R help - Jun 2010 - rules for optimizing samples in CLARA (size and numbers) ?

Hi,

With  a 9000 observations dataset, I have noticed a significant variability
in the silhouette index when I change the default value for samples (5
default value) and sampsize (40+2*clusters number) in CLARA.

Is there somes rules according to the number of cluster and observations to
fix samples and sampsize parameters efficiently, so as to avoid under- and
oversampling with CLARA in one hand and keeping a good time running in other
hand ?

I didn't not find any rules of this type on the web (except avoiding biaised
samples...).

Gratefully yours.
vincent

	[[alternative HTML version deleted]]