Author: Uri Blass
Date: 12:29:46 10/26/03
Go up one level in this thread
On October 26, 2003 at 15:02:11, Peter Berger wrote: >On October 26, 2003 at 13:15:41, Uri Blass wrote: > ><snip> >>>>>It was a surprisingly one-sided match where Movei was without any real chances. >>>> >>>>From looking at few game it seems that movei had chances but blundered. >>> >>>OK, agreed :). But the games showed me that until movei has improved further in >>>the endgame this opponent is just too tough, and it's better to try others >>>first. >> >>I am not sure >>I expected a win for Genius2 but a smaller win. >> >>Remember also that 6.5-1.5 is not a significant result >>Junior was leading 5-0 against Fritz and Fritz won the match. > >OK, I can't resist. My statistics is extremely rusty and I am lazy (there have >been some posts by Dieter B. some time ago that had some maths in Monte Carlo >experiments). But the normal distribution might be a bad guess for engine >matches. > >6.5-1.5 can be done in different ways, e.g. like in the match, or 5 wins and 3 >draws - the latter suggests a potentially better performance I think. I am sure >you know what I am talking about. > >The movei- Genius match is very close to being statistically significant (with >95% margin), even more as the single win was just book (that's for the humans ;) No Movei was out of book at the 5th move. Genius blunder in the opening but opening is also part of the game. >) > >When it is about the Junior-Fritz qualifier, I was surprised that there was >little discussion about the results/games (other than conspiracy theories :) ). >My theory is slightly different than bad luck. You can see the phenomenon nicely >and to the extreme in the SSDF Hiarcs games. A book has to be well prepared to >deal with the "optimal" book learner settings of the ChessBase interface, and it >is my impression that only the Fritz book really is. The learner works fine with >a wide default book of GM games in general - for a manual book there is some >danger that lines that are supposed to be only used rarely turn up much more >often than intended. > >For manual games played for fun, it's also a practical decision. In case movei >smashes other opponents there can always be another go against Genius 2 - it >would be a waste of time to try to get everything done in a statistical >significant way, when alternatives might produce more interesting data :). > >Cheers, >Peter I believe that movei of today is slightly weaker than Genius2 but I guess that 6.5-1.5 is too high to give the correct difference. Maybe in a long match it can be even worse because movei has almost no learning except learning to change first move after seeing a bad evaluation against itself(If I remember correctly more than 5 pawns against itself) that prevents losing the same game twice in a match of 4 games. 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.