2009 Poker Bot Competition Summary

Slideshow of the results: Click here.

The 2009 poker competition consisted of 14 competitors representing 10 different universities from 7 countries along with 4 independent entries. There were 13 bots submitted to the limit competition, 5 to the no limit and 7 to the ring.

About 150 CPU months of computation were used to run the competition. There was a results and discussion session at IJCAI 2009 on July 15, 2009 in Pasadena, CA.

A list of all the competitors and some info about each of them is available here

A description of each of the winning bots (from the authors) is available here

For the crosstables below:The units are "small bets/hand" for limit games, and "small blinds/hand" for no limit. To put this in perspective, in heads up games always folding loses 0.75 small bets/hand, or 1.5 small blinds/hand, while in 3 player games always folding loses 0.5 small bets/hand or 1 small blind/hand. 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. The +/- given is 1 standard deviation.

Some basic info:

  • Limit run-off and limit bankroll - 13 bots from 11 competitors
    75 matches of 3000 hands (duplicate) played between each bot, with an additional 45 matches played between some bots for better significance in results
  • No Limit - 5 bots from 3 competitors
    60 matches of 3000 hands (duplicate) played between each bot
  • Limit 3 player - 7 bots from 5 competitors
    110 matches of 1000 hands (6-way duplicate) played between each matchup of bots not from the same team. 40 matches of 1000 hands (6-way duplicate) were played for all possible matchups involving 2 bots from the same team at the same table. These were not used to compile the official results, however analysis using these matches is provided below.

Results

No limit run-off

First Place HyperboreanNL-Eqm (University of Alberta)
Second Place HyperboreanNL-BR (University of Alberta)
Third Place BluffBot4 - Teppo Salonen

No limit bankroll

First Place Hyperborean-BR (University of Alberta)
Second Place Hyperborean-Eqm (University of Alberta)
Third Place BluffBot4 (Teppo Salonen)

Crosstables:

limit run-off

First Place GGValuta (University of Bucharest)
Second Place Hyperborean-Eqm (University of Alberta)
Third Place MANZANA (Marv Andersen)

limit bankroll

First Place MANZANA (Marv Anderson)
Second Place Hyperborean-BR (University of Alberta)
Third Place GGValuta (University of Bucharest) - by 0.001 over Hyperborean-Eqm with +/- 0.002 as 1 standard deviation

Crosstables:

3 player limit run-off

First Place HyperboreanRing-Eqm (University of Alberta)
Second Place HyperboreanRing-BR (University of Alberta)
Third Place dpp (Technical University Darmstadt)

3 player limit bankroll

First Place HyperboreanRing-Eqm (University of Alberta)
Second Place HyperboreanRing-BR (University of Alberta)
Third Place akuma (Technical University Darmstadt)

Crosstables:

Note: The run-off stopped with 5 players left due to the fact that matches where 2 of the 3 bots were submitted by the same team were not considered in these results and the top 5 bots contained two entries from University of Alberta and two entries from Technical University Darmstadt.

3 player with teams at the same table

Below are the same crosstables as above for the case when bots submitted by the same team are allowed to sit at the same table. Only 40 iterations of these matches were run, as opposed to 110 for the rest of the bots. This actually changes the order of the bots: dpp is beaten by akuma, another bot submitted by the same team, by enough that in the run-off it drops out in 5th and CMU jumps to 4th. Also in the bankroll akuma now swaps places with dpp. Hyperborean-Eqm continues to edge out Hyperborean-BR for the top spot.

Crosstables with teams allowed at the same table:

Hand histories

Update 28th July: The limit hand histories below do not contain any matches involving dcurb or tommybot. This was an accident - these extra matches have now been added below. Also note that there may be a few log files of matches number 120 or greater. These are extraneous matches that were not used in calculating the results, and their correctness has not been verified. They were included by accident, and you may use them as you wish.

The hand histories and results files are available here. Please note that some of these files are quite large.

The crosstables for limit and no limit were compiled using "DataTest" from the pokerserver software (under statistics), and the ring crosstables were compiled using this script.

For 2008 results, click here