public ClusteringResult <TData> Cluster(int clustersCount, IEnumerable <ClusteringData <TData> > data)
{
var sw = new Stopwatch();
sw.Start();
var dataIndexes = this.CreateMapDataToIndex(data);
Log.DebugFormat("KMeans: start clustering {0} data items, number of clusters set to {1}", dataIndexes.Count, clustersCount);
// Number of clusters can not be more than number of data.
clustersCount = Math.Min(dataIndexes.Count, clustersCount);
Log.DebugFormat("KMeans: number of clusters changed to {0}", clustersCount);
var clustering = this.InitiallyDevideDataByClusters(clustersCount, dataIndexes);
// When clustersCount == 1 then do not need to clustering at all, just return given data as one cluster.
if (clustersCount > 1)
{
var means = new double[clustersCount];
var centroids = new ClusteringData <TData> [clustersCount];
this.CalculateCentroids(dataIndexes, clustering, means, centroids);
var interationNum = 0;
while (this.RecalculateClusters(dataIndexes, clustering, centroids) && interationNum++ < this.MaxIterationsCount)
{
this.CalculateCentroids(dataIndexes, clustering, means, centroids);
}
}
var result = this.BuildClusteredResult(clustersCount, dataIndexes, clustering);
sw.Stop();
Log.DebugFormat("KMeans: end clustering {0} data items, took {1} ms", dataIndexes.Count, sw.ElapsedMilliseconds);
return(result);
}
public int CompareTo(ClusteringData <TData> other)
{
return(this.Value > other.Value
? 1
: this.Value < other.Value ? -1 : 0);
}
public bool Equals(ClusteringData <TData> data)
{
return(data != null &&
(ReferenceEquals(data, this) ||
(Math.Abs(this.Value - data.Value) < Epsilon)));
}
