Not surprised. It is still roughly an O(n^3) algorithm (n = number of users).

A lower number will increase speed. You might want something far lower
like 0.05.

One other way to speed this up is to use a different metric for
cluster similarity, like computing a centroid for the cluster and
comparing that rather than the O(n^2) comparison done now.

This sort of approach, I haven't followed up on much, to be honest.
There are some areas for improvement to be sure.

On Fri, Oct 31, 2008 at 7:21 PM, Otis Gospodnetic
<[EMAIL PROTECTED]> wrote:
> Quick report that, sadly, TreeClusteringRecommender 
> (TreeClusteringRecommender2 actually!) is a no go, too.  It's been running 
> for well over an hour over the same amount of data, and this is where its 
> been spending its time:

Reply via email to