Author: Andrei P
Date: 16:11:56 06/16/04
Go up one level in this thread
On June 16, 2004 at 18:59:55, Mike Byrne wrote: >On June 16, 2004 at 17:27:17, Andrei P wrote: > >>has anybody tried to astimate top engine ratings vs humans by discarding the >>drawn games from the analysis and looking only at the decisive games? > > >I don't understand the objective why one would want to do that. The way ELO >ratings were defined by Dr. Elo, you would always factor drawn games in. His >methodology does not allow for discarding drawn games. Therfore, ratings >without drawn games would be without the same statistical significance as >developed under his system. I am not a statistician, so I am mainly speculating. it seems to me that taking both draws and decisive games is good when the players are closely matched. when the rating spread is large, than draws favor the lower-rated opponent. so it seems more fare to use the decisive games, especially in computer-human games where on one hand you see games where computer easily crushes a stong player, and, on the other hand, games where computer just shuffles his king hopelessly and draws or loses in a pityfull was. because, there is a lot of uncertainty how closesly humans and computer are matched today, so it seems better to use the decisive games only (assuming my guess about more fairness in this case is valid)
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.