Poker Research

ok 2+2ers... ive been a longtime reader but i rarely post.... i leave it up to the pros.

well im a masters student in computer science and i have to pick a thesis... well i thougth why not incorporate the game i love. so im fishing for ideas on wut to do research on... but please keep in mind that it has to be a legit masters thesis or else i dont graduate, so please add some intellectual/Artificial intelligence/computer science related theme to your ideas.


hope u guys can help me out... and i can spend the next year doin some cool project where i have a poker bot that learns or soemthing along those lines.


peter

Someone in canada is already doing the learning poker bot thing... so I'd skip that one.

http://www.cs.ualberta.ca/~games/poker/

I am also a master's CS student, but luckily my program doesn't require a final project/thesis of any type.


Anyhow, this is something I started to do, but gave up on.

It may be good for a thesis since it could incorporate AI, compilers, and even some statistics.


Create a rule based HE simulator with the following properties.


Each player would be his/her own object with a set of logical rules. , i.e. If (I have top pair best kicker) and (Bets to me =JQ) or (players hole cards >=98s) then (increment respect upto 80)

My idea for this was that increment would ++ a respect var, but upto would put a limit on the max. An alternative to this was two ratings, tightness and agressiveness. These ratings could then be used in the players decision. For instance


define group 2 AJs ATs JJ TT 99


if (hand = group2) and (respect(raiser(1))

J, did this get cut off, or was that your complete post?

yah ive read about the research in alberta... thanks for your responses, and keep them coming.

i also was pondering some algorithms for the learning HE bot including a tight/loose scale and a passive/aggr scale on determing which hands to play against certain players.


peter

Oh..The third part was to be able to feed the program data from an actual game and have it figure out players equations. I think that would be really cool, useful, and worthy of a Thesis(if it was fairly successful at prediction.)


-J

How about trying neural nets.


It's worked very well for backgammnon

and I don't think anyone's done it yet

for Hold'em.

Hello Peter:


I don't think the last word on EV has been written.


For example: To take the simplest case, AA heads up has an EV of 1.70 (wins c 85% against random hand - it does not vary much when the hand randomization is limited to group 8 or better). But as the board grows the EV of succeeding bets change, sometimes drastically depending on the board's content (as well as the opponents' play). Some of this could lend itself to simulation.


The above EV is academic since it is generally at a high positive value and does not often enter into decision making (although it could shed some light on the use of slowplaying AA).


But there are a variety of situations that are highly practical such as the response to preflop raises depending on position; the analysis of actions from the steal position; the utility of various hands depending on the type of opponents and their positions.


Note that serious players have clear ideas how to handle these events but most of the analysis stops before the flop comes. What post flop analysis exists, it is not well quantified for many examples particularly in multiway instances.


Good luck,


Charlie K