Author: Uri Blass
Date: 16:07:01 10/26/00
Go up one level in this thread
On October 26, 2000 at 18:25:33, Eelco de Groot wrote: >On October 26, 2000 at 16:53:58, Uri Blass wrote: > >>On October 26, 2000 at 15:33:17, Marc van Hal wrote: >> >>>On faster sytems the depth of plys, looks ahead with much chess knowledge is >>>about the same, as with the normal chess knowledge settings. >>>so putting it higher should improve all chess programs on faster systems. >>> I only have an 800 amd, so you can imagine what hapens on fater systems. >>>Or am I wrong here? >>>only a try to help >>> >>>Marc van Hal >> >>I do not understand. >> >>In most chess programs there is no parameter that is called chess knowledge. >>I know only about one program with this parameter(Rebel) >> >>Increasing this parameter does not help Rebel to get better results and the time >>control is not important. >> >>Uri > >I thought too that maybe Marc was referring to Rebel's "Chess Knowledge". >Increasing this parameter means that rebel needs more time to evaluate every >position. So the time needed to complete a ply should in theory go up a lot, >unless there would be very unexpected side-effects that somehow Rebel can find >refuting lines quicker now and the search would become more efficient that way >or something like that. I never really tried it but I know other people did. But >in theory time to complete a ply should increase and for tournament games this >is not compensated enough by the better understanding because of extra >knowledge. But especially for analysis the exact time needed is not so important >so there it is worth a try I think. I disagree. The exact time is important also for analysis. Another point is that the problem with chess knowledge=500 is not only smaller number of nodes per second but also the fact that Rebel changes its mind more often and the result of this is that Rebel need more nodes and not only more time to finish the same depth. to But we do not know much from Ed in which >kind of positions this extra 'Chess Knowledge' can be used. I think that this is not extra knowledge but less lazy evaluation. Rebel without big chess knowledge simply does not use the evaluation function in part of the cases(if the score of lazy evaluation tells Rebel that probably the full evaluation will not improve the real score significantly). The result is that Rebel is not sensitive to small changes in the evaluation so it can search deeper. 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.