Monday, February 2, 2015

Kaggle Competition & Machine Madness

The second annual Kaggle competition has been announced:
Kaggle & HP are hosting our second annual NCAA March Mania prediction competition. All methods, flavors, credos, and styles of prediction are legal, guessing included. Here's how it works:
  • You're given team-level data going back three decades, and are welcome to use whatever data you want
  • We have a historical "warm up" stage so you can get used to the data format and have some motivation to work on the problem before March
  • On Selection Sunday (when teams are announced), we wipe the leaderboard and collect 2015 predictions
  • Entries are judged on the log loss of predicted probabilities for every possible matchup in the tournament.
  • We update scores as the games unfold
Have fun!
The home page for the competition is here.

I'll also remind everyone that I'll be hosting the annual Machine Madness Competition. Machine Madness has been going on since 2010. Unlike the Kaggle competition, you make one entry into a Yahoo pool and compete under the usual pool rules. So the strategy is very different from the Kaggle competition.  And you're probably not going to win either competition, so it's worth it to take a few extra minutes and join the Machine Madness competition as well :-)


  1. Just a reminder that if you'd like we can run the Machine Madness on my site ( instead of Yahoo (I don't make any money or anything off of my site). As you know and we've discussed in the past, the normal 1-2-4-8-16-32 scoring is pretty boring AND heavily weighted to picking the champion. With my site we can customize scoring and reduce late-round points (to reduce the reliance on picking the champ) and institute some sort of underdog bonus (to spice things up but more importantly to increase diversity among picks). There's only 63 games and in a normal scoring bracket pool we "waste" probably 1/3 to 1/2 of the games since everyone just picks the same team...changing the scoring a bit should allow us to maximize the chances the better predictors actually come out on top, though of course with March Madness you can never fully escape the unpredictability.

  2. Yes, Monte, let's use your site. I'll send you an email. And we can discuss the scoring system as well.