private Cluster <T> CreateCluster(HashSet <int> processed, BidirectionalGraph <Cluster <T>, ClusterEdge <T> > tree, IList <ClusterOrderEntry <T> > clusterOrder, int startIndex, int endIndex)
{
var subclusterDataObjects = new HashSet <T>();
var subclusters = new List <Cluster <T> >();
foreach (Cluster <T> subcluster in GetSubclusters(processed, tree, clusterOrder, startIndex, endIndex))
{
subclusterDataObjects.UnionWith(tree.GetAllDataObjects(subcluster));
subclusters.Add(subcluster);
}
for (int i = startIndex; i < endIndex; i++)
{
processed.Add(i);
}
var cluster = new Cluster <T>(clusterOrder.Skip(startIndex).Take(endIndex - startIndex).Select(oe => oe.DataObject).Except(subclusterDataObjects));
tree.AddVertex(cluster);
foreach (Cluster <T> subcluster in subclusters)
{
tree.AddEdge(new ClusterEdge <T>(cluster, subcluster));
}
return(cluster);
}
public IBidirectionalGraph <Cluster <T>, ClusterEdge <T> > GenerateClusters(IEnumerable <T> dataObjects)
{
var tree = new BidirectionalGraph <Cluster <T>, ClusterEdge <T> >(false);
var clusters = new List <Cluster <T> >();
foreach (T dataObject in dataObjects)
{
var cluster = new Cluster <T>(dataObject)
{
Description = dataObject.ToString()
};
clusters.Add(cluster);
tree.AddVertex(cluster);
}
var distances = new Dictionary <UnorderedTuple <Cluster <T>, Cluster <T> >, double>();
var heights = new Dictionary <Cluster <T>, double>();
for (int i = 0; i < clusters.Count; i++)
{
for (int j = i + 1; j < clusters.Count; j++)
{
double distance = _getDistance(clusters[i].DataObjects.First(), clusters[j].DataObjects.First());
if (double.IsNaN(distance) || double.IsInfinity(distance) || distance < 0)
{
throw new ArgumentException("Invalid distance between data objects.", "dataObjects");
}
distances[UnorderedTuple.Create(clusters[i], clusters[j])] = distance;
}
heights[clusters[i]] = 0;
}
while (clusters.Count >= 2)
{
int minI = 0, minJ = 0;
double minDist = double.MaxValue;
for (int i = 0; i < clusters.Count; i++)
{
for (int j = i + 1; j < clusters.Count; j++)
{
double dist = distances[UnorderedTuple.Create(clusters[i], clusters[j])];
if (dist < minDist)
{
minDist = dist;
minI = i;
minJ = j;
}
}
}
Cluster <T> iCluster = clusters[minI];
Cluster <T> jCluster = clusters[minJ];
distances.Remove(UnorderedTuple.Create(iCluster, jCluster));
var uCluster = new Cluster <T>();
tree.AddVertex(uCluster);
double height = minDist / 2;
heights[uCluster] = height;
int iCount = tree.GetAllDataObjects(iCluster).Count();
double iLen = height - heights[iCluster];
if (iLen <= 0 && !tree.IsOutEdgesEmpty(iCluster))
{
foreach (ClusterEdge <T> edge in tree.OutEdges(iCluster))
{
tree.AddEdge(new ClusterEdge <T>(uCluster, edge.Target, edge.Length));
}
tree.RemoveVertex(iCluster);
}
else
{
tree.RemoveInEdgeIf(iCluster, edge => true);
tree.AddEdge(new ClusterEdge <T>(uCluster, iCluster, Math.Max(iLen, 0)));
}
int jCount = tree.GetAllDataObjects(jCluster).Count();
double jLen = height - heights[jCluster];
if (jLen <= 0 && !tree.IsOutEdgesEmpty(jCluster))
{
foreach (ClusterEdge <T> edge in tree.OutEdges(jCluster))
{
tree.AddEdge(new ClusterEdge <T>(uCluster, edge.Target, edge.Length));
}
tree.RemoveVertex(jCluster);
}
else
{
tree.RemoveInEdgeIf(jCluster, edge => true);
tree.AddEdge(new ClusterEdge <T>(uCluster, jCluster, Math.Max(jLen, 0)));
}
double iWeight = (double)iCount / (iCount + jCount);
double jWeight = (double)jCount / (iCount + jCount);
foreach (Cluster <T> kCluster in clusters.Where(c => c != iCluster && c != jCluster))
{
UnorderedTuple <Cluster <T>, Cluster <T> > kiKey = UnorderedTuple.Create(kCluster, iCluster);
UnorderedTuple <Cluster <T>, Cluster <T> > kjKey = UnorderedTuple.Create(kCluster, jCluster);
distances[UnorderedTuple.Create(uCluster, kCluster)] = (iWeight * distances[kiKey]) + (jWeight * distances[kjKey]);
distances.Remove(kiKey);
distances.Remove(kjKey);
}
clusters.RemoveAt(minJ);
clusters.RemoveAt(minI);
clusters.Add(uCluster);
}
return(tree);
}
