Author: GuyHaworth
Date: 02:49:01 06/20/02
Go up one level in this thread
Reviewing (?), based on four answers then, we have:
1) the ELO system is intrinsically ok 'within the closed community of players
that it rates' as long as there is 'free mixing' within the pool.
2) It is agreed that there are more computer-computer games than human
Therefore, the inflation effect is worse in the SSDF list.
[ Therefore, the SSDF list needs adjusting downwards if anything. ]
3) Super-tournaments are creating a problem on the 'human' side.
They are creating a pool within a pool, or rather two pools instead of one
4) Match-ratings and tournament-ratings can be wildly 'out' - too few games.
5) Although there are better rating schemes (Glick*?), it would be better not
to rate computers by scheme X and humans by scheme ELO. There is more chance of
getting two ELO schemes alongside each other, than scheme X and scheme ELO.
6) I'd be a bit surprised if nothing could be learned from a cross-calibration
of a set of humans against a set of computers. Could one not say:
"Suppose the computer ratings were 'as_is minus N'. What is the value of N which
makes the results most likely. What is the standard-deviation on N?
I think that k*k times the number of calibration games would reduce the
standard-deviation of N by k.
7) there is an old human-computer SSDF list going back to 1992 ... but there is
a much more up to date and maintained list - excellent.
Thank you all
g
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.