Author: Uri Blass
Date: 17:19:45 12/29/01
Go up one level in this thread
On December 29, 2001 at 19:50:13, Sune Fischer wrote: >On December 29, 2001 at 19:13:33, Christophe Theron wrote: > >>On December 29, 2001 at 18:48:32, Severi Salminen wrote: >> >>>>>>If your full QS eats 80% of the time I would say that limiting QS is indeed a >>>>>>must. However my QS (full) takes about 10-15% (25-30% with checks) and so the >>>>>>picture is quite different then. >>>>> >>>>>Ok, but what about nodes? Does your qsearch take only 10-15% of all the nodes? >>>>>That's quite low... >>>>> >>>>>Severi >>>> >>>>Yes that is very impressive, I must have lots of room for improvement ;) >>> >>>Yeah, but no way it can be 10% of all the nodes: that is just too low. Or then >>>it prunes like hell... >>> >>>Severi >> >> >> >>Believe it or not, above 30% you have a big problem somewhere. > >I think R. Hyatt said crafty spends 50% of the time in QS. I think that it includes the time that Crafty spends on evaluation function. > >>That's the reason why using a SEE instead of the QSearch is not an improvement. >> >>Remember: we do not count the horizon nodes as being in the QSearch, because >>with or without QSearch you have to visit them anyway. >> >> Christophe > >I'm not counting nodes just the time, the program just runs 3-5 times slower to >the same depth (for the base tree) when it does a full QS. It means that you probably have some bug. The program should run faster to the same depth when you do qsearch because of better order of moves. I don't understand >why this is so unexpected, say there are two possible captures at every leaf >node on average after I stop my QS at depth 3. Then that alone is 3 times as >large a tree, and there could be more captures after that! There are captures that should not be analyzed because they are not good enough. If you are a rook down and capture a pawn when alpha=0 then you can avoid making the move because you can know that it is not good enough. I can guess that your problem may be that you make these captures only to evaluate them later and discover that they are bad. Maybe my guess is wrong and your problem is different. >In the midgame I think it is quite normal for the QS to do 5-6 plies. It is not rare but it is also not something that happen often enough to cause the problem that you talk about. > >BTW Christophe, are you not doing SEE at all? I think that Tom already explained in his reply to christophe that Christophe does DEE and not SEE. Uri
This page took 0.01 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.