Author: Bas Hamstra
Date: 08:52:03 06/19/01
Go up one level in this thread
On June 19, 2001 at 08:44:19, Gian-Carlo Pascutto wrote: >On June 19, 2001 at 08:21:38, Bas Hamstra wrote: > >>>For sorts of futility pruning, this is my general observation which is >>>confirmed by Vincent. >> >>No wait, this is a different matter. Vincent just doesn't like FP. And I don't >>either. For Crafty-like qsearches it works ok. But I don't like such a qsearch. > >What evidence do you base this on? Base what on? a) That Vincent doesn't like FP b) That I don't like it c) That it works for Crafty d) That I don't like Crafty's qsearch But I think you want me to explain why I don't like FP? For one thing I do not evaluate, when InCheck. I always GenOutOfCheck. This does not go well together with FP, unless you check if a certain capture is "checking" first. But then the benefits become so marginal, it is hardly worth doing it. Plus it makes errors, of course. Think of an endgame, where a pawncapture makes the enemy king stand outside the square of the pawn. In one move you can gain +8.00 or so. FP will skip that capture happily. FP gives you speed and depth. I chose to sacrifice some, to get better quiescence. >My program used a crafty-like qsearch and when I started adding more eval >I had to up the FP margins too to get more benefit. Maybe you should check if you get even more benefit, when switching it off. When using big windows the reduction in nodes is also much smaller. I think FP is overrated. Big windows, small gains. Small windows, bad play. >The FP in crafty dates back to 1996(!) and has not been altered or tested >since. I believe Robert is reevaluating it now. Also Extended FP/Limited >razoring has been found not to work. I know. There are many ideas that sound brilliant, but when you test them they disappoint. Such as: at depth=1 why would you search all moves, if eval is already >= Beta? Sounds fine. Extended FP sounds fine too. Razoring sounds fine (I think in an old paper razoring had a diffent meaning than today?!). But I have become a bit sceptical. >>Whether the nullmove version would have been weaker tactically, I am not so >sure about that. Personally I think not. Take you own program, give it a fixed >>time per position. Then compare with and without nullmove. Which one sees more >>tactically? For my program it is absolutely no question. > >At four hours per move (needed to get comparative to DB), I think it is >going to make little difference, especially when SE comes to play. How can you tell? Best regards, Bas.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.