Author: Chessfun
Date: 00:41:17 12/27/03
Go up one level in this thread
On December 26, 2003 at 15:36:45, Mike S. wrote: >On December 26, 2003 at 13:54:53, Kurt Utzinger wrote: > >>Please refer to (only in German) >>http://f23.parsimony.net/forum50826/messages/86999.htm >>I try a short and abbreviated translation of what Michael >>Scheidl wrote: >> >>A new, "less shorter" short book for ChessBase programs 5moves.ctg. >>Book depth 5 moves or 10 half-moves, 65.243 positions, data base of >>the book generated out of 250.000 games with players having at least >>2350 Elo and game length between 20 and 80 moves, no manual adjustments >>but Michael let the book learn with over 30.000 computer games (1990-2003). >>The book option (F4) should be set to "normal" to allow enough variety. >>Download: >>http://members.aon.at/computerschach/5moves.zip >>(1,43 MB zipped, almost 5 MB on HD) > >Thanks Kurt, you have metioned all the important infos. I can add that I've >tried to check carefully for illegal moves in the database it's based on, >removing quite a number of them (I don't know if these would have been imported >into the tree, causing problems). It includes moves from historical games too, >and even Elo performances for them, because I've let calculate Elo ratings for >these. Also, I want to emphasize that it is indeed crucial to use the *normal* >F4 book options for a usage that makes sense. > >@Sarah, some elaborations: It is intended to be used in "book-neutral" engine >matches and tournaments, where every engine uses the same book. I think it is >more attractive than usual sets of openings variants, giving more variety, and >also the unpleasant effect of too long book variants is avoided. Kurt and ohters >have critzised this, when virtually no middlegame exists in computer games. With >the 5move.ctg, engines have theory support up to the 10th ply (sometimes a bit >less), and must finish the opening themselves. This may be the downside from the >engines viewpoint, but OTOH hand they can develope their own middlegame in these >games. Also, the user gets more engine creativity to see right from the >beginning of the game. > >Due to the creation process and because I'm not theory expert, some newer >refutations of common moves of the past may be missing (or may not have proper >probabilities, as refutations typically will have appeared much less often). No >opening theory which wasn't tried in master games yet, is included. - But I hope >these are minor problems. Some blitz tests have been made, i.e. by Kurt with 100 >(!) games, and only very few evals bigger than +/- 1.00 have been found among >the first calculated moves. > >Regards, >Mike Thanks Mike, Kurt, for your explainations. I'd agree the normal draw, remis and 2600 ctg's do generally lead themselves to books without much middlegame play. The test is interesting I was just curious where and how the book had developed after seeing all the + and - to book scores for moves. I'll run a couple of hundred myself, I'm curious to see score out of book also. Sarah.
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.