Author: Pat King
Date: 14:55:59 03/21/02
Go up one level in this thread
On March 20, 2002 at 12:49:54, Miguel A. Ballicora wrote: >On March 20, 2002 at 07:44:18, Sune Fischer wrote: >>On March 20, 2002 at 05:49:56, Gian-Carlo Pascutto wrote: >>>On March 20, 2002 at 05:47:28, stuart taylor wrote: >>>>The worst would have to be about 0 elo. >>> >>>No. ELO is a relative scale. You can have ELO ratings from -inf to +inf >>>if you want to. >>> >>>-- >>>GCP >>Of cause you can't _actually_ reach +-inf, for several reasons. >It depends on how the rating is calculated. In other words, it depends on >the implementation. The idea of a rating is derived from the normal probablility curve. The various implementations of ratings will differ from the "real" rating that reflects the actual probability of victory (to the extent that the normal probablility curve reflects the performance of real chess players). If a chess player really will lose every time (and one could design a program that bad) then his/her/its "real" rating would be -inf. It would take an infinite number of games for the "approximate" rating generated by the various systems to approach the "real" rating. >The classical way is to give a starting rating to a player >and it is updated with each tournament. That is what you are referring. However, >if you do not have a "starting" rating you can certainly reach -inf. Playing the >first 100 games a losing them all gives an estimated rating of -inf. >Anyway, inf is a limit so in any case there no problem to say that elo scale >goes from -inf to inf, whatever the reference is. > >Regards, >Miguel > > >> >>It would take an infinite number of games, which would require an infinite amout >>of time. >>And even the perfect player must be rated below +inf because chess is limited >>yada yada yada random player etc.... Easy to make the perfect loser, hard to make the perfect winner, but such a player's "approximate" rating would approach the "true" rating of +inf, reflecting P(win) = 1.0. >> >>However, it seems a deterministic engine could lose _all_ the games, simply by >>playing another deterministic engine, which just happend to win that particular >>game. >> >>-S. Which is why matches against one opponent, or tournaments agaisnt a very limited number of opponents, is not a valid measure of playing strength. Pat King
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.