Clustering with 'agnes'

Hello,

I had a question regarding clustering using the agnes() function from the 'cluster' package.

I was wondering if anyone knew how I can identify cluster points after running the agnes function. 

For example, I created a dataset with points randomly scattered around (0,0), (0,1) and (1,0). After clustering, the dendrogram shows all the clustered points and I get the ordering and height and the agglomerative coefficient. But nowhere do I see the three actual points listed. Although agnes clusters until there is one main cluster, it is clear that at three clusters, each of the clusters consist of points around the three main points. I was wondering if there was any way in which I can have R give me the actual cluster points at three (or any number, for that matter) clusters, ie (0,0), (0,1) and (1,0). A visual display of the clusters would be even better.

I have tried using idenfity after converting the agnes object to an hclust object, but that only gives me a listing of the points in each cluster.

I hope this question is clear. I am a little new with both clustering and using R for clustering, so please ask me to clarify if anything is unclear.

Your help would be most appreciated!

With regards,
Arnav



	[[alternative HTML version deleted]]

Arnav Sheth wrote:

Hello,

I had a question regarding clustering using the agnes() function from the

'cluster' package.

I was wondering if anyone knew how I can identify cluster points after

running the agnes function. 

For example, I created a dataset with points randomly scattered around

(0,0), (0,1) and (1,0). After clustering, the dendrogram shows all the
clustered points and I get the ordering and height and the agglomerative
coefficient. But nowhere do I see the three actual points listed. Although
agnes clusters until there is one main cluster, it is clear that at three
clusters, each of the clusters consist of points around the three main
points. I was wondering if there was any way in which I can have R give me
the actual cluster points at three (or any number, for that matter) clusters,
ie (0,0), (0,1) and (1,0). A visual display of the clusters would be even
better.

I have tried using idenfity after converting the agnes object to an hclust

object, but that only gives me a listing of the points in each cluster.

I hope this question is clear. I am a little new with both clustering and

using R for clustering, so please ask me to clarify if anything is unclear.

Your help would be most appreciated!

See the example of ?agnes, where the points are labeled.
Most easily use agnes() on a data.frame with rownames.

Uwe Ligges

With regards,
Arnav



	[[alternative HTML version deleted]]

Hi Uwe,

Thanks for the tip. I already have row labels. My problem is, (referring to the 
example below) how can I get R to tell me that upto three clusters, the points 
are clustered around (0,0), (1,0) and (0,1)?

Perhaps it is not even possible, I am not sure.

With regards,
Arnav

Hi,

the underlying principle of hierarchical clustering is *not* that the
clusters can be represented by some centroid points. Most methods are
distance based, i.e. they can be calculated also in absence of any R^p
representation of the points.

If you want to recover centroids, you should do kmeans, normal mixture
clustering (mclust) or pam/clara. Of course you can also take the points
belonging to an agnes cluster and compute the mean vector (or any other
summary statistic), but that's not what hierarchical clustering is
meant to do (it may be reasonable with Ward's method, though).

Christian

On Wed, 4 Feb 2004, Arnav Sheth wrote:

Hi Uwe,

Thanks for the tip. I already have row labels. My problem is, (referring

example below) how can I get R to tell me that upto three clusters, the

are clustered around (0,0), (1,0) and (0,1)?

Perhaps it is not even possible, I am not sure.

With regards,
Arnav

***********************************************************************
Christian Hennig
Fachbereich Mathematik-SPST/ZMS, Universitaet Hamburg
hennig at math.uni-hamburg.de, http://www.math.uni-hamburg.de/home/hennig/
#######################################################################
ich empfehle www.boag-online.de