You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@mahout.apache.org by Ted Dunning <te...@gmail.com> on 2010/04/26 23:38:25 UTC

Re: [jira] Commented: (MAHOUT-305) Combine both cooccurrence-based CF M/R jobs

On Mon, Apr 26, 2010 at 1:46 PM, Sean Owen (JIRA) <ji...@apache.org> wrote:

> Ted how do you like to pick which items to pay attention to for
> co-occurrence? I'm looking for something simple to start.
>

LLR is my standard answer.

>
> Though it's running pretty well (well a lot better than it was) at the
> moment, with the aggressive combiner chucking out low-frequency
> co-occurrence.
>

That still worries me.  I would expect that you would get better by
down-sampling high frequency items.