Author: Graham Laight
Date: 04:46:28 01/05/00
Go up one level in this thread
On January 05, 2000 at 06:10:07, blass uri wrote: Blass - thankyou for the effort you've put into this good looking explanation. The first thought that springs to my mind is that it seems churlish to argue about inaccuracy in computer ratings, when it appears there are so many "imperfections" in the human ratings! -g >1)I know that the rating system is based on some assumptions that are not based >on games(I think they assume that the ability of players is ditributed normally >with standard deviation of 200 elo and in Israel there is a linear formula that >is only close to this assumption to do it easy for humans to calculate their >next rating). > >I am sure that the assumptions are not correct and the only question is what is >the distance between the assumption and the truth. > >2)You cannot predict the result only by knowing the difference in the elo >because some humans are unstable and have bigger chance to win good players and >bigger chance to lose against weaker players but this is not the only problem. > >3)The rating is also not perfect in predicting the playing strength of humans. >Suppose an 1600 player based on previous topurnaments train some years against >computers without playing in tournaments and after many years of training go >back to tournaments and gets performance of 2600 in 9 games(including draws and >wins against GM's). > >It is logical to assume that he deserves more than 2200 but >the rating system is not going to give this player even 2000. > > >4)It is possible to do the following test for rating systems. > >a)give prediction for every future game between 2 players based on the rating of >players(the prediction can be 0.7 points for the stronger player) > >b)calculate the sum of the squares of the differences between the prediction and >the real result > >The rating system that gives the smallest sum of squares is the best rating >system that humans find. > > >I guess that nobody is going to do it because people are not interested in a >better rating system but in a system that it is easy to calculate their elo and >you cannot earn money from finding a better rating system. > >Uri
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.