Author: Thomas Mayer
Date: 21:57:30 02/25/04
Go up one level in this thread
Hi Uri, >> By the way: the opening book and the handling of the opening book is a big >> part of the engine strength. So in my opinion Thoralf is correct when he >> says that it would be somewhat unfair to the others, when one engine uses >> the environment of the other products. E.g. when I think about my own >> engine just using general.ctg and the advanced book learning of say Fritz >> GUI this would boost it in such matches about 100 Elo, I am quite sure >> about that number. I would never consider this as fair. > Based on what is your confidence? based on my own experience. 2 years ago I made a little test between a Quark engine with a random book playing EVERY line that is in book (even bad ones), a Quark without book and a Quark with a somewhat optimized book. They played around 100 Games against each other if I remember correctly. The version with the fully random book get last, around 100 Elos behind the version without book. The version with the good book was again around 100 Elos better then the version without book. A version with a booklearner would have done at least again 100 Elos better. A booklearner helps mainly in such long matches between engines. > I do not know and I would like to see the advantage that programs get from > book. Maybe I will repeat the above described match once again. Especially when I have a booklearner. I am myself interested in the outcome. > I understood based on results that there was no big difference between > chessmaster out of box and chessmaster with book under chessbase when it also > had learning enabled. Well, afaik not very much games with ChessMaster SKR were played. Also the different settings make a difference. Additionally we do not know how well the used book for ChessMaster SKR fits to the engine. There are to many unknown factors to make any conclusion. > Quark does well in Leo's tournament without the advanced learning of Fritz. Anyway you might see a difference. The book Quark uses at Leos tournament is very good. There is nearly no tourney where Quark scores as good as in Leos tournaments. Of course the time control and ponder=on might interfear the result, but I think that the results at WBEC clearly show the differences of books. For a booklearner the 4 games played against every opponent might be to less to really help. They may fix some very wrong lines, but not optimize the book for an opponent. With e.g. 40 games (like SSDF usually plays) against the same opponent this is a different story. > It is significantly weaker than the top programs but I doubt if it performs > better in nunn match experiments. Quark IS definitely book dependant. With some kind of positions it has big problems. So the job of the book is to avoid those kind of positions as good as possible. Greets, Thomas
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.