Author: Robert Hyatt
Date: 05:55:04 05/18/00
Go up one level in this thread
On May 17, 2000 at 22:17:19, Michael Fuhrmann wrote: >Do you have a hunch which is true? > Before you can ask (or answer) this question, you have to stop and think a bit. Let's take two angles: (1) the Elo rating improvement is linear. This seems impossible. Because the Elo formula is based on the "normal bell-curve shape." If the program keeps getting better and better, the rating difference between it and a human with a fixed rating slows down. IE if I win 3 of 4, I am about 200 points better. If I win 15 of 16, I am about 400 points better. The formula isn't linear, so a "fixed Elo improvement" doesn't make sense... (2) we take a weak program vs a GM. It is likely that for a long time, every doubling will produce at least a 60 point rating climb. If you measure the gain before the program catches the GM, you will see big jumps. But once the program catches the GM, then (1) takes over. I believe that additional depth _always_ helps. But it is difficult to measure a precise term. Everyone wants a simple linear function to predict this rating gain. I don't believe that is reasonable or possible.
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.