R help - Feb 2011 - kmeans: number of cluster centres must lie between 1 and nrow(x)

Dear R,

Can't I cluster a dataset into k clusters where k is exactly the number of
observations? I have version 12.2 installed. See this example
> a <- matrix(1:100, 20)
> kmeans(a, 20)
Error: number of cluster centres must lie between 1 and nrow(x)

This is a bit ad-hoc but I known R from version 2.12 allows number of
clusters to be one. So I guess allowing number of clusters to be nrow(x)
should be also possible in the future release?


Feng

-- 
Feng Li
Department of Statistics
Stockholm University
106 91 Stockholm, Sweden
http://feng.li/

	[[alternative HTML version deleted]]

Thank you for the suggestion and it is exactly as you said only one
observation in each cluster. I know I can avoid this anyway and I am just
out of curiosity of the error.

I am writing a special algorithm to cluster some datasets with different
numbers of observations. For some particular datasets, there is only one
observation(e.g. people died of a rare disease). kmeans() thus will not work
at this situation.


Feng

2011/2/2 Rafael Björk <rafael.bjork@gmail.com>
> If you change the algorithm, the function allow you to do this:
>
> > kmeans(a, 20,algorithm="Lloyd")
>
> Maybe i'm missing something here, but why would you want to create as
many
> clusters as there are observations? Won't the outcome just be one
> observation in each cluster?
>
>
>
>

-- 
Feng Li
Department of Statistics
Stockholm University
106 91 Stockholm, Sweden
http://feng.li/

	[[alternative HTML version deleted]]