Author: Robert Hyatt
Date: 11:21:04 03/18/99
Go up one level in this thread
On March 18, 1999 at 10:23:23, KarinsDad wrote: >On March 17, 1999 at 22:21:09, Robert Hyatt wrote: > >>On March 17, 1999 at 17:43:21, KarinsDad wrote: >> >[snip] >>>Robert, >>> >>>Are you saying that the servers do not decrease the ratings changes (i.e. halve >>>them just due to the fact that a higher rated player is playing) for the higher >>>rated players? >>> >>>If so, this is strange and disturbing. >>> >>>KarinsDad :| >> >> >>No... the Elo system has a 'max change' paramater K, that seems to be set to >>32 for most places. And that works well in that the most your rating can change >>with any one game is 32 points. But it is proportional to the rating difference >>between you and your opponent. >> >>Other servers use a smaller K (chess.net for example) as suggested by the >>'glickman' proposal for rating calculation. If you reduce K, then you reduce >>the maximum rating change delta. and you limit rating 'swings'... > >Robert, > >As I'm sure you are aware, the USCF modifies the K for higher rated players. > >For players 2100-2399, the change is multiplied by 0.75. >For players 2400+ , the change is multiplied by 0.5. > >It would seem logical for the chess servers to do something similar since you do >want some more major level of stability at the higher ratings than you need at >the lower ratings. > >Do any of the servers publish their K values and potential modifications to it? > >Thanks, > >KarinsDad :) I have never asked on ICC, but two 2900 programs (exactly 2900) will end up 32 points apart after one game that is not a draw... So it seems to 'scaling' near the top occurs...
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.