Speaking of spinning, Mike, there is a bit of a move afoot to use the 0.3
release to do some *really* big SVD in order to claim a size record of
sorts. The goal is to find some realistic and interesting matrix with about
5 x 10^9 non-zero elements.
On Fri, Mar 5, 2010 at 8:05 PM, Jake Mannix wro
Hi Mike,
Welcome to the long journey down the road of dimensional reduction. :)
On Fri, Mar 5, 2010 at 5:05 PM, mike bowles wrote:
>
> Really large matrices require using one of the randomizing methods to get
> done.
"Require" is a strong term. Really really large (but still sparse!)
matric
Mike,
http://issues.apache.org/jira/browse/MAHOUT-180 might be of interest. Jake
has done a fair bit of work beyond that.
Next up is a stochastic decomposition version. You can see the seeds of
that in Jake's other JIRA's.
On Fri, Mar 5, 2010 at 5:05 PM, mike bowles wrote:
> ... I thought i
I've been trying to figure out how to code an svd algorithm and I've seen
some questions about svd algorithms floating around the Mahout mailing
lists. I thought it might be helpful to share what I've found so far.
Really large matrices require using one of the randomizing
