Re: DM: equal-size clustering

[John Aitchison]

> > > > From: Hukan <hukan@cs.hku.hk>
> > > >    I have a special clustering problem. Given a set of points 
>in the
> > > > multidimensional space, we want to cluster these points under 
>the
> > > > limition that the sizes of clusters are (almost) equal. Could 
>anyone
> > > > give me some suggestions?

There have been various suggested approaches to this problem .. Snob, 
neural nets etc.

Might I make a simpler suggestion. Since presumably there is some 
administrative or external reason that the cluster sizes must be 
equal (and therefore that external 'reality' is known to require 
equal 
sized clusters, or that a sub optimal but equal-cluster-sizes 
solution is acceptable) would not a simple exchange algorithm 
suffice?. (along the lines of .. partition  the data into M equal 
sized 
groups, exchange case pairs until some criterion is minimized etc). 

If running time is a problem, some obvious heuristics can be 
employed. Spath (Cluster Dissection and Analysis) has some algorithms 
along these lines.

fwiw




 

 
John Aitchison <jaitchison@acm.org>
Data Sciences Pty Ltd
Sydney, AUSTRALIA.