Transcript bt-opponent_modeling_in_bayesian_poker3

Outline

Background




Aims





What is Texas Hold'em?
What are Bayesian Networks?
What is BPP?
Initial opponent model
Adaptive opponent model
Performance testing
Further Work
Conclusion
Opponent Modeling in Bayesian Poker
Brendon Taylor (BSE)
http://www.allposters.com/-sp/Poker-Pups-II-Posters_i1611677_.htm
Supervisors: Ann Nicholson
Kevin Korb
What is Texas Hold'em?
Poker Hands
From strongest to weakest
Poker Bayesian Network
What is BPP?





Bayesian Poker Program
1993: Initial version (Jitnah)
1999: First publication (Korb, Nicholson, Jitnah)
2000: Decision network (Carlton)
2003: Adapted to Texas Hold'em (Boulton)
Personality Types
Aggressive behaviour
 More likely to bet/raise
 Conservative behaviour
 More likely to fold/check/call

AAAI 2006 Results - Bankroll
Hyperborean
(U Alberta)
Hyperborean
(U Alberta)
Bluffbot
(Finland)
Monash
(Monash U)
Teddy
(USA)
0.0514
±0.0171
0.7227
±0.0161
0.4067
±0.0247
0.5271
±0.0197
-0.1895
±0.0289
Bluffbot
(Finland)
-0.0514
±0.0171
Monash
(Monash U)
-0.7227
±0.0161
-0.5271
±0.0197
Teddy
(USA)
-0.4067
±0.0247
0.1895
±0.0289
1.1678
±0.0427
-1.1678
± 0.0427
Initial opponent model
CONSERVATIVE
AGGRESSIVE
New Network Structure
New node
Final opponent model
Generating different opponents
using Betting Curves
Aggressive
Conservative
Adapted from Carlton (2006)
Results - Opponent Type
Further Work

BPP's game play




Improved bluffing strategy.
Adding sand bagging.
Avoiding predictable game play
Network structure



Adding a OppTight node to the network.
Adding a OppBluff node to the network.
Adding a BppBehaviour node to the network.
Conclusion



BPP is an ongoing research project and still
requires further work.
The improved opponent model has improved
BPP's ability to adapt to an opponent.
This project has been challenging and taken
me outside my comfort zone.
References





AAAI Computer Poker Competition (2006).
http://www.cs.ualberta.ca/~pokert/2006/index.html
Aces High Casino Parties and Rentals San Antonio Texas
(2007). http://www.aceshighcasinoparties.com
Carlton, J. (2000). Bayesian poker, Honours thesis, School
of Computer Science and Software Engineering, Monash
University.
Poker Pups II Prints by Jenny Newland at AllPosters.com
(2007). http://www.allposters.com/-sp/Poker-Pups-IIPosters_i1611677_.htm
Taylor, B. (2007). Opponent Modeling in Bayesian Poker,
Honours Thesis, School of Computer Science and Software
Engineering, Monash University.
Aggressive opponent model
Conservative opponent model
Lessons Learnt





Honours is more challenging than undergraduate units.
Artificial Intelligence and decision making.
Machine learning and structures.
How to effectively research a topic.
What to expect if I was to undertake further
post-graduate studies.