RECLUSTK

Recluster multilevel clustering to determine exactly K clusters

   LABOUT = RECLUSTK(LABIN,K,LABREF)

Input
LABIN	[M,L] array of indices for L clustering levels of M objects.  Indices should point to the prototype objects of its cluster.
K	Scalar or vector of length N with desired numbers of clusters.  Default is a set of N clusterings between 2 and M/10 clusters.  To reduce computing time small K's (eg K < 100) are recommended.
LABREF	[M,1] vector of a high-resolution clustering to be used for  estimating cluster distances. If omitted a high-resolution  clustering from LABIN will be used.

Output
LABOUT	[M,N] array with the results of the multilevel clusterings of the
M	objects. The columns refer to the N clusterings. They yield for  the objects the prototype indices of the clusters they belong to.

Description

The multilevel clustering LABIN is used to find a set / sets of exactly K clusters. LABREF should be a set of)high-resolution clustering(s),  obtaining many more clusters than demanded in K. It is used to estimate  distances by CLUSTDIST between the clusters in a clustering in LABIN that  has slightly more clusters than desired by K. In this clustering the  clusters are combined using a single linkage procedure (DCLUSTH) based on  the computed cluster distances.

This routine might be usefule for clustering procedures like CLUSTM and  CLUSTS for which the number of desired clusters can not exactly be  defined.

For large values of K this routine might be very time consuming.

See also

datasets, mappings, dclusth, cluste, clustf, clustk, clusth, clustkh, clustm, reclusth, reclustn, testclust, clustcerr, clustc, clustnum, plotdg,