AAAI Computer Poker Championship

[July 2006] The University of Alberta team has won the first ever AAAI poker bot competition, in convincing fashion, winning every match it played in both tournaments. The following summary crosstables give the average win rates, in small bets per game:

Normal Pace Alberta Bluffbot Monash Teddy Average
Alberta x +0.05 +0.72 +0.41 +0.39
Bluffbot -0.05 x +0.53 -0.19 +0.10
Monash -0.72 -0.53 x +1.17 -0.03
Teddy -0.41 +0.19 -1.17 x -0.46

Slower Pace Alberta Bluffbot C M U Monash Average
Alberta x +0.11 +0.18 +0.73 +0.34
Bluffbot -0.11 x +0.12 +0.52 +0.18
C M U -0.18 -0.12 x +0.65 +0.12
Monash -0.73 -0.52 -0.65 x -0.64

Numerous duplicate matches were played in each case, in order to obtain statistically significant results. In the following detailed crosstables, the "~" entry indicates the +/- one standard deviation bounds for each outcome.

Normal Pace Alberta Bluffbot Monash Teddy Average
Alberta x  +0.0514 
 ~0.0171 
 +0.7227 
 ~0.0161 
 +0.4067 
 ~0.0247 
 +0.3936 
 ~0.0193 
Bluffbot  -0.0514 
 ~0.0171 
x  +0.5271 
 ~0.0197 
 -0.1895 
 ~0.0289 
 +0.0954 
 ~0.0219 
Monash  -0.7227 
 ~0.0161 
 -0.5271 
 ~0.0197 
x  +1.1678 
 ~0.0427 
 -0.0273 
 ~0.0262 
Teddy  -0.4067 
 ~0.0247 
 +0.1895 
 ~0.0289 
 -1.1678 
 ~0.0427 
x  -0.4617 
 ~0.0321 

Slower Pace Alberta Bluffbot C M U Monash Average
Alberta x  +0.1145 
 ~0.0375 
 +0.1843 
 ~0.0079 
 +0.7344 
 ~0.0290 
 +0.3444 
 ~0.0248 
Bluffbot  -0.1145 
 ~0.0375 
x  +0.1200 
 ~0.0561 
 +0.5214 
 ~0.0408 
 +0.1756 
 ~0.0448 
C M U  -0.1843 
 ~0.0079 
 -0.1200 
 ~0.0561 
x  +0.6512 
 ~0.0397 
 +0.1156 
 ~0.0346 
Monash  -0.7344 
 ~0.0290 
 -0.5214 
 ~0.0408 
 -0.6512 
 ~0.0397 
x  -0.6357 
 ~0.0365 

Due to the short timeframe we had to prepare for the competition, we simply used a combination of two Sparbot programs (PsOpti4 and the new PsOpti6). The second place finisher was Teppo Salonen, a hobbyist programmer who built his own version of Sparbot for the Poker Academy software platform.

A highly-publicized entry from Carnegie Mellon University did not perform up to expectations, losing to the U of A and to Bluffbot. No doubt the talented researchers at CMU will be back with a strong contender next year. Their new program will also need to play substantially faster to maintain the normal pace of Limit Hold'em. The special slow event (one minute per game) that was created to accomodate the CMU entry may not be repeated in future tournaments.

More details of the event are available in this press release, and at the competition website. A full analysis of all matches will be performed over the coming weeks.


[Computer Poker Research Group] [GAMES Group Home Page] [CS Department Home Page]

Last Update: August, 2006