There is an implementation of an algorithm called ALS-WR which was
developed for the Netflix-Prize. It contains a parallelized matrix
factorization, but the algorithm itself is iterative which is not a good
fit for the MapReduce paradigm, so this implementation should be
considered experimental at best currently.
--sebastian
On 09.05.2011 19:26, Sean Owen wrote
What you see in trunk now is basically what you'll get.
I know Sebastian has done a lot of work on this piece since 0.4, yes,
so I imagine the answer is 'yes' it's new and different and perhaps
better for your purpose.
On Mon, May 9, 2011 at 6:03 PM, James James<[email protected]> wrote:
Hi,
I was wondering if the new release 0.5 will include the new implementation of a
SVD-based recommender? I know there exists an old implementation in Mahout, but
it does not really scalable.
Thanks,
James
________________________________
From: Sean Owen<[email protected]>
To: Mahout Dev List<[email protected]>
Sent: Mon, May 9, 2011 4:30:34 AM
Subject: Re: [VOTE] Code freeze for 0.5 release May 20
OK, hearing no objections in a few hours, this will have passed. If
so, only fixes related to 0.5, or test/doc/checkstyle improvements,
from here to release.
On Fri, May 6, 2011 at 3:25 PM, Sean Owen<[email protected]> wrote:
https://issues.apache.org/jira/secure/IssueNavigator.jspa?reset=true&jqlQuery=project+%3D+MAHOUT+AND+fixVersion+%3D+%220.5%22+AND+status+%3D+Resolved+ORDER+BY+priority+DESC&mode=hide
e
127 issues resolved for 0.5, and 6 months since 0.4 was released. Yep,
pretty good time for 0.5.
I suggest we enter a code freeze upon passing of this vote. The only
changes will be fixes, more tests, doc improvements. I'll act as
release engineer then to head up the release (with helping hands along
the way) per https://cwiki.apache.org/MAHOUT/how-to-release.html
targeting a release on May 20.
What say?
Sean
