[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

BOT COMPARISON 400 / 500 GAMES

Posted By: Michael Depreli
Date: Sunday, 28 March 2010, at 9:51 a.m.

In Response To: BOT COMPARISON 400 / 500 GAMES (Daniel Murphy)

Daniel:

Michael, I was just reading up on http://www.extremegammon.com/SearchInterval.asp . So I'm wondering (1) what search intervals the 10 bot-instances compared in your table were using for their evaluations, and (2) whether variant search interval settings for any one of the bots has (or would have, since I suspect you did have the time to devote time to that) a significant effect on its total error.

The short answer is that the search intervals are infinite.
For XG the SI are the default Huge Settings.
For S4 3-Ply Precise
For BGB Frank sent me analysis with the top 7 moves
For GNUBG 2-ply is World Class, 3-Ply is 10 moves within 0.12 and 4-ply 8 within 0.08.

However providing I don't overlook them, I try to make sure all bots evaluate all the other bots' top choices. So for example if say XGR+ second best move doesn't make it through gnubg's 4-ply filter, I force it to evaluate it.

Also let's say for the top 6 moves the bots only like a,b,c,e and f based on gnubg 2-ply evals (the bot/ply I use for rollouts), play d still gets rolled out too. Re your second question, I haven't done any testing that compares SI to error rate. I spoke to Xavier about "optimizing" the SI. By that I mean testing multiple combinations e.g 7/0.18 versus 8/0.16 versus 9/0.14 etc and comparing accuaracy and speed. He said he may look at that after he has improved the net for XG.

Michael

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.