Image Image Image




Post new topic Reply to topic  [ 7 posts ] 
Author Message
 Post subject: Game theory-based opp. modeling in large imperfect-info. games
PostPosted: Sat Feb 11, 2012 2:43 pm 
Offline
Level1 member
User avatar

Posts: 46
Favourite Bot: ---
Game theory-based opponent modeling in large imperfect-information games

Authors: Sam Ganzfried, Tuomas Sandholm
Published: AAMAS '11 The 10th International Conference on Autonomous Agents and Multiagent Systems - Volume 2
Date: May 2011

Abstract:
We develop an algorithm for opponent modeling in large extensive-form games of imperfect information. It works by observing the opponent's action frequencies and building an opponent model by combining information from a precomputed equilibrium strategy with the observations. It then computes and plays a best response to this opponent model; the opponent model and best response are both updated continually in real time. The approach combines game-theoretic reasoning and pure opponent modeling, yielding a hybrid that can effectively exploit opponents after only a small number of interactions. Unlike prior opponent modeling approaches, ours is fundamentally game theoretic and takes advantage of recent algorithms for automated abstraction and equilibrium computation rather than relying on domain-specific prior distributions, historical data, or a handcrafted set of features. Experiments show that our algorithm leads to significantly higher win rates (than an approximate-equilibrium strategy) against several opponents in limit Texas Hold'em --- the most studied imperfect-information game in computer science --- including competitors from recent AAAI computer poker competitions.

http://www.cs.cmu.edu/~sganzfri/AAMAS2011_Opponent.pdf


Top
 Profile E-mail  
 
 Post subject: Re: Game theory-based opp. modeling in large imperfect-info. games
PostPosted: Thu Feb 23, 2012 11:54 pm 
Offline
Senior member
User avatar

Posts: 134
Favourite Bot: my (incomplete) bot
I'm a little puzzled about the part where the author discusses the issue of an opponent mucking their hand. Specifically how to narrow the range hands that opponent could have had. Don't most sites write this information to the hand history document? Which made me wonder... if mucked cards are NOT in the GUI but ARE in the hh file has anyone modeled that feature in their simulator? Do you think it's worth incorporating for training the AI?

_________________
TorChat ID: "irqnpurfeuebdwmd"


Top
 Profile E-mail  
 
 Post subject: Re: Game theory-based opp. modeling in large imperfect-info. games
PostPosted: Fri Feb 24, 2012 6:06 pm 
Offline
Junior member
User avatar

Posts: 39
Favourite Bot: nonameasofrightnow
Quote:
if mucked cards are NOT in the GUI but ARE in the hh file has anyone modeled that feature in their simulator? Do you think it's worth incorporating for training the AI?


mucked cards are not in hh file.
if you call a bet on the river (or if you are last to check) and your hand is worst than your opponent your cards will not be shown on the table, but they will be visible on hh. (this are not mucked cards)

when you win a pot without showdown you can just muck your cards and no one will ever know which exact cards it was.


Top
 Profile E-mail  
 
 Post subject: Re: Game theory-based opp. modeling in large imperfect-info. games
PostPosted: Sat Feb 25, 2012 9:17 am 
Offline
Senior member
User avatar

Posts: 134
Favourite Bot: my (incomplete) bot
I accidentally posted my original question in the wrong papers forum, meant to ask it for: http://pokerai.org/pf3/viewtopic.php?f=64&t=2664&sid=914e6ffe5b2d879fdaf020f63968d6ca anyway on page 4 the author writes
Quote:
This approach has the advantage that mucked hands still hold some information and can be attributed to bins with ranks lower than that of the current winning hand.

So i'm assuming the author is considering the losers hand at showdown to be classified as "mucked". Also I just double checked some hand history files and at showdown it labels two losing players as mucking their hands but then their hole cards are listed in the hh file. In any case when you said:
veryfastT wrote:
mucked cards are not in hh file. if you call a bet on the river (or if you are last to check) and your hand is worst than your opponent your cards will not be shown on the table, but they will be visible on hh. (this are not mucked cards)
Whatever that scenario is called, that's what I wanted to know if anyone has modeled in their simulator

_________________
TorChat ID: "irqnpurfeuebdwmd"


Top
 Profile E-mail  
 
 Post subject: Re: Game theory-based opp. modeling in large imperfect-info. games
PostPosted: Tue May 08, 2012 9:45 am 
Offline
New member
User avatar

Posts: 4
Favourite Bot: mine
I haven't read the paper yet but I would presume that the issue is where the data was obtained.

Observed hands mucked at showdown should not appear on the hand history.

Likely the hand histories were 'data mined' and and so the information was not available.


Top
 Profile E-mail  
 
 Post subject: Re: Game theory-based opp. modeling in large imperfect-info. games
PostPosted: Thu May 10, 2012 2:58 pm 
Offline
Senior member
User avatar

Posts: 134
Favourite Bot: my (incomplete) bot
qwertyuiop wrote:
Likely the hand histories were 'data mined' and and so the information was not available.
That makes sense.

_________________
TorChat ID: "irqnpurfeuebdwmd"


Top
 Profile E-mail  
 
 Post subject: Re: Game theory-based opp. modeling in large imperfect-info. games
PostPosted: Mon Jul 09, 2012 7:34 pm 
Offline
New member
User avatar

Posts: 9
Favourite Bot: none
Hi,
This paper does not actually use any historical data. We basically start with a prior that the opponent is playing a precomputed approximate-equilibrium strategy, and then exploit observed deviations from this strategy (by computing an approximate best response in real time). For example, if Nash opens 70% of hands from the button preflop, and our opponent only opens 40%, then we will construct an opponent model by basically taking out the "worst" 30% of hands from the Nash range. Our current implementation does not assume that we ever see the opponent's hand at showdown -- this is an obvious avenue for improvement in the future.


Top
 Profile E-mail  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 7 posts ] 


Who is online

Users browsing this forum: No registered users and 3 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to: