Thursday, December 30, 2010

2010 computer poker comptetition

Source: http://www.computerpokercompetition.org/index.php?option=com_content&view=article&id=64:participants-2010&catid=35:participants&Itemid=62&limitstart=1

Participants: 2010 - Heads-up Limit Texas Hold'em
Friday, 19 March 2010 19:39
Last Updated on Thursday, 29 July 2010 18:24
Written by Administrator
Article Index
Participants: 2010
Heads-up Limit Texas Hold'em
Heads-up No-limit Texas Hold'em
3-player Limit Texas Hold'em
All Pages
Page 2 of 4
Heads-up Limit Texas Hold'em
Arnold2

Team Leader: Victor Saase
Team Members: Victor Saase
Affiliation: Independent
Location: Zurich, Switzerland
Technique: Monte Carlo Regret Minimization with Imperfect Recall
ASVP

Team Leader: Nan-chen Chen
Team Members: Christine Pei-jinn Chai, Nan-chen Chen, Chun-yen Ho, I-yen Su
Affiliation: National Taiwan University
Location: Taipei, Taiwan
Technique: The only idea we use is sampling. We use sampling to estimate the probability of winning. If it is pretty high, we will raise. If it is pretty low, we will fold. Others we will call.
GGValuta

Team Leader: Mihai Ciucu
Team Members: Mihai Ciucu, Stefan Popescu, Leoveanu Mihaita, Diana Dulgheru
Affiliation: University of Bucharest
Location: Bucharest, Romania
GS6

Team Leader: Sam Ganzfried
Team Members: Sam Ganzfried, Andrew Gilpin, Tuomas Sandholm
Affiliation: Carnegie Mellon University
Location: Pittsburgh, Pennsylvania, USA
Hyperborean

Team Leader: Michael Bowling
Team Members: Nolan Bard, Michael Bowling, Neil Burch, Josh Davidson, Richard Gibson, John Hawkin, Rob Holte, Michael Johanson, Boyan Marinov, Dustin Morrill, Jonathan Schaeffer, Nathan Sturtevant, Duane Szafron, Martha White
Affiliation: University of Alberta
Location: Edmonton, Alberta, Canada
Technique: All of our two player entries are Nash equilibrium strategies within abstract games, generated using CFR - counterfactual regret minimisation (Zinkevich et al., NIPS 2007). For our limit bankroll entry we did some ad-hoc post-processing of the strategies in an attempt to eliminate defensive play.
Related Papers:
Martin Zinkevich, Michael Johanson, Michael Bowling, Carmelo Piccione. Regret Minimization in Games with Incomplete Information. In Advances in Neural Information Processing Systems 20 (NIPS), 2007.
Michael Johanson. Robust Strategies and Counter-Strategies: Building a Champion Level Computer Poker Player. M.Sc. Thesis, 2007.
Jester

Team Leader: François Pays
Team Members: François Pays
Affiliation: Independent
Location: Paris, France
Technique: The equilibrium is the solution of a scaled-down minimax problem.
sequence form problem formulation with k-means cards abstraction
cards bucketing: preflop:169, flop:400, turn:50, river:25
dedicated minimax (convex-concave) interior-point solver
newton system solved by PCG-like Krylov-subspace iterative method
solved in 5 days on Cuda / Nvidia Fermi GPU
LittleRock

Team Leader: Rod Byrnes
Team Members: Rod Byrnes
Affiliation: Independent
Location: Lismore, New South Whales, Australia
Technique: Little Rock is based on the regret minimization technique but with some important differences. It has been designed to be able to be used for games with any number of players, and as such the perfect recall of player actions that is used in the published technique is not suitable, as the number of states for games with more than 3 players becomes too large to compute. Instead Little Rock uses two abstractions. The first is a bucket abstraction, whereby each hand is placed into one of 31 buckets based on its potential (similar to the EHS squared metric). The second is a game state abstraction that takes into account number of players, player position, various actions that have taken place, as well as information about the pot size, bet/call ratio etc. By using these two abstractions much larger games are able to be solved, with the drawback that the agent may not be as effective against algorithms specifically tailored to games with fewer players (ie. 2 and 3 player games). Tests with Poker Academy show that Little Rock is able to convincingly beat Sparbot heads-up, and Pokibot in all game sizes from 3 to 9 players.
longhorn

Team Leader: Alan Lockett
Team Members: Alan Lockett, Risto Miikkulainen
Affiliation: University of Texas at Austin
Location: Austin, Texas, USA
PLICAS

Team Leader: Christian Friedrich
Team Members: Christian Friedrich, Dr. Michael Schwind
Affiliation: Technical University Kaiserslautern
Location: Kaiserslautern, Germany
Technique: PLICAS (latin for "you fold") is a Texas Hold'em Fixed Limit Heads-up Bot. It uses besides a classical rule-based decision process additional units which use case-based reasoning, pseudo-optimal play for specific situations and dynamic (adaptive) preflop hand selection as well as postflop agression control. The adaptive playing style is realized on basis of a detailed opponent model.
Related Papers: Friedrich, Schwind. PLICAS - A Texas Hold'em Poker Bot. Not yet published.
PULPO

