[Numpy-discussion] Faster

Keith Goodman kwgoodman@gmail....
Sat May 3 20:24:16 CDT 2008


On Sat, May 3, 2008 at 5:56 PM, Hoyt Koepke <hoytak@gmail.com> wrote:
> You could also try complete linkage, where you merge two clusters
>  based on the farthest distance between points in two clusters instead
>  of the smallest.  This will tend to get clusters of equal size (which
>  isn't always ideal, either).  However, it also uses sufficient
>  statistics, so it will be trivial to change your code to use that
>  merge criteria if you want to try it.

Thanks for the tip. The cluster sizes are much more reasonable now.


More information about the Numpy-discussion mailing list