Author: Rolf Tueschen
Date: 11:25:39 02/24/06
Go up one level in this thread
On February 24, 2006 at 12:20:26, Roger Brown wrote: >On February 24, 2006 at 10:53:57, George Tsavdaris wrote: > >> >> You should avoid that kind of matches for obvious reasons! And by obvious >>reasons i don't mean the funny score but the possible vitiation the results may >>have. The top leader's ELO rating may be wrongly calculated if we match them >>with very much weaker opponents.... > > > >Hello George, > >Actually, given that Fritz 5.32 has played so many games, its rating is probably >quite accurate. This is wrong. Accuratess isnt an objective term, so a rating belongs in its time and surrounding. > >The aim of matching engines with Fritz 5.32 is to obtain a means of calibration. No, completely wrong. You could calibrate machines on humans and their Elo, but you cant take an old Fritz 5.32 and then calibrate all new programs, this is absolutely impossible because Fritz 5.32 belongs to an older, long time overcome generation of programs. And now the important hint: you just dont have a normal distribution because you have gaps through completely different technology. Learning features or table bases or whatever. And we dont have started to talk about differences in strength... > Should Elo values be computed from a changing pool then the error bars on the >ratings would make them useless for comparison purposes. Exactly and therefore I dont understand the proposals above. > >This would seem to be an approach consistent with sound statistical analysis to >me. > >Later.
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.