Team Leader: Marv Andersen
Team Members: Marv Andersen
Affiliation: Independent
Location: London, UK
Technique: This bot is a neural net trained to play like a mixture of psuedo-equilibrium strategies
Rockhopper

Team Leader: David Lin
Team Members: David Lin
Affiliation: Independent
Location: New York, New York, USA
Sartre

Team Leader: Jonathan Rubin
Team Members: Jonathan Rubin, Ian Watson
Affiliation: University of Auckland
Location: Auckland, New Zealand
Technique: Sartre uses a case-based approach to play Texas Hold'em. Hand history data from the previous years top agents are encoded into cases. When it is time for Sartre to make a betting decision a case with the current game state information is created. The case-base is then searched for similar cases. The solution to past similar cases are then re-used for the current situation.
Related Papers:
Jonathan Rubin & Ian Watson. (2010). Similarity-Based Retrieval and Solution Re-use Policies in the Game of Texas Hold'em. In International Conference on Case-Based Reasoning (ICCBR 2010). To Appear.
Jonathan Rubin & Ian Watson. (2009). A Memory-Based Approach to Two-Player Texas Hold'em. In Proceedings of AI 2009: Advances in Artificial Intelligence, 22nd Australasian Joint Conference, pages 465-474, 2009.
Slumbot

Team Leader: Eric Jackson
Team Members: Eric Jackson
Affiliation: Independent
Location: Menlo Park, California, USA
Technique: Slumbot employs the fictitious play algorithm to find a near-equilibrium solution for (an abstraction of) heads-up limit Texas Hold 'Em. In the abstraction Slumbot uses, the game tree has roughly 3x10^13 nodes. Bins have been manually specified (e.g., two pair, aces and eights, on a monotone unconnected board) rather than automatically derived from some metric (e.g., hand strength squared).

Heads-up No-limit Texas Hold'em
c4tw

Team Leader: Thorsten Spieker
Team Members: Thorsten Spieker
Affiliation: Universitat Bamberg
Location: Bamberg, Germany
Technique: Strategy is based on the creation of an action sequence tree. Action sizes have been selected with expert knowledge. Nodes of the tree contain information about occurences of their state showdown nodes contain information about previously showndown hands after that particular action sequence. Decisions are based on calculating the EV of different action possibilities by simulating current holding versus the observed hands in the showdown nodes. Total Bankroll strategy was trained with a database of observed real-money hands. Instant Run-off strategy is a freshly created tree which updates during the game.
Hyperborean

Team Leader: Michael Bowling
Team Members: Nolan Bard, Michael Bowling, Neil Burch, Josh Davidson, Richard Gibson, John Hawkin, Rob Holte, Michael Johanson, Boyan Marinov, Dustin Morrill, Jonathan Schaeffer, Nathan Sturtevant, Duane Szafron, Martha White
Affiliation: University of Alberta
Location: Edmonton, Alberta, Canada
Technique: All of our two player entries are Nash equilibrium strategies within abstract games, generated using CFR - counterfactual regret minimisation (Zinkevich et al., NIPS 2007). For our limit bankroll entry we did some ad-hoc post-processing of the strategies in an attempt to eliminate defensive play.
Related Papers:
David Schnizlein. State Translation in No-Limit Poker. M. Sc. Thesis, 2009.
Martin Zinkevich, Michael Johanson, Michael Bowling, Carmelo Piccione. Regret Minimization in Games with Incomplete Information. In Advances in Neural Information Processing Systems 20 (NIPS), 2007.
Michael Johanson. Robust Strategies and Counter-Strategies: Building a Champion Level Computer Poker Player. M.Sc. Thesis, 2007.
PokerBotSLO

Team Leader: Bojan Butolen
Team Members: Bojan Butolen, Mitja Cof
Affiliation: University of Maribor & University of Ljubljana
Location: Maribor, Slovenia
Technique: In general the agent plays by a simplified decision tree. It uses the information received from the server to determine which phase of the game it is (PRE-FLOP, FLOP, TURN, RIVER). The decision for our agent’s next move is based on an evaluation function from the combination in our hand. Based on the evaluation we check our opponent’s last move and try to decide accordingly to some predefined limits for our own move. Additionally we observe our opponents playing style by making in game statistics of how many times the opponent viewed the flop, raised pre-flop, called pre-flop, etc. We use those statistics to view if the opponent plays more loose or tight, so we can try to steal blinds.
Related Papers: Jonathan Schaeffer, Darse Billings, Lourdes Peña, Duane Szafron. Learning to Play Strong Poker.
SartreNL

