Sunday, March 27, 2011

2011 Algorithmic March Madness: Machines Lock in Victory over Humans

Posted by Danny Tarlow
Well that was an exciting and surprising weekend of basketball. (8)Butler beat (2)Florida in overtime, and (11)Virginia Commonwealth handily took care of (1)Kansas, eliminating the last remaining #1 seed. Rounding out the Final Four are (3)UConn and (4)Kentucky.

March Madness is usually good for some Cinderella stories, but this Final Four seems particularly improbable. There are no #1 seeds remaining (this has been the case 3 times in March Madness history, according to the TV announcers), and never before have teams seeded as low as #11 and #8 met in a Final Four game.

None of the entries in our second annual March Madness Algorithm Challenge saw this amount of madness coming. Two entrants correctly predicted that (3)UConn would make the final four (Team Delete Kernel and The Pain Machine), but no other algorithms or baselines got any Final Four teams correct. In fact, the only entry that has any more chance at points is The Pain Machine, which has UConn winning one more game.

So the question of which algorithm will win the contest is still not settled: a UConn victory on April 2, and The Pain Machine walks home with the prize; a UConn loss, and Team Delete Kernel is our winner.

What is settled at this point is that a machine will claim victory over the human-aided competition. The human baselines include our commissioner Lee's bracket; the Higher Seed bracket (where the human intervention came via the committee that chose seeds); and the Nate Silver baseline, which was a part-human, part-computer effort.

To give you an idea of the potentially winning methodologies, Scott Turner (The Pain Machine) describes his approach here, and Kevin Lee (Team Delete Kernel) based his model on the method described here.

So it's premature to congratulate a winner yet, but let me tritely say that I, for one, welcome our new March Madness algorithm overlords.
ENTRANT                         R1  R2  R3  R4  R5  Winner      Pts  Possible
Team Delete Kernel              23  20  16  8   -   Ohio St.    67   67
Human (Lee)*                    25  18  16  0   -   Kansas      59   59
Baseline (Higher Seed)*         25  20  12  0   -   Ohio St.    57   57
The Pain Machine                19  18  8   8   -   Kansas      53   69
Baseline (Nate Silver)*         25  20  8   0   -   Ohio St.    53   53
InItToWinIt                     22  20  8   0   -   Kansas      50   50
Baseline (TrueSkill)            26  18  4   0   -   Ohio St.    48   48
Danny's Dangerous Picks         22  16  8   0   -   Duke        46   46
Baseline (LRMC)                 25  16  4   0   -   Ohio St.    45   45
DukeRepeats                     23  16  4   0   -   Duke        43   43
Point Differential Centrality   23  16  4   0   -   Ohio St.    43   43
dirknbr1                        23  8   4   0   -   Ohio St.    35   35

* Denotes human-involvement.
You can see the full brackets here (Update: actually, it looks like Yahoo took them down). Also, there is a second, Sweet 16 contest that we haven't mentioned lately. Stay tuned for an update on that front.


Scott Turner said...

This is only the second year that no #1 has made it to the Final Four. The only other time it happened was during GMU's run, and in that case there was a #2, #3, and #4. So this year is very anomalous, and maybe not the best year for judging predictive programs! The PM has already wrapped up the Second Chance Contest, but in large part its success is due to riding UConn to the Final Four (and hopefully the Championship game :-), so I'm not sure that proves anything other it happened to get lucky. It's overall record in the tournament certainly isn't heartening.

Scott Turner said...

I was slightly wrong -- there was also no #1 seed in 1980. UCLA was a #2 that year, losing to eventual champion Florida.