Poker-AI.org
http://poker-ai.org/phpbb/

Slumbot NL: Solving Large Games with Counterfactual Regret..
http://poker-ai.org/phpbb/viewtopic.php?f=25&t=2531
Page 1 of 1

Author:  Isildur11 [ Tue Jul 16, 2013 7:27 pm ]
Post subject:  Slumbot NL: Solving Large Games with Counterfactual Regret..

Abstract

Slumbot NL is a heads-up no-limit hold'em poker bot built with a distributed disk-based implementation of counterfactual regret minimization (CFR). Our implementation enables us to solve a large abstraction on commodity hardware in a cost-effective fashion. A variant of the Public Chance Sampling (PCS) version of CFR is employed which works particularly well with our architecture.

http://www.aaai.org/ocs/index.php/WS/AA ... /view/7044

Author:  cantina [ Tue Jul 16, 2013 7:52 pm ]
Post subject:  Re: Slumbot NL: Solving Large Games with Counterfactual Regr

Unless I missed it, they don't make note of the size on disk. I'm guessing it was roughly that of their HUL bot last year: 500gb.

Author:  thewannabe [ Thu Sep 12, 2013 7:01 pm ]
Post subject:  Re: Slumbot NL: Solving Large Games with Counterfactual Regr

Nasher wrote:
Unless I missed it, they don't make note of the size on disk. I'm guessing it was roughly that of their HUL bot last year: 500gb.


I asked Mr. Jackson a couple questions about this, this was his response:

Eric Jackson wrote:
I used nine machines, but most of them were pretty low-end. All running Linux. One machine had 8 gigs of RAM, but most just had 2 gigs. Didn't use multithreading, but on the more powerful machine had two processes running at once. Trained for about four months. Combined size of the data was probably around 150 GB.

Page 1 of 1 All times are UTC
Powered by phpBB® Forum Software © phpBB Group
http://www.phpbb.com/