Author: Stefan Meyer-Kahlen
Date: 08:51:52 12/08/00
Go up one level in this thread
On December 08, 2000 at 09:44:48, Uri Blass wrote: >On December 08, 2000 at 08:07:48, stuart taylor wrote: > >>On December 07, 2000 at 17:39:38, Stefan Meyer-Kahlen wrote: >> >>>On December 07, 2000 at 17:02:38, Chris Taylor wrote: >>> >>>>On December 07, 2000 at 14:16:04, Christophe Theron wrote: >>>> >>>>>On December 07, 2000 at 06:16:59, Stefan Meyer-Kahlen wrote: >>>>> >>>>>>On December 06, 2000 at 21:08:54, stuart taylor wrote: >>>>>> >>>>>>>On December 06, 2000 at 14:57:34, Stefan Meyer-Kahlen wrote: >>>>>>> >>>>>>>> >>>>>>>> >>>>>>>>Thanks for playing this games. >>>>>>>> >>>>>>>>Before you continue with Shredder I ask you to download the patch I have put on >>>>>>>>our homepage today. There is a bug in the book learning in Shredder5 which gets >>>>>>>>fixed by the patch. >>>>>>>> >>>>>>>>Stefan >>>>>>>> >>>>>>>> >>>>>>>Do you really mean to say that that might have been the whole reason why CTs >>>>>>>results got better as the match progressed? >>>>>> >>>>>>No. >>>>>> >>>>>> >>>>>>> I think that both the opening book competition and the learning component are >>>>>>>artificial ways of gaining points, and spoil the true evaluation of playing >>>>>>>strength. >>>>>> >>>>>>Yes, but the book and all the learning stuff are part of the program. >>>>>> >>>>>>Stefan >>>>> >>>>> >>>>> >>>>>That's right. Every program nowadays has a book designed specially for it, and a >>>>>learning system in order to block the opponent from replaying won games. >>>>> >>>>>The book and the learning are weapons, but everybody has them now, so they just >>>>>counter each other and in the end we are back at evaluating real engine >>>>>strength. Actually a much better book can make a significant difference, but in >>>>>the case of Tiger and Shredder the books have been written by the best authors >>>>>you can find, so I don't expect a big difference in quality between them. >>>>> >>>>>The book and the learning system will hurt mainly programs that have a poor book >>>>>(or an old one) and no learning at all. That is, it will hurt only older >>>>>programs. >>>>> >>>>> >>>>> >>>>> >>>>> >>>>>>> Fscher randoms, and no-book openings, and equal number of K-pawn and Q-pawn >>>>>>>openings (both book, and no -book) and equally for each colour should be >>>>>>>included in tests (and, entire proccess should even be repeated at Tournament, >>>>>>>Rapid and Blitz-equally) >>>>>>>S.Taylor >>>>> >>>>> >>>>> >>>>>That's a good idea, and it would be nice if you could do these kind of tests. >>>>>I'm very interested in the results. My guess is that you are going to discover >>>>>that the results will be approximately the same than with the standard way of >>>>>testing (books ON). >>>>> >>>>> >>>>> >>>>> Christophe >>>> >>>>I would not mind running a Fischer random auto232 at 60 mins blitz. It I could >>>>do such a thing with all the different programs! >>>>Or even a themed opening. >>>>See how the programs play in say a Caro, or Budapest..... >>> >>>I'd prefer anything but Fischer random. >>> >>>Stefan >> >>Why? Maybe you don't like the name? >>But it would test things which books off (normal) might not test. It could show >>what would be a good opening if that were the normal opening position, if the >>computer truly understands the opening. >> It's all part of the program? But not the thinking part. >>But maybe, that is why chess ratings inflate much quicker than human ratings. >>Because the programs with new books crush the programs with the old books, even >>before thinking skill is improved. >>S.Taylor > > >I do not think that the main improvement in chess program is the engine and not >the opening book. > >The problem with fisher random opening is that you get positions that usually >are not from regualr chess games and it is possible that some improvement in >regular chess is not an improvement in fisher random opening. Exactly, that's my point. Playing with random positions after 10-15 moves is fine, also replaying the same opening the next game with reversed colors is very good. Stefan >For example the sides often cannot castle in random chess so knowledge about >castling cannot be used. > >I think that it is better to use random positions from practical games if you >want to test the strength of the engine without opening book. > >Of course the 2 programs should play both sides of the positions. > >I think that even playing games from positions like 1.a3 a6 is more similiar to >chess than random chess because the sides can castle and the ideas in the game >are more similiar to chess. > >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.