PDA

View Full Version : Any mountain clustering method in Weka?



campaxic
02-14-2008, 07:58 AM
Dear all,
I wonder which Weka algorithm could perform a mountain clustering in an spatial grid, in other words, having the coordinates of a map (x,y ... or even more) and then the altitude/height at each point I would like to localise the peaks of that topological surface. Which Weka algorithm can perform this? I see it as a clustering problem where the peak should be at the center of each cluster.


Thanks a lot,

Josep Maria,
[SIZE=2][FONT=Arial]
---------------------------------------------
Josep Maria Campanera Alsina
Investigador Juan de la Cierva
Departament de Fisicoqu

Mark
02-21-2008, 05:44 PM
Weka doesn't have an implementation of mountain clustering. It does have an implementation of a naive EM-based density estimator for clustering. It is naive because attributes are considered to be independent of one another given the cluster.

Cheers,
Mark.

campaxic
02-22-2008, 07:56 AM
Thanks Mark,

I'd like to reformulate my question:

I wonder which Weka algorithm could perform a search of local maxima in an spatial grid, in other words, having the coordinates of a map (x,y,z... up to 6 coordinates) and then the altitude/height at each point (h) (in total 7 numerical variables) I would like to localise the peaks (local maxima) of that topological surface. Which Weka algorithm can perform this?

Thanks a lot,

Josep Maria,

Mark
02-24-2008, 05:02 PM
Hi Josep,

I would expect that the SimpleKMeans clusterer would find clusters centered around the peaks in this data. The only problem is that you need to tell it how many maxima (i.e. clusters) to find. You could also try the EM clusterer. This algorithm has an option for automatically finding the best number of clusters according to goodness of fit (maximum likelihood) via cross-validation.

Cheers,
Mark.