Author: Uri Blass
Date: 10:42:22 08/13/05
Go up one level in this thread
On August 13, 2005 at 13:04:02, Joachim Rang wrote: >On August 13, 2005 at 12:54:08, Thomas Logan wrote: > >>On August 13, 2005 at 11:04:06, Joachim Rang wrote: >> >>>On August 13, 2005 at 10:47:18, Thomas Logan wrote: >>>> >>>>I think computer chess will experience a leap foward when the evaluation >>>>techniques of Fruit are incorporated into other engines >>>> >>>>Tom >>> >>>why people think that the evaluation of Fruit is so superior (Uri also believes >>>it). I think the evaluation is nothing fancy but not speculative (contrary to >>>many commercials which have speculative eval in order to let them find "human" >>>sacrifices). >>> >>>regards Joachim >> >>I don't know but something has to accout for Fruits playing strength and >>comments concerning search bugs in Fruit leads one to discount the idea it's >>strength resides in it's search >> >>Tom > > >Ah well search bugs ;-). Fabien himself mentioned this but when it comes to code >quality of his own engine you should believe Fabien as much as Tord (both are >very modest of their own engines). There might be some earch bugs but most >probably there are evaluation bugs too (no program is without bugs). These bugs >however seem not to affect playing strength substantially. I think Fruits eval >is as good as his search - both are very clean and well-designed but not >outstanding. Outstanding is Fruits playing strength without outstanding >features. ;-) > >regards Joachim Movei also has search bugs and I do not claim that fruit's search is not superior to Movei but I think that most of the advantage of fruit is thanks to better evaluation. I did not work much on Movei lately but I think that inspite of not having things like Internal iterative deepening and pruning based on hash tables it is better to work first on evaluation and not on search and there is a lot to improve in Movei(for example pawn shield evaluation is too small,backward pawns and candidate passed pawns are not evaluated,king safety can be improved thanks to fruit's code) I guess that Movei can close most of the gap to fruit2.1 if I implement the relevant knowledge in Movei My opinion is that it is better to implement evaluation first. It does not mean that search is not important but only that my guess is that bigger gain can be earned from evaluation and also that better evaluation open opportunities to earn more from better search because part of my search tricks are based on evaluation and this is the reason that Movei evaluate every node and not only the leaves. 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.