Author: Uri Blass
Date: 05:54:13 08/26/02
Go up one level in this thread
On August 26, 2002 at 08:31:02, Andreas Guettinger wrote: >On August 26, 2002 at 08:07:26, Uri Blass wrote: > >>On August 26, 2002 at 07:22:28, Steve Coladonato wrote: >> >>>There is a very long thread here concerning the use of a book developed for one >>>program being used by another program. I am not a programmer, so I don't >>>understand the ramifications of this. But, I do have a question(s). If the >>>same book were used by all engines, would that not be a fair comparison of the >>>engines strength? As long as a program is "in" book, it is not using any of its >>>internal algorithms so the moves it is making are recognized as "best" for a >>>particular line/opening. >> >>I doubt if it is a good idea. >> >>I think that there should be a small book when the program plays moves in 0 >>seconds and a big book when the program read the book but does not play the >>moves in 0 seconds. >> >>I saw cases of big tactical mistakes by program because of a mistake in a book >>happen and it is easy at least to prevent it(if we know that a move is good and >>the program evaluates it as a blunder we can put it in the small book). >>> > >I think in general it is the other way round. A lot of critical opening moves in >complex chess openings will not be found by engine pondering even in hours. They >are the work of GMs and hundreds of assistants over decades. As in your case, >that would only proof to be a poor or out of date opening book, where bad lines >are not eliminated. I am not talking about finding moves but about finding blunders. I saw cases when programs played tactical blunders in the opening that they could avoid by searching. I do not say not to play a move that is 0.2 pawns weaker but not to play a move that is at least 1 pawn weaker unless it is in a special small book. It is possible to start by analyzing every position in the big book for 1 second in order to find moves that are suspected to be losing more than 0.5 pawn. The number of these moves is going to be relatively small and it will be possible to analyze these move for longer time. If the computer still evaluates the move as a blunder then the programmer may look at the moves and decide if to include them in the small book. > >An interesting thought crossed my mind when I read the original post in this >thread. What humans actually do in tournaments is choose the opening in respect >to the current opponent, not to fit his style. Anybody designed an engine yet >plays book moves according to its current opponent? > >Andy I think that the ssdf forbid knowing the opponents and I think that this should not be allowed. It is possible to play a killer moves against every opponent and I do not consider it as fair because it is against old programs. In this case programmers can autoplay a lot of comp-comp games against old programs and play the lines that it won when the program play in the ssdf games against the old programs. It is going to increase inflation in the ssdf list. I think that even in the condition of today the book may be a big problem in the ssdf because it is possible that someone is going to have a killer book based on thousands of games against Fritz7 and Junior7 and his program is going to beat them by repeating lines that Fritz7 or Junior7 have no idea about because they do not know the secret games that he played against them. Thanks god that it did not happen but I think the only way for the ssdf to prevent it is simply to stop playing with programs after they become public. It may be better if every programmer that want the ssdf to test his program will have to promise not to release his program in the next year and his program is going to be tested only against programs that were not released and got rating.
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.