Author: Aaron Tay
Date: 02:42:16 09/01/02
Go up one level in this thread
On August 31, 2002 at 05:29:21, Mogens Larsen wrote: >On August 31, 2002 at 02:10:25, Aaron Tay wrote: > >>I agree. You can find out if games stay in book longer with longer time controls >>if you keep an average of the number of book moves that the engine stays in >>book. It depends on the engine also. I have the impression Crafty finds book >>moves better than Yace at the same time control. And The King is also pretty >>good. > >That matches my observations as well. > >>Of course if this trend holds it may imply that autogenerated ametuer books >>might even be weaker than no books when faced against Professional books, since >>the autogenerated books will stay in book longer with greater chances of falling >>into a trap! > >That is true to a certain extent. Though I suspect that the deficit isn't >horrendous given that the book is something more than an ordered PGN file, where >main attributes are wide and random. This means choice by statistics, controlled >opening repetoire and learning are requirements for decent results IMO. At least >against your average commercial book, which is quite wide compared to those used >at championships (or so I imagine). It's not unlikely that an "out of book" book >could yield better results than your average autogenerated book. True. But i would think that the average autogenerated book might end abruptly at certain critical points where the chess engine would still need guidance. I would think a handtuned book would not have this problem. So if your autogenerated book abruptly stopped at move 20 of a complicated yuogslav dragon sicilian, the engine might come up with plausible moves but those which have being known in theory to be bad (this is where the wide commericial books with replies to even suspect moves shine). This is somewhat similar to a engine without opening books coming up with the first 20 moves unaided and stumbling on the 21th. Also as mentioned before by many others, tricky lines (!?) that are not played in GM practice , could be sprung on unsuspecting amaetuer using autogenerated books. In championship games, I could see people preparing such lines by testing the latest versions of their opponents to see if the engines react badly. If they do it's going to be placed in the books.. Autogenerated books based only on success percentage or times plays have this flaw, they only include lines that work or are promising. Like a human who memorises only modern GM moves, they can be upset by opponents who play non-book moves that have a drop of venom, but are harmless if the correct reply is known. The fact that the move is never played at GM level is because the correct reply is known at GM level, but the average human or computer might not know how to respond. Some engines maintain opening books for white and black, this can be used to store replies to tricky moves. But you are unlikely to find those moves in games..At most once... >Regards, >Mogens
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.