Author: Dann Corbit
Date: 18:07:54 08/13/99
Go up one level in this thread
On August 13, 1999 at 20:56:33, Terry Ripple wrote: [snip] >Hi once again James, > Lets assume that SapphireII just made Master strength at "2200". Now lets >assume what you say is accurate in regard to your rating fluctuating from 1700 >to 1800 USCF. By these figures you should win between 5.3% to 9.1% of your games >against SapphireII. Now just say that it is only 2050 USCF, that would mean that >you will win between 11.8% to 19.2% against SapphireII. My Expert friend won >12.5 out of 46 games which gives him a win factor of 27.2%. If these figures are >correct that are stated by the USCF,that would make my friend's win % show that >his opponent was 170 USCF points stronger than him (2065+170= 2235 USCF Rating >for SapphireII).This of course is only accurate if my friend is still rated >2065, but i can say that he is still active in our club but didn't play any >recent tournaments in the past year. > I hope that the future brings us the answers to these questions of man >vrs.-machine and how strong the Monsters really are. Maybe the Rebel matches >will help us understand a little more about our favorite beasts! You cannot predict from a small set of matches the ELO of any single opponent. You need a large number of games against a large number of opponents. Therefore "Opponent A did better than predicted against X" is not a proof of ELO one way or the other. Neither is "Opponent A did worse than predicted against X" They both lend a small data point one way or the other but it is not verified until a broad spectrum of tests have been completed. Your final point is well put. We really don't know much about human/machine competition because we have so little data. Experiments like the Rebel GM matches are groundbreaking in this regard.
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.