Author: Frank Schubert
Date: 07:27:38 11/09/99
Go up one level in this thread
On November 09, 1999 at 08:05:30, blass uri wrote: >On November 09, 1999 at 07:44:23, Frank Schubert wrote: > >>On November 09, 1999 at 06:41:16, blass uri wrote: >> >> >>> >>>The nunn match is also not the right way to decide about the level of chess >>>programs. >>> >>>The main problem is that the positions are known. >>> >>>Uri >> >> >>Hi Uri, >> >>You cannot compare the Nunn-Test with simple positional tests. >>Please show me one programmer who is able to tune his program >>for the Nunn-test WITHOUT increasing the general playing strength. If he is >>able to do that, he will produce a better chess program. That's one of >>the main ideas of the Nunn test. Another important point is that it's >>not possible to suppress games. So I think that the Nunn test is a very >>accurate indicator of playing strength within the 20 games limit. >>But of course 20 games are not much, and therefore results like 9-11 or >>even 13-7 do not really show which program is definitely the best. > >I think that it is not so hard to do it. > >You can change the piece square tables based on the opening and it will only >change the level of the program in the opening but will not increase the general >level of the program. > >I believe that you can create a different chessmaster personality for every nunn >position and improve the result of chessmaster in the nunn positions. > If you want to tune a program for a single opening position then you must know what is the right way (i.e. the BEST way) to handle this position. In all Nunn positions opening theory doesn't give a clear answer to this question and therefore tuning cannot be effective. >I did not try to do it because it was not important for me(I have nothing to >earn by doing it). > >Some programs today can learn by position(Hiarcs) and you can give them to play >nunn position games and learn from them and use the knowledge in future games. > Yes, you're right here. You have to switch the learning options off if you want to get reproducable results. Frank
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.