Author: Mark Young
Date: 03:34:32 10/03/98
Go up one level in this thread
On October 03, 1998 at 05:51:33, blass uri wrote: > >On October 03, 1998 at 05:04:07, Mark Young wrote: > >>On October 03, 1998 at 04:09:23, blass uri wrote: >> >>>In the ssdf list a program are playing long matches and I do not like it because >>>it gives an advantage to programs that are strong in learning a specific >>>opponent >>> >>>I suspect that fritz5 is a program that is strong in learning a specific >>>opponent >>> >>>I am not against learning but I think that it is better not to play against the >>>same oponent again and again. >>> >>>It is better for example that fritz5 will play the first game against Genius5, >>>the second game against Mchesspro7, the third game against Nimzo98... >>> >>>If it has not enough opponents and have to play against the same opponent many >>>times then at least it should play against other opponents before it plays >>>a game against the same opponent. >>> >>>Uri >> >>What the hell. Lets just outlaw all the learning features in chess programs when >>testing, that way SSDF does not have to go though contortions, as you suggest, >>when doing their testing. Why should we give any rating credit to programs that >>try to adapt their opening play to an opponent as a human would? > >The problem is that the rating of the ssdf list are not relevant for tournaments >when you cannot play against the opponent many games before the tournament > >for example in the world micro computer championship there are many new programs >that you cannot play against them many games before the tournament > >If the opponent is a human then you cannot play against him or her many games >before the game. > >I am not against learning if the learning help practically in tournaments but >this kind of learning from many games against a specific opponent does not help >practically. > >learning from database against other opponents can help practically >but the point is that fritz5 learned from its games against the same opponent in >the SSDF list games and not from database of the opponents. > >I am not sure that Fritz5 is number 1 because of this but it is a possibility >and if it it the case then the first place of Fritz5 is not relevant for >tournaments with many unknown opponents > >Uri > >Uri So, what if the rating is not relevant in tournaments. The SSDF rating is for the consumer to compare programs in match play. How many people buy their chess programs to play in tournaments? I want a chess program to play me, and I would want the program to try to adapt to me. Therefore, I think it is very relevant for SSDF to express learning features in their ratings. I would suggest that the program(s)that should be #1 on the SSDF rating list get with it and make their programs better learners. That way we do not have to dumb down the SSDF rating list for them to compete. I think that is a much better solution then the one you suggest. Its better for the consumer, and it is better for them.
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.