public Clusters Cluster(ClusterDistanceStrategy strategy, int k)
{
// build a clustering only with singleton clusters
this._BuildSingletonCluster();
//build the dissimilarity matrix
this._BuildDissimilarityMatrixParallel();
// build the hierarchical clustering
this.BuildHierarchicalClustering(_clusters.Count(), strategy, k);
return(_clusters);
}
private void _UpdateDissimilarityMatrix(Cluster newCluster, ClusterDistanceStrategy strategie)
{
double distanceBetweenClusters;
for (int i = 0; i < _clusters.Count(); i++)
{
// compute the distance between old clusters to the new cluster
distanceBetweenClusters = ComputeDistance(_clusters.GetCluster(i), newCluster, _dissimilarityMatrix, strategie);
// insert the new cluster's distance
_dissimilarityMatrix.AddClusterPairAndDistance(new ClusterPair(newCluster, _clusters.GetCluster(i)), distanceBetweenClusters);
//remove all old distance values of the old clusters (subclusters of the newcluster)
_dissimilarityMatrix.RemoveClusterPair(new ClusterPair(newCluster.GetSubCluster(0), _clusters.GetCluster(i)));
_dissimilarityMatrix.RemoveClusterPair(new ClusterPair(newCluster.GetSubCluster(1), _clusters.GetCluster(i)));
}
// finally, remove the distance of the old cluster pair
_dissimilarityMatrix.RemoveClusterPair(new ClusterPair(newCluster.GetSubCluster(0), newCluster.GetSubCluster(1)));
}
private void BuildHierarchicalClustering(int indexNewCluster, ClusterDistanceStrategy 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)
_clusters.RemoveCluster(closestClusterPair.Cluster1);
_clusters.RemoveCluster(closestClusterPair.Cluster2);
_UpdateDissimilarityMatrix(newCluster, strategy);
//add the new cluster to clustering
_clusters.AddCluster(newCluster);
// 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);
}
}
private static void AgnesClusteringPrint(IrisData[] irisData, DistanceDelegate calculateDistanceFunction, ClusterDistanceStrategy strategy)
{
var irisDataHier = irisData.Select(d => new double[] { d.PetalLength, d.PetalWidth, d.SepalLength, d.SepalWidth }).ToHashSet();
var hc = new AgnesClustering(calculateDistanceFunction, irisDataHier);
var clusters = hc.Cluster(strategy, 3);
var i = 0;
Console.ForegroundColor = ConsoleColor.DarkYellow;
Console.WriteLine($"Agnes {calculateDistanceFunction.Method.Name} {strategy.ToString()} Predicted:");
Console.ForegroundColor = ConsoleColor.Gray;
var res = new IrisData[clusters.Count()][];
foreach (Cluster cl in clusters)
{
var pp = cl.GetAllPatterns();
var clusterResult = pp.Select(p => p.GetAttributes()).Select(p => _irisData.First(x =>
x.PetalLength == p[0] &&
x.PetalWidth == p[1] &&
x.SepalLength == p[2] &&
x.SepalWidth == p[3])).ToArray();
res[i++] = clusterResult;
}
PrintResult(res, irisData.Length);
Console.WriteLine();
}
private double ComputeDistance(Cluster cluster1, Cluster cluster2, DissimilarityMatrix dissimilarityMatrix, ClusterDistanceStrategy strategy)
{
var distance1 = dissimilarityMatrix.ReturnClusterPairDistance(new ClusterPair(cluster1, cluster2.GetSubCluster(0)));
var distance2 = dissimilarityMatrix.ReturnClusterPairDistance(new ClusterPair(cluster1, cluster2.GetSubCluster(1)));
switch (strategy)
{
case ClusterDistanceStrategy.SingleLinkage: return(distance1 < distance2 ? distance1 : distance2);
case ClusterDistanceStrategy.CompleteLinkage: return(distance1 > distance2 ? distance1 : distance2);
case ClusterDistanceStrategy.AverageWeightedPairGroupMethodArithmeticMean: return((distance1 + distance2) / 2);
case ClusterDistanceStrategy.AverageUnweightedPairGroupMethodArithmeticMean:
return(((cluster2.GetSubCluster(0).TotalQuantityOfPatterns *distance1) / cluster2.TotalQuantityOfPatterns) +
((cluster2.GetSubCluster(1).TotalQuantityOfPatterns *distance2) / cluster2.TotalQuantityOfPatterns));
default: return(0);
}
}
