2007 Poker Bot Competition Summary
The 2007 poker competition consisted of 15 competitors from 7 countries and 43 bots. Matches were played on 32 machines running for a month, playing over 17 million hands of poker. The results were announced at AAAI 2007 on July 24, 2007 in Vancouver BC.
Limit Series (Equilibrium) and Limit Bankroll (Online) - 33 bots from 13
13.7 million hands played
No Limit - 10 bots from 8 competitors
3.4 million hands played
Note: some bots played in both the equilibrium and online competitions
|BluffBot2.0 (Teppo Salonen)
Teppo Salonen (who designed Bluffbot, one of the top limit bots from last year) developed BluffBot2.0, the only no-limit bot this year which went undefeated heads-up. BluffBot2.0 used a far wider variety of betting amounts which worked to its advantage.
|GS3 (Carnegie Mellon)
GS3 used automated abstraction equilibrium computation techniques to develop a very strong competitor this year. Due to a bug discovered a week before the competition, they had to re-learn an equilibrium from scratch in a week. This team can be expected to put forward an even stronger program in future years.
|Hyperborean07 (University of Alberta, Canada)
The University of Alberta put forward another very solid program: amongst the top three, Hyperborean07 No-Limit had the highest average bankroll on all its hands.
|Hyperborean07 Online2 (aka Qeqbot, University of Alberta, Canada)
The top online bot was "qeqbot" (Hyperborean Online 2), a static "quasi-equilibrium" strategy based on pdf-cutting. Although not as close to a true equilibrium as the larger solutions, the system has the flexibility to invoke a wide variety of playing styles.
|Hyperborean07 Online1 (University of Alberta, Canada)
The second place Hyperborean online bot was a solid equilibrium bot. Amongst the top 4 bots, this bot had better performance than the top bot. However, against medium range and weaker bots, it did not fare as well.
|GS3 (Carnegie Mellon)
GS3's bot was the same in both limit competitions. Its equilibrium play was enough to do moderately well against all the weaker bots.
Top three programs in the equilibrium competition were very solid. Each abstracted the game of limit poker and then solved that abstraction. The edge obtained by the top bot was due to the fact that it did not abstract the betting space.
|Hyperborean07 (University of Alberta, Canada)|
|IanBot (aka INOT, Ian Fellows, University of California, San Diego)|
|GS3 (Carnegie Mellon)|
The units were "small bets/hand". To put this in perspective, always folding loses 0.75 small bets/hand. Each individual series is summarized below. The number is the amount (in terms of small bets/hand) the row player won from the column player. Green (positive) indicates a series where the row player won money. Red (negative) indicates a series where the column player won money.
Hand histories and data
The hand histories, results and data files can be found on the data page