Author: José de Jesús García Ruvalcaba
Date: 09:21:08 02/25/99
Go up one level in this thread
On February 24, 1999 at 19:29:02, Dann Corbit wrote: >On February 24, 1999 at 19:20:28, David Eppstein wrote: > >>On February 24, 1999 at 18:04:34, Vincent Diepeveen wrote: >>>On February 24, 1999 at 17:08:41, Dann Corbit wrote: >>>>Another interesting prize-winning category might be ELO per dollar. This would >>>>probably be very interesting to the end-consumer. >>> >>>Better is prizewinning category for elo per node. >> >>It makes no sense mathematically to divide Elo by anything, >>even ignoring the question of whether you can define a "node". >> >>The zero-point of the various Elo systems (USCF, FIDE, SSDF, etc) >>is completely arbitrary -- you would get essentially the same system if you >>subtracted say 1500 from everyone's number, which would change your proposed >>ratios dramatically. The only thing it makes sense to do with Elo scores is to >>look at the difference between two of them, which should be (some function of) >>the expected outcome of games between the two players. >> >>So, I have to say, elo per anything is a really stupid basis for awarding any >>kind of prize. Especially, since most of the programs in these competitions are >>going to have an Elo within 100 or so of each other, what you are really >>awarding is "minimum price" or "minimum number of nodes". We already know how >>to do that without playing any tournament games. >I disagree. While ELO per node will simply award the prize to slow searchers >and penalize the fast searchers, this sort of measure (ELO/Dollar) will allow >people who have to come with wimpy machines a shot at something. IIRC, last >year some people were competing against 767MHz alphas using K200's. Seem fair >to you? > >It would be a measure of some sort for the price/performance mark. And it is >not strictly a measure of the price of the machine. If that were true, then two >different programs would perform identically on identical hardware. This is >clearly not the case, as a glance at any extended contest or the SSDF results >will show you. Hi Dann, it seems you are missing David's point. Chess ratings do not have a meaning by themselves, what is meaningful is the difference between ratings. If you take a rating list and add any arbitrary number to all the ratings, the new list has the same meaning than the original one, as the differences remain the same. But if you divide those ratings by price or nodes or whatever, then you get very different results.
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.