has anyone ever tried implementing sibling prediction pruning from Jeroen Carolus thesis?
http://chessprogramming.wikispaces.com/Jeroen+Carolus
http://www.google.it/url?sa=t&rct=j&q=& ... 0087,d.ZGU
My last attempt is giving me some little enanchement (+10 elo) in 6s selfplay games.
I'm just opimizing value and then I'll try on longer time control.
sibling prediction pruning
Moderators: hgm, Rebel, chrisw
-
- Posts: 4367
- Joined: Fri Mar 10, 2006 5:23 am
- Location: http://www.arasanchess.org
Re: sibling prediction pruning
Unless I am mistaken this is not a new idea but just an implementation of classic futility pruning.
--Jon
--Jon
-
- Posts: 855
- Joined: Sun May 23, 2010 1:32 pm
Re: sibling prediction pruning
I think the same, but adding it to my engine with futility pruning is giving me some improvement.
there is a little difference between futility pruning and sibling prediciotn pruning. The former don't test a move because it will not raise the eval, the latter as soon find a quiet move below a threshold, don't test ALL the quiet move related to the move.
there is a little difference between futility pruning and sibling prediciotn pruning. The former don't test a move because it will not raise the eval, the latter as soon find a quiet move below a threshold, don't test ALL the quiet move related to the move.
-
- Posts: 4367
- Joined: Fri Mar 10, 2006 5:23 am
- Location: http://www.arasanchess.org
Re: sibling prediction pruning
If your futility pruning threshold is the pre-move static score + a fixed margin, which is commonly done, then you get the same behavior, because one quiet move looks like any other quiet move to the futility code.
However, some implementations try to get a better estimate of each move's score by doing at least a partial calculation of its positional gain/loss, or using historical data that can estimate this.
--Jon
However, some implementations try to get a better estimate of each move's score by doing at least a partial calculation of its positional gain/loss, or using historical data that can estimate this.
--Jon
-
- Posts: 855
- Joined: Sun May 23, 2010 1:32 pm
Re: sibling prediction pruning
you are rightIf your futility pruning threshold is the pre-move static score + a fixed margin