Author: Peter Berger
Date: 17:12:15 10/01/04
Go up one level in this thread
On October 01, 2004 at 02:33:56, Peter Skinner wrote: >On September 30, 2004 at 23:52:18, Robert Hyatt wrote: > >>The thing is, they play X games with one computer, so learning ought to fix this >>up unless the Nimzo lines are so well plotted that everything it tries fails. >> >>Making a special SSDF opening book is not something I have time to do. Nor >>anyone else I know of. IE Peter has some stuff from the WCCC, but we'd not want >>to reveal that via SSDF testing as we will likely play in the next WCCC event >>and will need a unknown book for it... > >I don't think it would require a drastically changed book. Just something to >differ from 1.d4. Just to shake things up a bit, and vary from what is being >expected from Crafty. > >I found while playing on FICS that when I modified the book to play either 1.d4, >1.e4, 1.Nf3, there was a greater winning percentage in the games, using lower >hardware. > >The main problem with all the Crafty clones out there is they might make their >own book.bin files, but almost always have your books.bin, and bookc.bin. When >you play 50 Crafty clones in a row and all open basically the same, ie. 1.d4, it >is easy to gain learning. So when I varied the opening book, I found the >previous learning or "expected" openings played weren't forthcoming, I had an >advantage. > >I think the main thing hurting Crafty in current computer-computer games is the >opening standard. During the last CCT and the WCCC Crafty was running on purely >massive hardware, and due to that was able to overcome some possible short >comings from book openings. The average user doesn't have a Quad Opteron in >their back pocket, thus the opening selection has to be just that much better. > >I could be wrong, and might have a completely different view than yours, but I >think it deserves a serious look. > >Peter This is a question of approach, not of right or wrong. E.g. in the WCCC game Diep-Crafty , Crafty was out of book after 1. d4 d5 2. c4 e6 3. cxd5. This was a completely unexpected move order, so Crafty had to deal with it on its own - similar as a human who is out of book would have to. It's the question if you trust the engine to deal with some problems on its own, or rather have it fall back to repeating unchecked moves played previously by some human masters that might be dubious. Crafty would have even been out of book after 1. g4 e.g., while being booked up reasonably well against more usual anti-book setups. Some of this was due to limitted preparation time, but in reality it can even happen with a thoroughly prepared book ,just because of basic choices. Also learning is meaningless in tournaments or can be overruled by the superior ways of learning a human can use. I don't buy your hardware argument here - the main opponents of Crafty at the WCCC tournament had no hardware disadvantage at all. And while book problems *look* obvious if the engine is out of book at move 3, there is also another side of the story. Crafty got an extremely promising position out of book against Shredder, and against both Fritz and Junior both sides agreed that at some point close to the end of book it was doing very well. Such a book is basically useless for SSDF tests though . You have to be wide to compete in such an event, while having weeded out all the weak lines at the same time. This might be a much tougher job to do, but at least it is defnitely very much different. Peter
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.