Team Leader: Jonathan Rubin
Team Members: Jonathan Rubin, Ian Watson
Affiliation: University of Auckland
Location: Auckland, New Zealand
Technique: SartreNL uses a case-based approach to play Texas Hold'em. Hand history data from the previous years top agents are encoded into cases. When it is time for Sartre to make a betting decision a case with the current game state information is created. The case-base is then searched for similar cases. The solution to past similar cases are then re-used for the current situation.
Related Papers:
Jonathan Rubin & Ian Watson. (2010). Similarity-Based Retrieval and Solution Re-use Policies in the Game of Texas Hold'em. In International Conference on Case-Based Reasoning (ICCBR 2010). To Appear.
Jonathan Rubin & Ian Watson. (2009). A Memory-Based Approach to Two-Player Texas Hold'em. In Proceedings of AI 2009: Advances in Artificial Intelligence, 22nd Australasian Joint Conference, pages 465-474, 2009.
Tartanian4

Team Leader: Sam Ganzfried
Team Members: Sam Ganzfried, Andrew Gilpin, Tuomas Sandholm
Affiliation: Carnegie Mellon University
Location: Pittsburgh, Pennsylvania, USA

Participants: 2010 - 3-player Limit Texas Hold'em
Friday, 19 March 2010 19:39
Last Updated on Thursday, 29 July 2010 18:24
Written by Administrator
Article Index
Participants: 2010
Heads-up Limit Texas Hold'em
Heads-up No-limit Texas Hold'em
3-player Limit Texas Hold'em
All Pages
Page 4 of 4
3-player Limit Texas Hold'em
Arnold3

Team Leader: Victor Saase
Team Members: Victor Saase
Affiliation: Independent
Location: Zurich, Switzerland
Technique: Monte Carlo Regret Minimization with Imperfect Recall
Bender

Team Leaders: Johannes Fürnkranz, Frederik Janssen, Sang-Hyeun Park, Eneldo Loza Mencía, Jan-Nikolas Sulzmann, Lorenz Weizsäcker
Team Members: Johannes Fürnkranz, Frederik Janssen, Sang-Hyeun Park, Eneldo Loza Mencía, Jan-Nikolas Sulzmann, Lorenz Weizsäcker, Christian Brinker, Jan Bücher, Johannes Dorn, André Hoffmann, Alexander Juling, Benjamin Kahl, Tobias Krönke, Patrick Metzler, Mateusz Parzonka, Lavong Soysavanh, Tobias Wieschnowsky, Erkan Yüksel, Michael Zohner, Ruimin Zou
Affiliation: Technical University Darmstadt
Location: Darmstadt, Germany
Technique: Simulation-based Approach with Online Decision Tree Opponent Modelling
dcu3pl

Team Leader: Neill Sweeney
Team Members: Neill Sweeney, Dr. David Sinclair
Affiliation: Dublin City University
Location: Dublin, Ireland
Hyperborean

Team Leader: Michael Bowling
Team Members: Nolan Bard, Michael Bowling, Neil Burch, Josh Davidson, Richard Gibson, John Hawkin, Rob Holte, Michael Johanson, Boyan Marinov, Dustin Morrill, Jonathan Schaeffer, Nathan Sturtevant, Duane Szafron, Martha White
Affiliation: University of Alberta
Location: Edmonton, Alberta, Canada
Technique: The three player games use CFR within abstract games to generate stratagies (Abou Risk, 2009). Heads up experts are added for common lines of play with only two players. These experts are also generated using CFR, within a game with more card knowledge than the base three player game. In the bankroll entry, the experts use an alternate payout structure to encourage more aggressive play.
Related Papers:
Nicholas Abou Risk. Using Counterfactual Regret Minimization to Create a Competitive Multiplayer Poker Agent. M. Sc. Thesis, 2009.
Martin Zinkevich, Michael Johanson, Michael Bowling, Carmelo Piccione. Regret Minimization in Games with Incomplete Information. In Advances in Neural Information Processing Systems 20 (NIPS), 2007.
Michael Johanson. Robust Strategies and Counter-Strategies: Building a Champion Level Computer Poker Player. M.Sc. Thesis, 2007.
LittleRock

Team Leader: Rod Byrnes
Team Members: Rod Byrnes
Affiliation: Independent
Location: Lismore, New South Whales, Australia
Technique: Little Rock is based on the regret minimization technique but with some important differences. It has been designed to be able to be used for games with any number of players, and as such the perfect recall of player actions that is used in the published technique is not suitable, as the number of states for games with more than 3 players becomes too large to compute. Instead Little Rock uses two abstractions. The first is a bucket abstraction, whereby each hand is placed into one of 31 buckets based on its potential (similar to the EHS squared metric). The second is a game state abstraction that takes into account number of players, player position, various actions that have taken place, as well as information about the pot size, bet/call ratio etc. By using these two abstractions much larger games are able to be solved, with the drawback that the agent may not be as effective against algorithms specifically tailored to games with fewer players (ie. 2 and 3 player games). Tests with Poker Academy show that Little Rock is able to convincingly beat Sparbot heads-up, and Pokibot in all game sizes from 3 to 9 players.