mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jeff Eastman <jeast...@Narus.com>
Subject RE: Kmeans clustering in multiple levels
Date Thu, 10 Feb 2011 17:52:10 GMT
I doubt one iteration will give you what you are seeking. Better to use a larger convergence
value to terminate the iterations sooner rather than a small maxIterations limit. This will
likely require some experimentation to arrive at the best values.

-----Original Message-----
From: Veronica Joh [mailto:vj8211@hotmail.com] 
Sent: Thursday, February 10, 2011 9:46 AM
To: user@mahout.apache.org
Subject: Kmeans clustering in multiple levels


Hi
In the manning book, it is suggested that we run clustering in multiple levels when clustering
large number of articles.  For example, if we were to cluster 1 million articles, we would
first cluster it into 100 giant clusters and cluster again to get smaller clusters.
My question is when we run cluster for each level, what should be the maxIteration?  Is doing
one iteration enough for each level?
Thank you,Veronica 		 	   		  

Mime
View raw message