1 / 23

2006 AAAI Computer Poker Competition

2006 AAAI Computer Poker Competition. Michael Littman Rutgers University. Martin Zinkevich Christian Smith Luke Duguid U of Alberta. What is Poker?. The OR View A partial information game with over 10 18 states. The AI View A huge opponent modeling challenge. The Public View

saman
Télécharger la présentation

2006 AAAI Computer Poker Competition

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 2006 AAAI Computer Poker Competition Michael Littman Rutgers University Martin Zinkevich Christian Smith Luke Duguid U of Alberta

  2. What is Poker? • The OR View • A partial information game with over 1018 states. • The AI View • A huge opponent modeling challenge. • The Public View • Really popular! Lots of fun!

  3. What is the Game? • Play 1000 hands of Heads-up Limit Texas Hold’em Poker against an opponent • Reset bots, switch seats, and play again with the same hands. • lower variance • more fair • Repeat 6-20 times • Can treat the outcome of the duplicate match as a random variable which we have sampled several times.

  4. How do you win? • Make money! • Bankroll Competition • Against overall, have the maximum total bankroll • Highlights opponent modeling and learning • Series Competition • Against individuals, have positive bankroll • Highlights the OR problem of “solving” the game

  5. The Teams • Hyperborean: University of Alberta Edmonton, Canada • Bluffbot: Finland • Monash: Monash University, Australia • Teddy: USA • Gs2: Carnegie Mellon University, Pittsburgh, USA

  6. Before I Begin • There is a period that is allocated for competitors to contest the results that has yet to expire.

  7. Bankroll Results

  8. Bankroll Results

  9. Bankroll Results

  10. Bankroll Results

  11. Bankroll-Heads Up

  12. Bankroll-Heads Up (Small Bets/Hand)

  13. Bankroll-Overall Significance • Difference between Bluffbot and U of Alberta • 0.2982 small bets/hand • 0.0190 standard deviations

  14. The Most Interesting Result • In the bankroll competition, in head-to-head, BluffBot beat Monash who beat Teddy who beat BluffBot • A practical example of the non-transitivity of poker

  15. Results-Series (Small Bets/Hand)

  16. Results-Series (Small Bets/Hand)

  17. Things to Never Assume • All bots will work on the competition machines the first time • The server code is bug-free • Everybody has a common idea of the rules of poker (or even heads-up Texas Hold’em) • People can write code instantaneously

  18. Exhibitionary Aspects of the Competition • Bots submitted late • Bots debugged after the deadline • The time limit was very large for the series competition

  19. Maybe Next Year • More advance notice • Competitors need access to one of the machines they will use • Server code needs to be frozen before the competition begins • More variants of poker need to be included, especially: • >2 players • >1000 hands • There has to be a 7 sec/hand time limit

  20. Is Poker “Solved”? • No one has ever solved a four-round abstraction of poker without a partition into the early game and the late game. • The game of poker is also about your opponent. For instance, playing rock-paper-scissors against a four-year-old is different than against an adult.

  21. Next Year • This summer: write the rules • This winter: write the code • Next summer: next competition

  22. Are You Interested? • Come talk to us • maz@cs.ualberta.ca • bowling@cs.ualberta.ca

  23. Summary • The poker competition brought together five teams in two competitions. • The competition was very close, and very interesting, non-transitive (A beats B beats C beats A) performance was observed. • A freeware codebase was developed for future competitions.

More Related