Author: Robert Hyatt
Date: 07:07:37 02/18/05
Go up one level in this thread
On February 18, 2005 at 05:38:26, Tony Werten wrote: >On February 17, 2005 at 16:15:14, Robert Hyatt wrote: > >>On February 17, 2005 at 14:50:57, John Merlino wrote: >> >>>On February 17, 2005 at 14:23:20, Robert Hyatt wrote: >>> >>>>On February 17, 2005 at 14:15:57, Sune Fischer wrote: >>>> >>>>>On February 17, 2005 at 14:03:30, Tord Romstad wrote: >>>>> >>>>>Well 700 Elo is equivalent to about 5-6 pawns material advantage, >>>>>I don't think I have ever seen that in an actual game much less >>>>>seen it on average. >>>>> >>>>>I can believe in a good book giving half a pawn or ~50 Elo, >>>>>not much more than that is realistic IMO. >>>>> >>>>>Perhaps the person you refer to is talking about a book >>>>>with "perfect chess" reaching 80 plies deep? :) >>>>> >>>>>-S. >>>> >>>>No, the person he is talking about simply lives in an alternate universe where >>>>our normal rules of physics and math do not apply... >>>> >>>>I don't see why anyone would even bother participating in that particular >>>>discussion, much less running tests. I claim that water freezes at 12.7C, >>>>who is going to run a detailed test to see if that is right or not? Or is >>>>common sense enough? :) >>> >>>I can't believe I'm going to do this. But, to defend Vincent and Arturo to some >>>degree, I'm PRETTY SURE they were referring to a book that was specifically >>>designed to be played against a single opponent. Somebody please correct me if >>>I'm wrong. >> >>Even so, 750 points? Against a program using a random unknown book with >>learning? I personally don't buy it. Vincent's quote "you lose the first game >>and the next 1980 games after that." I can only say my program would not lose >>the next 1980 games by playing the same again and again, with or without the >>opening book. > >Doesn't "no book" also imply "no book learning" ? From Vincents quote, he seems >to mean it that way. No idea. With no book, the only program I know of that will still do "book learning" is MchessPro. But position learning is a different matter, and is independent of book learning... > >Maybe positional learning will prevent you from loosing the same game time after >time, but not from loosing the same 5 games time after time. It will eventually prevent me from losing _any_ game an excessive number of times, again assuming that in general my opponent is not significantly stronger than my program. Vincent said "1920 times in a row". Won't happen. > >Tony > >> >>> >>>So, the only accurate way to test this (regardless of your argument that it >>>doesn't need to be tested at all due to "common sense" -- which may be a fine >>>argument but I'm not too sure it holds up scientifically :-) would be to create >>>a book that is designed to exploit the weaknesses in Hiarcs' book, and then test >>>with that. Then compare the results to using NO book, which, I believe, Vincent >>>was arguing reflected the other end of the 700-point range. >>> >>>Will it show the possibility of a 700-point ELO gain? I very highly doubt it. >>>But I do think it will result in a much bigger difference than the 3 points out >>>of 100 that came from the first test. >>> >>>jm >> >> >>Here is the fly in the ointment. I'll be happy to run the current version of >>Crafty, using the current book, for 3 months. Vincent or anyone can book up >>against it until their hearts are content. Then I will enter a match with them, >>either using a book or not using one. But I won't necessarily use that exact >>version of Crafty for the match since I _always_ enter something more recent >>than what has been publicly distributed. What is the probability I will get >>killed by a program that normally plays equal with me if we both use random or >>no books, but I lose 99 of every 100 games (750 rating points is somewhere >>around that win rate) when they use their super book and I use either no book or >>a random book? I'll bet the two programs will play within reasonable bounds >>about the same as they did earlier. No way to get a 99:1 win rate by just a >>book, particularly when you are claiming you can extend that streak for 2000 >>games. It just won't happen... A pre-planned opening against a specific >>opponent can win a single game, no question about it. I have done this in the >>past, multiple times. But not repeatedly against a book with randomness and >>learning, or against a version of the program that might well play different >>from the first move out of book.
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.