Geschrieben von: / Posted by: Thomas McBurney at 03 March 2004 14:40:03:
Als Antwort auf: / In reply to: Re: Does Movei 0.08.174 deserve a better opening book!? geschrieben von: / posted by: Wael Deeb at 03 March 2004 12:00:26:
new movei is probably better than the movei in that ranking list and may be better than the baron but I doubt if it is also better than Abrok.
Maybe the conclusion should be different and the nunn openings are simply not good for movei or different time control caused different results.
Uri
Abrok is still slightly stronger than Movei.It's very interesting because I realy intended to include Abrok 5.0 in this tournament,now I'll do it
I fully agree with you
Once I told you that the nunn openings are not identical to the real game in a tournament considering the engine performance!
Regards,
Dr.WAEL DEEB
P.S.Please,we need more opinions about this subject!
After playing around with opening books for a while with my engine, I have come to the conclusion that an opening book does provide an advantge over one that doesn't. But I don't think it is a clear advantage.
An opening can provide a time advantage. Lets say the game had a time control of 40/40 and the chess engine with an opening book played the first 10 moves from book. This means the engine can now play the next 30 moves in 40 minutes which will give it an average of an extra 20 seconds per move.
A chess engine with an opening book can vary the games it plays. I have seen chess engines play the exact same game over and over because it and the opponent didn't have an opening book. You also have to think about the person who simply uses your engine to play games against, you don't want to bore them with the same opening played every time.
Chess engines with a small opening book or no opening book can be exploited more easily by engines that have learning. A chess engine with a small opening book probably doesn't vary the games it plays as much as an engine that has a large opening book. So a chess engine with learning would not need as many games to learn which book moves lead to a win against engines with a small or non existant opening book.
Opening books can help an engine avoid or set traps/pit falls which the engine would otherwise have to perform a search to find. The faster the time control and/or the slower the computer, the less likely a search will find these solutions.
The problem with opening books is that the moves they contain were usually made by grandmasters. This sounds like a positive thing at first and I think it is generally but this often leads to some problems. Grandmasters do not play the same way as computers and this can lead to positions the chess engine does not understand. Gambits are probably a common source of problems for many engines. Generally, opening books are created by extracting moves from PGN's containing human vs human games, and the problem with humans is they do make mistakes. So it is very easy to add blunders to an opening book even if that particular move lead to a win and was made by a super grandmaster. An opening book containing thousands of moves can be very time consuming to check every move for blunders, which is why I guess many authors don't it.
So I have come to the conclusion that for a chess engine to benefit from an opening book, the author has to put in a bit of effort in to get it right. The author may even have to tailor the opening book for the engine. The goal is to come out of the opening book with a positive score and a good time advantage. Even if the author has put in a lot effort in the opening book, I think it only provides a small improvement overall to the engine, but then, you have to watch out for those engines that have learning. Once they find a hole, they will happily keep exploiting it.
Here is an experiment I completed today.
Kanguruh v1.82 vs TSCP (40 games @ 40/40, pondering off)
Kanguruh with learning off...
won 11, lost 22, draw 7
Kanguruh with learning on...
won 34, lost 2, draw 4
As you can see, Kanguruh benefitted from learning which was helped by TSCP's small opening book.
Cheers,
Tom.
