Author: Thorsten Czub
Date: 11:52:31 01/24/00
Go up one level in this thread
On January 24, 2000 at 14:19:53, Christophe Theron wrote: >Thorsten, an elo calculation can be done for a new program only if its opponents >have a known, well established elo (with many games, and a small error margin). correct. >This is why it is better to play against older program than against new programs >on 450, because the new programs have a too big error margin. no problem with me. the problem i have is with very old prgs. >It's maybe the 3rd or 4th time I repeat this to you. The SSDF results are VALID. its valid but senseless. i have not said they shall only play vs. NEW programs having NO elo so far. my point was: they shall play all on 450 Mhz and let the new programs play against the older ones. the older ones have the accurate elo, they have played many many games. but they shall not let them play against the very old programs on very slow machines, since this does not give precise results. >Of course you want to see the recent programs play against each other, but this >is not the job of the SSDF. Their job is to compute ratings as accurate as >possible. you misunderstand me. i do critisize that they always only let the top play vs. the top. so i call this incest. they shall let the new prgs. play against the older. and the new against the new.- but not the very old against the new, and not the very slow and old against the new. as accurate as possible means, not to let the latest programs on fast machines play against very old prgs on slow machines. even latest engines on slow machines against very old on fastest machines does not give precise elo. believe me. they do not measure strength than. >This is the job of passionated testers, like you, Shep, Enrique and others, to >play the recent programs against each other. nonono. you misunderstood my 3 points. i never said they should only play latest. big variance in programs = many many different programs... not only top NO variance in computer-speed = always 450 to get realistic results. THIS is my wish. >Let the SSDF do the less interesting work, and do yourself the interesting >stuff. We are listening carefully to both the SSDF and testers. its not a question of interesting or uninteresting. the results are more precise when you do not influence them by using strange settings. >But don't say SSDF results make no sense. not to me. and i said why. testing a porsche with me driving against schumacher in a trabbi (very slow and old eastern-german car) does not tell us much about how good i or schumacher drives car, or ? i say: give schumacher a porsche, and me a porsche, and than we will see.
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.