Author: Ricardo Gibert
Date: 23:34:05 06/15/01
Go up one level in this thread
On June 15, 2001 at 19:10:07, Dann Corbit wrote: >On June 15, 2001 at 18:41:00, Gian-Carlo Pascutto wrote: > >>On June 15, 2001 at 18:32:18, Ricardo Gibert wrote: >> >>>Why ressurrect this thread? Are you trolling? >> >>Because the issue at hand was never resolved. >> >>They said sorry-sorry, but the thread itself died >>out because of the name-calling. >> >>>Hopefully, your post will be ignored by the parties involved. >> >>If they are unable to talk about this in a mature manner, >>that's whay they should do. >> >>That's why I chose the subject as it is. I'd like to discuss >>this further, but I realize it may not be possible. > >ELO is a very simple system. It measures exactly this: >"On a broad average, what is the probability of points scored between two >opponents in a tested pool of talent?" This absolutely incorrect! ELO measures the *performance* of a player within a *pool* of players. It is particularly *not* intended to predict performance between *individuals*. That is a mistake. It is not even intended to predict the performance of an individual within a pool of players, though that may be a plausible application. > >If the opponents are from different pools, the results are suspect. >If the opponents played under different conditions, the results are suspect. >If the number of trials is small, the results are suspect. > >So exactly what is it useful for? > >Imagine a USCF tournament. Two players have played hundreds of games against >various opponents. If they should play against each other, what is the most >likely outcome? [NOT to be confused with the actual outcome]. What >classification should a player be in? Things like that. > >Imagine a SSDF test run. A program is tried against two dozen other programs, >which have each played hundreds of games against each other. Many trials are >run. We can now predict (with measured certainty) what will happen if we match >this new program against another program from this pool of known strength, under >exactly the same conditions of hardware, time control, etc. > >It does not answer this question: >What happens if a USCF player plays a game against an SSDF program? > >It does not answer this question: >What happens if Paul Morphy played against Garry Kasparov? > >It does not answer this question: >Why will people never get the hang of the ELO system? >;-) > >Of course, it was never designed to address any of those.
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.