Author: Gordon Rattray
Date: 16:02:06 02/03/02
Go up one level in this thread
On February 03, 2002 at 17:39:51, Mike S. wrote: >On February 03, 2002 at 13:50:40, Gordon Rattray wrote: > >>(...) >>I also disagree that getting computers to play Fischer random chess is a good >>way to test their strengths at *chess*. > >True, if you are talking about the "complete definition" of chess including the >standard starting position. But I think, Fischer random and shuffle chess can be >a good way to test their *chess skills*, especially opening skills. It may >happen that if a program lacks basic opening skills (because the programmer did >rely on having an opening book always), it will often fail early in it's shuffle >chess games. IOW. there will probably not be much skill testing with such a >program after the opening. But I don't expect that this happens too often, or in >every shuffle position. > >>For example, what if a program >>understands very well about the pros and cons of a fianchetto bishop. Such >>knowledge won't be so effective in Fischer Random chess because there will be >>less opportunities to fianchetto a bishop. > >Probably not, but in practise of "normal" computer chess the books decide if >there is a fianchetto or not (and play 10 more moves from the book until the >engines start). Knowledge how to use the fianchetto bishop can come into effect, >but not a knowledge about developing the bishop like that (or at least very >seldom). > >In f.r. or shuffle, there will quite often be positions, where the engine can >show if it is able to play a fianchetto by itself (or similar, i.e. sometimes a >g1 bishop is developed by h2-h3, g1-h2 etc.). > >Some time ago, I let play a little shuffle chess tournament between old chess >programs and chess computers. If you are interested: > >The games (PGN): >http://meineseite.i-one.at/PermanentBrain/oldies2/old2_pgn.zip > >Description (in german language) >http://meineseite.i-one.at/PermanentBrain/oldies2/oldies2b.htm#modus > >Includes the FEN of 18 shuffle chess positions which have bishops of different >colors and allow the standard castlings (one of which is the standard position). > >Regards, >M.Scheidl > > >P.S. Shuffle Chess Oldies Tournament, results: > > ----------------- Novag Sapphire 4.0/6 > | -------------- Mephisto risc 1 MB 3.5/6 > | | ----------- Mephisto Lyon 16Bit 3.0/6 > | | | -------- CXG Super Enterprise 0.0/6 > | | | | ----- SciSys Turbo 16K 0.0/6 > | | | | | -- CXG Sphinx Dominator 0.0/6 > | | | | | | > 1 Socrates 3.0 1 1 1 1 1 1 6.0/6 > 2 Sargon V 0 0 1 1 1 1 4.0/6 > 3 Kallisto 97 Aegon 0 1 0 1 1 1 4.0/6 > 4 Chessmaster 3000 1 0 0 1 1 1 4.0/6 > 5 Kasparov's Gambit 0 0 1 1 1 1 4.0/6 > 6 LChess 3.0 0 ½ 0 1 1 1 3.5/6 > >The board computers were running at 2h/40, the oldies programs were set at >20:00/60 moves on a PII/333. You've mentioned a lot of good points. Also, thanks for the test results. My opinion is based on the fact that I don't think it is unreasonable for a programmer to rely on an opening book. The same applies to endgame tablebases. I accept that other people are interested in a program's performance without such pre-computed knowledge. That's fair enough, but I myself don't view it as being unfair or false. I just view it as exploiting the strengths of a computer and/or the programmer(s). As a further example to the "fianchetto bishop" example that I mentioned, what if a programmer researches common pawn formations (e.g. isolated queen pawn, closed centres, etc.) and then invests effort in getting his/her program to "understand" such middlegames. Then, I'd guess that Fischer random chess will result in a different distribution of typical middlegame positions and it may be entirely different knowledge that the program requires. So, what can results for Fischer random tell us? I'd guess very little. Gordon
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.