#3 #16 I agree, player ratings are a poor predictor of game outcomes [1].
1: http://en.chessbase.com/post/sonas-overall-review-of-the-fide-rating-system-220813
1: http://en.chessbase.com/post/sonas-overall-review-of-the-fide-rating-system-220813
Very interesting article.
Hence my proposition. Which was meant to point out that ratios between 2 players is often far away from the theorical table.
Using the ratio of won/lost points for established players was a poor way to illustrate the "theorical" odds we had so much difficulties agreeing upon. Still, I maintain it's true for the elo rating.
In order to be rigorous, we should go back to the gaussian table. Don't we all that elo and glicko tend to a similar gaussian law? A simple gaussian table with a step of 50 or 25 pts, would be fine, as ratings change every game.
Hence my proposition. Which was meant to point out that ratios between 2 players is often far away from the theorical table.
Using the ratio of won/lost points for established players was a poor way to illustrate the "theorical" odds we had so much difficulties agreeing upon. Still, I maintain it's true for the elo rating.
In order to be rigorous, we should go back to the gaussian table. Don't we all that elo and glicko tend to a similar gaussian law? A simple gaussian table with a step of 50 or 25 pts, would be fine, as ratings change every game.
#22 Unfortunately using Bayesian inference (based on previous games) presumes that game outcomes are related! And you get dependency issues: game 1 is used to predict every other game, but game N-1 is only used to predict game N.
(Then try explaining the numbers to Beret Guy! http://xkcd.com/1252/ )
What might be fun is an "odds predictor" for the current game considering all preceding games with the same opponent, time control, variant, color, opening, etc.
(Then try explaining the numbers to Beret Guy! http://xkcd.com/1252/ )
What might be fun is an "odds predictor" for the current game considering all preceding games with the same opponent, time control, variant, color, opening, etc.
Support your second amendment : refuse background checks for swimming dogs !
#22 Might be fun but that would be a little overcooked, it'd take a real lot of games between two players to be efficient. My idea was a simple normal law table.
I don't know why you want that bad to introduce Bayesian inference. Maybe i should just bow out and leave you my nerdiest nerd crown for your collection ;-)
#22 Might be fun but that would be a little overcooked, it'd take a real lot of games between two players to be efficient. My idea was a simple normal law table.
I don't know why you want that bad to introduce Bayesian inference. Maybe i should just bow out and leave you my nerdiest nerd crown for your collection ;-)
This topic has been archived and can no longer be replied to.