Author: blass uri
Date: 22:03:25 12/31/98
Go up one level in this thread
On December 31, 1998 at 18:04:50, KarinsDad wrote: >On December 31, 1998 at 17:38:06, blass uri wrote: > >>I think that the ssdf guys should give the programs time to learn(12 hours >>between every 2 games should be used for learning and not for playing) >> >>I cannot learn much without analyzing my games and I think that computer >>programs should have the right to analyze their game before they play a new >>game. >> >>programs can use the time for learning to analyze their games(for example if >>they discover that they lost not because the opening but because they missed a >>win then they can repeat the same opening again). >> >>Uri > >I disagree. Do you have 12 hours to "learn" between games at a tournament? >Usually not. Usually I have more than 12 hours between 2 tournament games In most of the tournaments I have even a week or 24 hours between the beginning of 2 games I do not use 12 hours to learn but I think that programs can use the time to learn > Even if you had the time, a person usually has to recover between >games anyway. > >Computer chess is strangely different than human chess. Each human chess machine >has only one version. With the learning capability, each version of a computer >chess program can have multiple "subversions". > >Which subversion should be playing a given game? This question is not >deterministic, hence, the original (commercial) version should always be used. programs play long matches in the ssdf games so they can decide that in the first game the original version will be used and it will learn during the match. > >The purpose of the ssdf is to approximately compare (commercial or available to >the public) chess programs playing strength, not their learning strength. It >would be unfair to have "special" opening books for only playing in the ssdf for >a given program, even if those opening books were dynamically created (i.e. >learned) between games. > >The ssdf should not play special favors for anyone. Having the program "learn" >while it is playing a game is fine if it is a feature of the program. Having the >ssdf waste CPU power to allow the learning programs to get better between games >when the non-learning programs do not have this option is unfair. It's the same >as loading an opening book into the program due to which opponent it is playing >(this would be fine if the commercial program had the capability and did it >dynamically during the game). I do not understand why learning during games is fair and learning between the games is unfair. programs can instead of resigning in a lost position and instead of playing when they see that they win to use only 1 second for playing and the rest of the time to learn and use an hour or 2 hours to learn between the games. Do you think that it is fair? 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.