Author: Bob Durrett
Date: 08:22:02 11/16/03
Go up one level in this thread
On November 15, 2003 at 19:10:37, Georg v. Zimmermann wrote: >The book is part of the engine. But let us agree to disagree there for a moment. > >What other problems are there with your approach ? One other problem is that you >would be testing the analysing strength, not the playing strength of the >engines. > >Why is that ? >Suppose you have one engine which really likes open games, therefore hates 1.d4 >with white for example or 1.e4 c6 as black. When you now force it to play 1.d4 >in maybe 40% of the games, it will perform poorly as opposed to an engine that >does comparably well in open and closed games. >This performing poorly will show you the bad analysing skills (since it doesnt >understand closed positions), but will give you wrong information about playing >skills (since it in "a real match" would have never allowed itself to get into >such a position). > >Best regards, > >Georg > > >On November 15, 2003 at 17:47:04, Eye Witness wrote: > >> Though some do, i do not consider the opening book really apart of a >>program(meaning the engine), i consider it a seperate piece of software in the >>package with a program. Most people here are really concerned with the engines >>playing strength, and the engines ability to select correct moves. If all progs >>had the same opening book to start, this would more clearly indicate the >>strength of the chess program. I do not in anyway support matches with no >>opening book, as some opening knowledge is apart of chess,and even the most rank >>of tournament chess player has some type of opening book knowledge. It seems >>better to judge the programs analytical ability than to judge its preparation. >>It seems also that programmers would prefer this to see this so that they can >>know that there program is losing or winning matches, not because of bad book, >>but because of good or bad engine programming That is very true. The "total chess-playing program package" is what SSDF is testing. They are not testing chess engines. [Similarly for endgames and endgame tablebases perhaps.] There is another side to this coin, however. When a chess-oriented consumer purchases a chess-playing program, that consumer will want a program which can strongly play the middlegames from all popular openings. The consumer probably will not want a program which either never plays 1.d4 or does so rarely. Worse would be a program which would play poorly as White if forced into a 1.d4 opening. Marketing people understand these marketing considerations. On the other hand, a programmer embroiled in the guts of engine programming might lose sight of such considerations. To satisfy the needs and desires of most consumers of commercial chess-playing programs, the SSDF people COULD decide to try to encourage chess programmers to produce engines which would play well from all positions resulting from all popular openings. The best way to do this might be to mandate use of an opening book which gives weights to the various openings such that the popular openings would be seen most often during the testing. Having a single opening book for everybody to use would make sense in this case. SSDF can do whatever they wish, as is their right, but I would like to see them take an active role in promoting the development of chess engines which play equally well from all openings. Maybe doing so would have the additional benefit of minimizing some of the cutthroat infighting. Bob D.
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.