Author: Uri Blass
Date: 05:25:38 08/03/03
Go up one level in this thread
On August 03, 2003 at 07:50:23, Jonas Bylund wrote: >On August 03, 2003 at 07:22:47, Thorsten Czub wrote: > >>On August 03, 2003 at 06:37:45, Jonas Bylund wrote: >>>The point of Emerson Tan which i find interesting, is that for very long >>>analysis it would be nice to have a version of program X (besides the default >>>.exe, .eng etc) that has as much knowledge as could possibly be squeezed into >>>program X (which in this case could be Rebel, but any srtong engine would do >>>really). Now this would hurt it's performance in fast games, but what about say >>>3 days per move or more? >>> >>>Jonas >> >>the knowledge in programs is often wrong implemented and programmers >>don't know that the algorithm is not working as it should. >> >>often different parts of knowledge wipe out each other, so that putting >>3 very good ideas altogether makes the program losing. >> >>i doubt that it would make much sense to switch anything on you have. >>the right tuning is the important thing IMO. >> >>if you switch anything ON, the program would IMO play weaker as it is in >>default. > >You have to remember that i am talking about a completely different timecontrol >than what most testing is based on (standard, rapid and blitz). I am talking >about x days per move and that is not (to my knowledge) what most engines are >tuned for. Therefore there is not much we can do to guess the outcome of adding >more knowledge at LONG time controls, it has to be put to the test, in other >words: practical play will either support adding more knowledge at x days per >move or the opposite. > >Jonas I think that it is a sure way to waste a lot of time without significant results. I suggest that you first prove that the optimal knowledge parameter is higher in 30 minutes per game than in 1 minute per game. If it is not higher then I see no reason to assume that it suddenly starts to be higher when you use more time. 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.