/// <summary>
/// Finds the closest pair of clusters.
/// </summary>
/// <param name="a">The first cluster.</param>
/// <param name="b">The second cluster.</param>
/// <returns>Returns false if no more clusters can be merged.</returns>
private bool FindClosestPair(out LinkedCluster a, out LinkedCluster b)
{
// initialize the minimal distance
double minDist = double.PositiveInfinity;
a = b = null;
// look at all clusters
foreach (LinkedCluster f in clusters)
{
if (f.distanceList.Count == 0)
{
continue;
}
// take the minimal distance
if (f.distanceList.Keys[0] < minDist)
{
// store the minimal distance
minDist = f.distanceList.Keys[0];
// store the indices of the closest pair
a = f;
b = f.distanceList.Values[0];
}
}
if (a != null && b != null)
{
// closest pair found
return(true);
}
else
{
// no more pairs to merge
return(false);
}
}
/// <summary>
/// Computes the complete link clustering.
/// </summary>
/// <param name="vertices">The array of vertices to cluster.</param>
/// <param name="maxClusterDiameter">The maximal allowed distance between any points in a cluster.</param>
public int ComputeClustering(Vertex[] vertices)
{
this.vertices = vertices;
// initialize every vertex as a cluster
clusters = new List <LinkedCluster>(vertices.Length);
for (int i = 0; i < vertices.Length; i++)
{
LinkedCluster f = new LinkedCluster(i);
f.AddVertex(i);
clusters.Add(f);
}
// initialize mutual distances
for (int i = 0; i < vertices.Length - 1; i++)
{
for (int j = i + 1; j < vertices.Length; j++)
{
double distance = vertices[i].WeightedDistance(vertices[j]);
if (distance <= maxClusterDiameter)
{
clusters[i].distanceList.Add(distance, clusters[j]);
clusters[j].distanceList.Add(distance, clusters[i]);
}
}
}
double[,] distanceMatrix = new double[vertices.Length, vertices.Length];
int half = vertices.Length / 2;
// create distance matrix
for (int i = 0; i < vertices.Length; i++)
{
for (int jPartial = 0; jPartial < half; jPartial++)
{
// make the index start after i and rotate modulo Length
int j = (jPartial + i + 1) % vertices.Length;
// compute the distance
double distance = vertices[i].WeightedDistance(vertices[j]);
// store it to the matrix
if (distance <= maxClusterDiameter)
{
distanceMatrix[i, j] = distance;
}
}
}
LinkedCluster a, b;
// do the clustering
while (FindClosestPair(out a, out b))
{
// merge b into a
foreach (int vertexIndex in b.MemberIndices)
{
a.AddVertex(vertexIndex);
}
// keep only a, discard b
clusters.Remove(b);
// remove the distance between the merged pair
a.distanceList.RemoveAt(0);
// update distances
foreach (LinkedCluster f in clusters)
{
// skip a and b
if (f == a) // || f == b b is already removed from facilities
{
continue;
}
// find distance to both clusters of the pair
int indexOfA = f.distanceList.IndexOfValue(a);
int indexOfB = f.distanceList.IndexOfValue(b);
int indexOfF = a.distanceList.IndexOfValue(f);
if (indexOfA < 0 || indexOfB < 0)
{
// farther than the maximal allowed diameter
if (indexOfA >= 0)
{
f.distanceList.RemoveAt(indexOfA);
}
if (indexOfB >= 0)
{
f.distanceList.RemoveAt(indexOfB);
}
if (indexOfF >= 0)
{
a.distanceList.RemoveAt(indexOfF);
}
}
else
{
double newDistance = f.distanceList.Keys[indexOfA];
// keep the greater distance
// compare only indices, the array is sorted
if (f.distanceList.Keys[indexOfA] < f.distanceList.Keys[indexOfB])
{
// replace the key, keep the value
LinkedCluster value = f.distanceList.Values[indexOfA];
double key = f.distanceList.Keys[indexOfB];
f.distanceList.RemoveAt(indexOfB);
f.distanceList.Add(key, value);
newDistance = key;
}
// discard the other distance
f.distanceList.RemoveAt(indexOfB);
if (indexOfF >= 0)
{
// do this only if the above if passes
// update distance in a
LinkedCluster value2 = a.distanceList.Values[indexOfF];
a.distanceList.RemoveAt(indexOfF);
a.distanceList.Add(newDistance, value2);
}
}
}
}
return(-1);
}
