Author: Uri Blass
Date: 10:25:05 02/15/05
Go up one level in this thread
On February 15, 2005 at 12:59:09, Vincent Diepeveen wrote: >On February 15, 2005 at 12:40:22, John Merlino wrote: > >>On February 15, 2005 at 12:36:10, John Merlino wrote: >> >>>>I hope that you realize 750 rating points means roughly 0% chance. >>>> >>>>What is the chance in YOUR opinion a program without book in 2005 has to win a >>>>world champs event? >>>> >>>>If you say 0%, that means 700+ rating points. >>>> >>>>Vincent >>> >>>I'll try to answer all of your posts in this one response, so as to not have to >>>bounce around this thread. >>> >>>The obvious flaw in your argument above is that you are all of a sudden talking >>>about winning a championship, but *I* am talking about using ratings to >>>determine the statistical likelihood of SCORING POINTS in a single game, which >>>has been the point of this discussion, unless I am grossly mistaken. >>> >>>As somebody else alluded to in this thread, you can be less than 300 points >>>behind the highest ranked person in a tournament. But if you are the >>>lowest-ranked player in this tournament, and there are a lot of other >>>participants, your statistical chances of winning the tournament are practically >>>zero. >>> >>>So, I agree with you that a program going into the computer world championships >>>without a book has close to a 0% chance of winning the tournament. But this >>>would also be true if all programs were of theoretically equal strength, and not >>>having a book only decresed the strength by 300 points. >>> >>>You also refer to a human playing some very large number of games (I think you >>>said 5000?) against a program without a book, eventually allowing the human (or >>>engine with learning) to beat the program close to 100% of the time. This is >>>also WAY outside the boundaries of this discussion. But the clear refutation of >>>that argument would be to say "Maybe so, but what happens in the first 10-20 >>>games of that test?" I'd bet that your theoretical human, who, let's say, is 700 >>>points weaker than the engine (just to pull that number out of the air), would >>>lose almost all of those games. >>> >>>Finally, I'm not sure why you are all of a sudden talking about Chessmaster not >>>entering a world championships. Admittedly, I did bring up my very brief tests >>>with Chessmaster on ICC that took place well over two years ago, just to provide >>>some evidence that a strong program without a book can still perform decently >>>against other strong engines, even occasionally beating them. But as for the >>>reason that Chessmaster does not enter the WC, you should ask Johan what it is, >>>because it has always been his decision. >>> >>>I haven't been involved in Chessmaster in more than two years, so I can't >>>comment on the current situation. I wouldn't even venture to guess as to what it >>>might be -- but I'm sure you know his e-mail address, so why don't you just ask >>>him, instead of bringing up something that has nothing to do with this topic? >>> >>>jm >> >>One more point. Even Arturo has been referring to this "well-tuned book" being >>specifically prepared for a single opponent. And this is all well and good, and >>of course preparation for your opponent is vital. However, could this one book >>be used equally successfully against ALL opponents in a tournament. Clearly the >>answer is no, and it might even be detrimental against other opponents. >> >>So, once again, I think we may be talking about different things. You and Arturo >>(and others) are talking about a book that is designed to be played against >>another specific engine, and Uri and I (and others) are talking about one >>"generically strong book" that is intended to be used against all opponents. >> >>jm > >Uri doesn't know what he talks about anyway of course. He still thinks 1.h3 is a >good book to test ones engine with. >I never saw him take that back. > >This discussion goes way over his head. > >My point is very simple. > >You say: "close to 0%". > >The point being made is that a program at todays hardware (and not some >imaginary hardware from the year 2100, nor a hardcoded tournament book in the >executable like rebel had it) has a hard 0% chance to win the world champs. > >I specifically mean world champs as the strongest opponents show up there. Not >just some amateurs. > >So i very clearly want the discussion here that it is a hard 0% and not 'close' >to 0%. Close to 0% is also 10%. > >It is not 10%. It is not 5%. It is 0%. And not 0,001%. It just never has >happened. And it never WILL happen. There is only 3 possible results in chess. >You win a game, you draw a game, or you lose a game. > >So we cannot calculate with 0.00000001% if there never is going to be >0.0000001%. > >There have been last 9 years precisely 2 winners. Shredder or Junior. > >So chances 'near zero' is not a good definition at all. > >We want hard formulations. As you win a title or you don't win a title. > >It has 0% happened so far that an engine without book won that title. > >It has a 0% chance. > >A hard 0% chance. > >Vincent There is no proof to hard 0% chance. If the chances for movei with no book is only 0.001% you will probably do not see it winning even after 1000 tournaments(because it means one win for 100,000 tournaments). Uri
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.