Author: Dann Corbit
Date: 17:53:05 11/04/04
Go up one level in this thread
On November 04, 2004 at 17:26:55, Uri Blass wrote: >http://www.geocities.com/lyapko/rat30.htm > >rating updated to 26.10 > >Note that Fruit won at least 60 games in a row when it won League M and League L >with 30/30 > >I also suspect that it won before this sequence of 60 games some game in league >N and won later some games in League K This is the way those contests always go. And if you do the Elo calculations yourself, you will get the same funny numbers. In George's tests, the programs start at the very bottom of the weakest engines and play in leagues. The top engines promote (as many as new ones are injected) and the weak ones stay. So if an engine is terribly weak it may stay in the bottom division. If stronger, it will move up. When the strong engines have played nothing but pansies, they will have scores like 100 wins, 1 loss, 5 draws. This causes an inflated Elo, that always drops as the engines climb up the ranks of the tournaments. You can consider every Elo figure unstable until the engine has reached its final resting place. He does not include the Error bars, or you would see that the figures given are not unexpected. You can pull the games and do the Elo calculations yourself. I think that all of his results are not unexpected. If you look at the actual tournaments, you will see in each case the results you imagine will happen.
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.