Author: Sandro Necchi
Date: 13:43:22 03/29/05
Go up one level in this thread
On March 29, 2005 at 16:36:03, Thomas Mayer wrote: >Hi Sandro, > >On March 29, 2005 at 11:22:57, Sandro Necchi wrote: > >>On March 29, 2005 at 06:58:46, WAEL DEEB wrote: >> >>>Hi, >>>Thank you for your accurate observations.... >>>The information that the author didn't use a pgn files for creating his opening >>>book is a BIG speculation,as he thinks that we are some kind of idiots or >>>something :-( >>>The statistics from the saved games results of the pgn are there.... >>>Cheers, >>>Dr.Wael Deeb >> >>About statistics: >> >>if a move has 99% score out of 100 games? Is it good? >> >>If yes, what about if game 100th was the confutation? > >that's why I always think about kind of a 2 pass book compiler. After having >compiled the hole pgn set it could walk again through the openings and look for >refutations -> e.g. when a move scores 75% but one of the opponent moves scores >even higher for the other site, the automated compiler should consider this... >-> of course this is far away from being perfect, but I think this could help >book cookers a bit in there complicate task to optimize a book. > >Greets, Thomas Well, I have found statistics about 20% correct. Yes, 80% not correct. There are several reasons to explain this: 1. The opponent found the correct answer/replies and the variation/move is not playable anymore. 2. The positions are not well understood by the program/different human player (we are creating the book for). 3. Those positions do not suit the program/human player style. 4. The resulting endgames are not well handled by the program/human player. 5. There are better replies than those considered which are found by computers/available. This is why I do not rely on statistics about openings. Sandro
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.