Author: Peter Fendrich
Date: 10:34:32 07/31/98
Go up one level in this thread
On July 30, 1998 at 21:18:33, Shaun Graham wrote: >On July 30, 1998 at 12:42:52, Peter Fendrich wrote: > >>On July 30, 1998 at 00:48:21, Shaun Graham wrote: >> -- snip -- >This for the most part is true, except for when the ssdf compiles a rating using >different speed processors(especially since one should consider someprograms >benefit more from speed than others). Well, the point is that that no program gets its rating from different hardware combined. Each combination of a program and hardware is regarded as one individual with one rating. This means that a program gets different ratings on differnet hardwares. Of course every possible combination of software and hardware are not tested so the newest pograms gets the latest hardwares. All because of limited resources. Sometimes there is a discussion which program should get the best hardware before the others but that's a completely different issue. -- snip -- >In my calculation i set the base ratings of all opponents at 2400, for the >reason, that generally even the most biased people against computers, accept >that most of the current top of the line programs are AT LEAST 2400. This >causes a slight inacuracy of the rating, but no inaccuracy of the ranking which >is given to the programs. Thus the problem of having people claiming that the >ratings that you are assigning to the programs are inflated. If i played >chessmaster against Fritz and claimed that i should calculate the rating based >on the 2589 or so rating it has, there would be some people claiming my rating >was faulty because they don't believe the 2589 ssdf rating. However if i just >act as if fritz is 2400 a lot of those arguements are squelched. Chessmaster >incidently still performed at 2535 USCF even setting all opponents at 2400 and >this sounds indeed quite reasonable to me. > >Shaun What you should do is an iterative approach. It doesn't matter much if you start with 2400 or say 1800, for all combinations of program and hardware. 1. Compute the ratings by going through all played games. 2. Now keep the computed ratings and do another computation giving new ratings. 3. Continue in this way until he ratings are stabilised and not changing a lot between the computations. Easy! Now to the hard part: It is very important to know that the ELO formula doesn't compute fixed ratings. It computes the RAITNG DIFFERENCES between programs. If you know (or think you know the "real rating" for one or more of the programs, you should adjust the computed ratings in accordance to that. So if you increase/decrease their ratings with 100 points then you should increase/decrease the ratings of all the programs with 100 points. The best way is probably to use ratings from human tournaments for some of the programs. //Peter
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.