Author: Michael Yee
Date: 11:53:13 10/18/03
Go up one level in this thread
On October 18, 2003 at 14:35:25, George Sobala wrote: >On October 18, 2003 at 12:02:18, Uri Blass wrote: > >>On October 18, 2003 at 09:28:18, Michael Yee wrote: >> >>>On October 18, 2003 at 04:35:22, Uri Blass wrote: >>> >>>>On October 17, 2003 at 23:42:35, Mike Byrne wrote: >>>> >>>>>"there is no sign that the top-200 players are losing ground at all against the >>>>>top computers." >>>>> >>>>>http://www.chessbase.com/newsdetail.asp?newsid=1244 >>>>> >>>>>...an interesting read... >>>> >>>>You cannot compare apples with oranges. >>>>Time control of 30 minutes per game is different than 2 hours/40 moves and you >>>>cannot put both time control in the same group when you do comparisons. >>>> >>>> >>>>Uri >>> >>>Well, Sonas claims that rapid and blitz results do correlate well with FIDE >>>ratings... They just would need different weights than long time control results >>>if you wanted to use them for rating calculations. >>> >>> http://www.chessbase.com/newsdetail.asp?newsid=562 >>> >>>Michael >> >>I do not believe to everything that sonas claims. >> >>Uri > >I agree. He touts himself as a super statistician yet when deriving his Sonas >ratings, he validated his results by testing against the same dataset he used to >generate the ratings in the first place. Elementary statistical mistake, and no >response to queries about this. You're certainly correct about not Sonas not validating on a separate test set... But I still think his approach would turn out to improve the predictive ability of the standard ratings for standard outcomes. If it were true that (1) most players' standard, rapid, and blitz ratings were similar, and (2) rapid and blitz results are simply subject to more variance, then a technique like weighted least squares seems appropriate. Of course, assumption (1) could be questionable. Michael
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.