Sunday, July 26, 2009

Nail biting finish!

I wonder what I'm going to do with my life now?! :-)

It seems the Netflix Prize competition is finally over. Nothing is absolutely final yet - the judges have to verify the results and confirm the actual winning team, but the deadline has come and gone just an hour ago and "The Ensemble" ended up at the top of the leaderboard. Just!


As things turned out I ended up having an insider's view of what was going on inside "The Ensemble" over the final few days leading up to the deadline. That was very interesting even if I wish I could have contributed more directly to the final submissions! Instead I ended up frying my brain through lack of sleep, and turned into a zombie like creature instead. On the morning of the last full day (here in NZ) I finally realised what I needed to do to get my "silver bullet" post-processing tool into a potentially more useful state, but I ended up running out of time (not to mention the fact that zombies can't code). I will continue with that little project though - even if only to find out what its potential could have been, one way or the other.

Congratulations to all the members of the two teams that cracked the 10% barrier.

Saturday, July 11, 2009

Traffic from the Netflix Prize

This is all rather too late, but I thought I better put something here that is slightly more relevant to the Netflix Prize than Eigenratios of Self-Interpreters which is where I've been sending people from the Netflix Prize leaderboard or forums for the last 18 months or so. (Makes my traffic stats look ever so slightly more impressive but probably doesn't help them win the $1,000,000 USD. Sorry about that people!)

Some of you may have come here because you're wondering how "See No Evil" leapt from 214th position to 39th in a single mighty bound. Quite impressive eh? Would you like to know my secret? Perhaps even grab have a copy of my code?

Well, sorry, but I have to disappoint you. Where I am now is actually much higher than where I really belong. This is because earlier today I processed a set of predictions from another team through my own blending system and then (eventually) also submitted the results using my existing team identity. This was a test to see if my blending system might produce a significantly better quiz RMSE than that of the other team. They'd noticed that this was indeed the case with another set of inputs that I'd found via the forums and then contacted me.

So, should I have created a new team to submit this test blend? At first I thought so, and in fact even went through the motions of doing so. But then it turned out the other party was asleep (somewhere on the other side of the world) when it came to the time when they needed to confirm their membership on this new (and probably temporary) "team". Plus, they'd also said I should just go ahead and use my own existing team account anyway. And so I did. Seems a bit wrong to me, but if everybody created a new team every time they wanted to submit a test of this nature then the leaderboard would probably be very messy by now.

Anyhow, now you know. My actual personal best on the Quiz was 0.8902 (June 15, 2008) and I was in 71st position at about that time. Truly! (But that was also only after Netflix cleaned out a bunch of duplicate teams!)

If you still want to talk about using my blending and polishing system to help you leap into top spot in the next two weeks, then you know who to call. Actually I still think its pretty good even if the Grand Prize Team aren't beating down my door. So you probably won't have to promise me a huge cut of the prize to get me to join your team also... :-)