Other pages introducing user commands of ClusterTools: cluster routines, classification, reclustering, evaluation, support

The below routines change one multilevel clustering LABC_IN into another, LABC_OUT, e.g.

LABC_OUT = reclustn(LABC_IN);
LABC_OUT = LABC_IN*reclustn(LABC_IN);

Some of these routines have additional parameters. Below more information is given on multilevel clustering.

reclusth An an arbitrary, preferably diverse set of clusterings is used to create a hierarchical set of clusterings based on an agglomerative procedure.
reclustk A set of clusterings with different sizes is restructured into a set of user defined sizes.
reclustn An arbitrary set of clusterings is made nested. i.e. all objects in smaller clusters in one clustering are always in the same larger cluster in another clustering.

Other pages introducing user commands of ClusterTools: cluster routines, classification, reclustering, evaluation, support

Multilevel clustering

Cluster procedures in ClusterTools always generate a set of prototypes. A prototype is an object that is representative for a particular cluster. For every object a pointer is stored to the prototype of the cluster it belongs to. The resulting set of pointers is called a clustering and usually stored as a column vector. With the size of a clustering usually the number of clusters is meant.

All cluster routines in ClusterTools produce by default a set of clusterings of different sizes. These clusterings are different as the routines search for a different numbers of clusters, are based on different settings of smoothing parameters or are based on a set of thresholds in an agglomerative procedure.This is illustrated by the following example.

randreset;         % set random seeds for reproducability
A = gendatclust;   % 10 cluster example, 1000 objects
LABC = clustk(A,[2 3 5 10 20]); % K-Means cluster analysis
disp(size(LABC))   % size of array with clusterings
clustnum(LABC)     % number of clusters per clustering
clustsizes(LABC)   % cluster sizes in every clustering
scatn(LABC(:,4),A) % scatter plot with clusters for K = 10;