterPair];
}
else
{
clusterPairDistance = _distanceMatrix[new ClusterPair(clusterPair.Cluster2, clusterPair.Cluster1)]; // if not, look in distance matrix for an input of cluster2 and cluster1 (remember that distance matrix is symetric)
}
return(clusterPairDistance);
}
// get the lowest distance in distance matrix
public double GetLowestDistance()
{
double minDistance = double.MaxValue;
ClusterPair closestClusterPair = new ClusterPair();
foreach (var item in _distanceMatrix)
{
if (item.Value < minDistance)
{
minDistance = item.Value;
}
}
return(minDistance);
}
// get the closest cluster pair (i.e., min cluster pair distance). it is also important to reduce computational time
public ClusterPair GetClosestClusterPair()
{
double minDistance = double.MaxValue;
ClusterPair closestClusterPair = new ClusterPair();
foreach (var item in _distanceMatrix)
{
if (item.Value < minDistance)
{
minDistance = item.Value;
closestClusterPair = item.Key;
}
}
return(closestClusterPair);
}
private IEnumerable <ClusterPair> _ClusterPairCollection()
{
ClusterPair clusterPair;
for (int i = 0; i < _clusters.Count(); i++)
{
for (int j = i + 1; j < _clusters.Count(); j++)
{
clusterPair = new ClusterPair();
clusterPair.Cluster1 = _clusters.GetCluster(i);
clusterPair.Cluster2 = _clusters.GetCluster(j);
yield return(clusterPair);
}
}
}
// calcula a distancia entre todos os clusters e as armazena na matrix de dissimilaridade
private void CreateDissimilarityMatrix()
{
double distanceBetweenTwoClusters;
_dissimilarityMatrix = new DissimilarityMatrix();
ClusterPair clusterPair;
for (int i = 0; i < _clusters.Count(); i++)
{
for (int j = i + 1; j < _clusters.Count(); j++)
{
clusterPair = new ClusterPair();
clusterPair.Cluster1 = _clusters.GetCluster(i);
clusterPair.Cluster2 = _clusters.GetCluster(j);
distanceBetweenTwoClusters = ClusterDistance.ComputeDistance(clusterPair.Cluster1, clusterPair.Cluster2);
_dissimilarityMatrix.AddClusterPairAndDistance(clusterPair, distanceBetweenTwoClusters);
}
}
}
//remove a cluster pair from the clustering data structure
public void RemoveClusterPair(ClusterPair clusterPair)
{
this.RemoveCluster(clusterPair.Cluster1);
this.RemoveCluster(clusterPair.Cluster2);
}
public void AddClusterPairAndDistance(ClusterPair clusterPair, double distance)
{
_distanceMatrix.TryAdd(clusterPair, distance);
}
public void AddClusterPairAndDistance(ClusterPair clusterPair, double distance)
{
_dendrogram.Add(clusterPair, distance);
}
