On Aug 1, 2008, at 5:19 PM, Sean Owen wrote:
Oh OK let me try to run through the steps myself to see what happens.
Yeah I hadn't completely merged into the main build file. Well, I have
in the sense that the code compiles and builds through that. I left
the more project-specific targets in a separate file.
I am not against combining it all, in fact, I was just worried about
polluting the main build file.
No worries about polluting. We've got a few releases to work out
things, etc. I think we will have a clean up at some point.
What kind of compile errors do you get? maybe I can diagnose from
that.
Pretty much looks like missing Hadoop jars. I see the problem. We
upgraded the hadoop jar.
Yes I committed two integrations with Hadoop: one, a job that will run
n copies of a Recommender and use it to generate recommendations for
users. So this is a way to crudely parallelize recommending items to
all users at once in batch. This would be good for recomputing
everyone's recommendations, say, every night.
I also added a job to compute item-item pref diffs, which is an
expensive preprocessing phase for slope-one recommenders. This is very
amenable to Hadoop even if may be of somewhat narrow interest.
It works locally but in honesty I have not tried it on a real
cluster yet.
No worries, I _may_ be doing a session tomorrow at BarCampRDU on
Mahout and Hadoop and someone asked about Taste, so I thought it would
be cool to know a thing or two more about it than I currently do and
what it's status is.
-Grant
