Author: Uri Blass
Date: 03:00:49 04/12/02
Go up one level in this thread
On April 12, 2002 at 05:40:09, Sune Fischer wrote: >On April 12, 2002 at 05:10:30, Uri Blass wrote: > >>On April 12, 2002 at 05:05:23, Sune Fischer wrote: >> >>>On April 12, 2002 at 04:23:04, Uri Blass wrote: >>> >>> >>>>>Fritz 7 is a good program, without a doubt, one of the best for PCs. HOWEVER, >>>>>get real. 4M "wiser" nps isn't CLOSE to 200M "wise" nps. No matter how you >>>>>look at it. >>>> >>>>There is an assumption that the 200M are "wise" nps. >>>> >>>>I believe that if Crafty has no bugs then Crafty17.07 at 200M nps is slightly >>>>better than the Fritz that kramnik is going to play but only slightly better >>>>than it. >>> >>>200 versus 4 is a big factor corresponding to probably 3-4 plies, the only way >>>fritz can make up for that would be to search a much slimmer tree or use better >>>extensions I think, but maybe that is not unrealistic with nullmove and better >>>moveordering like SEE and hashmoves etc? >>> >>>-S. >> >>Another way is also to use better evaluation function. > >In what way? >I think DB had enough knowledge to prefer getting 2 pawns for a bishop over >losing 2 pawns (if that was your example). In that case it's search rules were not good enough to see that Kh1 is leading to winning 2 pawns or they were not good enough to see that Kf1 is leading to a bishop for 2 pawns. >AFAIK DB had a very good eval since they could throw anything in there for free, >so I wouldn't count on Fritz to have a huge advantage there, some maybe but not >huge. You may be right and I did not say that the evaluation advantage of Fritz is huge. > >Anyway there is probably also a diminishing return on the evaluation. >Some stuff is really important to have like material, mobility, double pawns, >but already you can see that material is more important than mobility and >mobility is more important than double pawns (IMO). >This list goes on I think, when you have the 20 most significant evalution >terms, then the rest will just be special knowledge that you can't even apply to >every position, most of it probably endgame knowledge. Endgame knowledge is the more easy knowledge to define. middle game knowledge may be also important but it is more hard to define it correctly. > >But I do think Fritz has a lower branch factor and better extensions, >seeming the tree grows exponentially this could amount to quite a bit of savings >compared to DB. 3-4 plies is not out of reach with those techniques. I agree. 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.