Poker-AI.org

Poker AI and Botting Discussion Forum
It is currently Mon Nov 13, 2023 5:47 pm

All times are UTC




Post new topic Reply to topic  [ 5 posts ] 
Author Message
PostPosted: Thu Aug 29, 2013 12:25 am 
Offline
Junior Member

Joined: Mon Aug 12, 2013 7:28 pm
Posts: 16
Hi guys,

I'm trying to write my first bucketizer.
I'm starting with something I think is simple, 10 buckets preflop, 10 flop, 10 turn and 10 river.
I'm using EHS^2 as described in Johanson's paper: http://poker.cs.ualberta.ca/publication ... on.msc.pdf
I'm planning to use the buckets in a CFR algorithm.

I have 2 questions:

1) It seems impossible to enumerate all the cards to exhaustively compute the transition probabilities. It would mean enumerating all my pocket hands (1326) and then enumerate all the flop, turn and river cards. And for each hand(private+board) I want to bucketize, I also need to compute its EHS^2 which I do using a simple sampling (MonteCarlo) approach.
Of course, I can change the number of samples in the MonteCarlo approach but even using a low value of 100, it seems to me that enumerating all these cards is not feasible. So I tried enumerating all my pocket cards, and then sample the rest of the board, let's say 1000 board samples per preflop hands. Is that a reasonable/correct way to do it ?
Also what's a faster/more accurate way to compute the EHS^2 ? Is there a good post about it maybe I missed it/which keywords should I look for ?

2) He says we need to know the probability of every bucket sequence winning a showdown against every opponent bucket sequence. That makes sense as I can see how this would fit in computing the utility for the CFR.
But this means that with my 10,10,10,10 buckets, I get 10x10x10x10=10000 different bucket sequences, so about 100million bucket sequence combinations. It's computable with my low bucket number, but it's very heavy already, am I missing something here ?
Or maybe we don't need to do this if buckets are strictly ordered and we just declare the same buckets on the river as a tie instead of considering the sequence leading there ? We'd loose some information there though ? What do you think ?

Thanks!


Top
 Profile  
 
PostPosted: Thu Aug 29, 2013 1:05 am 
Offline
Junior Member

Joined: Mon Apr 22, 2013 11:46 am
Posts: 34
you probably shouldn't deal with bucket sequences at all. it's a bit harder to implement and doesn't perform as well as imperfect recall.


Top
 Profile  
 
PostPosted: Thu Aug 29, 2013 8:14 am 
Offline
Junior Member

Joined: Mon Aug 12, 2013 7:28 pm
Posts: 16
Imperfect recall yes thanks, that makes sense, especially with all the posts i've read saying it holds better results in practice.

What about computing the buckets transition probabilities ? I suppose sampling is good enough ?
I suppose i'll just check for the probabilities converging actually.


Top
 Profile  
 
PostPosted: Sat Aug 31, 2013 11:55 am 
Offline
Veteran Member

Joined: Wed Mar 20, 2013 1:43 am
Posts: 267
Do you just forget everything on a new street? I thought it was state of the art to remember some informations from previous streets.


Top
 Profile  
 
PostPosted: Sat Aug 31, 2013 12:47 pm 
Offline
Junior Member

Joined: Mon Aug 12, 2013 7:28 pm
Posts: 16
It's my first attempt. First I'm going to completely forget about the previous streets. Then my idea is to synthesize the previous streets information into a couple of booleans or numbers. Ie: has someone raised preflop, did someone cbet...
I have no clue if this is a good idea, but it sounds good to me


Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 5 posts ] 

All times are UTC


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:
Powered by phpBB® Forum Software © phpBB Group