Author: Dann Corbit
Date: 01:15:01 01/04/06
Go up one level in this thread
On January 04, 2006 at 03:37:50, stuart taylor wrote: >On January 03, 2006 at 13:17:41, Dann Corbit wrote: > >>On January 03, 2006 at 12:00:45, stuart taylor wrote: >> >>>Is Hiarcs 7 below all those others mentioned? >>> >>>And can I find somewhere where Fruit has individual scores against Hiarcs 6 and >>>7 and others, since that period? >>>S.Taylor >> >>Get the full list: >>http://web.telia.com/~u85924109/ssdf/rlwww061.txt > >Thanks, Dan. > >But it looks like the lower listed, older programs, cannot stay correctly >calibrated, because they're not being tested any longer. >What I had been looking for was to see at which point Fruit wins ALL games. >But Fruit does not play those which are lower down, at all. There is too much Elo difference for the really old programs to impart information. If you have a 600 Elo difference, the result will be rather uninteresting. In order to get the best data, (IMO) the following sorts of things would be ideal: Programs that are 100 Elo or so weaker. Programs that are 100 Elo or so stronger (if any exist). Programs that have played lots of games are best. Programs of equal strength are liable to random walk more than programs that are quite a bit weaker or stronger. (E.g. Tyson fights Hollyfield 10 times, it might go 7-3 or 3-7 but a much weaker fighter or a (mythical) much stronger fighter won't be quite a liable to such fluxuations. But programs of the same Elo are not harmful, especially if they are well calibrated. And if you play enough games, eventually it will even out. So the best programs are those in the Elo ballpark, and with lots and lots of games played. Win expectency for a difference of 0 points is 0.5 (half of the points) Win expectency for a difference of 100 points is 0.359935 Win expectency for a difference of 200 points is 0.240253 Win expectency for a difference of 300 points is 0.15098 Win expectency for a difference of 400 points is 0.0909091 Win expectency for a difference of 500 points is 0.0532402 Win expectency for a difference of 600 points is 0.0306534 As you can see, at even up to 300 Elo, the opponent program will score some points. At 400 Elo, it will score less than ten percent, and at 500 Elo, less than 5%. Now the problems with that is that a couple random wins by the -500 Elo program will skew the results badly. Imagine if you play an opponent 2000 Elo weaker and it happens to win one game, by some fluke. It will make the whole set of games just a collection of not-so-good data. And so you can see why the SSDF plays a very large number of games before they report the results. One single batch really tells you very little.
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.