|
|
|
Topic: Opponent's rating used to calculate rating adjustment
| |
|
|
Author
| Message |
|---|
|
|
Still, predictions are quite complicated to make when a player's performance (of which his/her rating is also a sort of measurement) oscillates sometimes greatly during a CC game. I believe a more precise rating adjustment (let alone the concept of fairness) would be to use, in the computation, an average of the opponent's rating between the start end the end of a game. Hence, a rough (rating_start + rating_end)/2 or a more accurate interpolation based on the number of games played. Not too hard to compute. 
|
|
|
The rating scheme used here is essentially the one that the US Chess Federation uses for rating correspondence games and the USCF obviously feels that using the end-of-game rating is the best way to go. Having said that, they are geared toward old-fashioned moves-on-a-postcard correspondence chess so maybe people in general aren't playing so many simultaneous games there. So it's possible that the simple solution that works for the USCF isn't really appropriate here.
The classical argument would be that the rating at the end of the game contains more information (more of the player's history is encoded in it) so that's the one we should use. I think that the actual strength of a player isn't likely to change all that much over the course of a game so the rating at the end of the game should be a reasonable reflection of their strength throughout it.
But that's not to dismiss your suggestions entirely. It's certainly an interesting problem to work out what is the best way of rating correspondence games
|
|
|
The root cause of the problem is that players here are loosing many point in a very short time period by expired-time-losses. These players often recover afterwards. Assume a player with a normal rating of 2000, diving through a wave of 2000 (t1) -> 1500 (t2) -> 2000 (t3). Taking the players rating from the end of the game will be unfair for players that start a game at t1 and end up at t2, taking the rating from the beginning of the game would be unfair for those opponents that starts playing at t2 and finishes at t3. All average calculations will be "half as unfair", but then for both kinds of opponents.
The only thing that might help out would be to avoid this rating bumps at all, but this would definitely need a lot of new code and/or administration/correspondence/research.
As long as we have a "Bergman-Rating-Enabled"-System, I think we have bigger problems that would need less work to fix in place. But as I can see the only relevance of the rating is that it enables one to enter certain tournaments. Facing sometimes a certain kind of unfairness in some games doesn't makes a differences on the long term for most of us. And players that have established their rating entirely on this "bugs", must ask themselves about the meaningfulness. They are easy to detect if someone like to filter them out.
cheers
|
|
|
Thomas_A_Anderson wrote:
> But as I can see the only relevance of the rating is that it enables one to enter
> certain tournaments.
More generally, it allows one to choose opponents of roughly one's own strength, either by entering rating-restricted tournaments, or by posting or accepting challenges or open games.
Dave.
|
|
|
Sometimes things need to be done in the easiest way possible when there are no good solutions however you do it.
Take my yearly tournament championship list. I use the tourney finish date to group them into years. The start date makes more sense but has too many problems.
Taking the rating at the start (when the game was agreed upon) is more logical but provides little benefit as Thomas points out, and more complexity.
The rating we use is quite volatile and contains very little history. At best one only knows that a player that is rated at 2000 was between 1680 and 2320 ten games back. Much better to have something like a 100 Max number that averages your 100 best ratings. It would require minimal overhead using just a single stored number as a rolling average after the first 100 games.
Personally I would find it even better if such a number would be used in the tournament qualification so that elite players cannot enter lower tournaments regardless of what they do with their ratings (unintentional or otherwise).
|
| Previous 1 2 3 Next |
|
|
|
