|
From: | Joseph Heled |
Subject: | [Bug-gnubg] Re: Naw..panic or teach me what I am doing wrong !! |
Date: | Wed, 08 Jan 2003 18:32:52 +1300 |
User-agent: | Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.2) Gecko/20021202 |
There seems to be a discrepancy. My code gives me consistently 0.54213905334472656, 0.34824475646018982, 0.08864891529083252, 0.0, 0.0 When I use gnubg, I get Neil numbers or worse. Have no idea yet what is wrong.Can gnubg log the rollouts? I would like to analyze them with fibs2html to see where the differences are?
-Joseph Neil Kaz wrote:
0-ply 1296 games AWFUL -.8503 +/- .0327. Good night and I try again tomorrow :( ----- Original Message ----- From: "Joseph Heled" <address@hidden> To: "Neil Kaz" <address@hidden> Cc: "gnubg" <address@hidden> Sent: Tuesday, January 07, 2003 10:26 PM Subject: Re: No more need to panic about backgame play anymoreNeil Kaz wrote:Joseph wrote :2 ply filters are default. set evaluation movefilter 2 0 2 3 0.1 set evaluation movefilter 2 1 0 0 0Neil replies: I didn't know how to change the evaluation move filters.So Ijust ran it 2-ply default with the new version 13. As I type this (800 games), GNU is winning from the backgame side almost as much as Snowie 43ply did. And the cubeless equity for the backgame is -.576 (note SW4 was -.554) This is about what I'd expect since I do feel SW4 plays it just a little bit better, noting that this really isn't enough games to be statistically certain.Perhaps, but this is the figure I got for the 0 ply. I feel the 2ply should do better. My (very small) 96 games indicated a -0.50 figure, but it might have been lucky. Will try to generate a larger sample later. Neil, what does your GNU 0ply rollout says. It should take only a few minutes for a 12960 sample? -JosephAlso note that default settings don't have filters as large as SW 4.MikeDepreli told me the SW 4 filters and I'll match them exactly and run 50% speed tomorrow. (Should match SW 4 3-ply precise) Now that I know that GNU can now handle backgames, I'll take a look at several holding game reference positions I have where SW 4 is clearly overvaluing the defender's chances. SW 4 can be more than .100 cubelessinerror in some holding games positions where the defender has a blotbehindthe advanced anchor. Let's hope GNU isn't making the same mistakes. Please forward this to the rest of the GNU team and thx for your help. ..neilkaz..
[Prev in Thread] | Current Thread | [Next in Thread] |