Author: Uri Blass
Date: 09:04:40 12/04/03
Go up one level in this thread
On December 04, 2003 at 11:29:52, Robert Hyatt wrote: >On December 04, 2003 at 10:21:09, Uri Blass wrote: > >>On December 04, 2003 at 09:38:36, Robert Hyatt wrote: >> >>>On December 03, 2003 at 19:14:36, James T. Walker wrote: >>> >>>>I also have a question about Crafty 19.6. Why is it such a poor Blitz player? >>>>It seems strong at long time controls but still loses to Fritz 5.32 at Blitz. >>>>Is there something in the search that only works at deeper depths than usually >>>>reached at blitz? On my XP2400+ Crafty usually hits 9 ply minimum and 10-12 is >>>>normal most of the time (G/5min). >>>>Jim >>> >>> >>>One thing is the q-search. Shallower searches give null-move a better >>>opportunity to hide a key tactic. At longer depths, this gets pushed >>>farther out into the tree, where it goes beyond what is "forced" and into >>>what is "optional" in the real game, most of the time. >> >>I remember that when you compared crafty with null move and crafty without null >>move you found results that suggested that the main benefit of null move is at >>blitz and not at long time control. > >No, that is backward. What I found was that at blitz, Null-move did worse >while at longer time controls it did better. But as depth increased due to >faster hardware, the depth at blitz (today) is as good as the depth at >long games 8 years ago.. > >The problem is overlooking fairly shallow tactics by letting the null-move >hide a non-capturing threat that the q-search can't find by itself... > >> >>This means that null move is not a reason for relatively good results at long >>time control. >> >>Note that I do not think that Crafty is stronger at long time control relatively >>to other programs and the problem is probably with Fritz5.32 that is weaker at >>long time control. > >I believe it is. For example look at my ICC standard rating, made almost >exclusively against computers. Most programs have a blitz rating 200-400 points >above their standard rating. Mine isn't. I do not trust the rating list. Crafty of today has 2803 in blitz Crafty of 9.6.2000 had 3388 in blitz Do you believe that Crafty of 2000 on the hardware of 2000 was almost 600 elo better in blitz than Crafty of today on the hardware of today? I suspect that difference in rating on ICC is too much influenced by the question how much games you play and if you try to inflate your rating. It seems that there was a big deflation since 2000 and I guess that programs who try to inflate their rating simply do not play blitz because the operators expect to lose rating. Uri
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.