Another way of looking at the ecf grading method(?)

General discussions about ratings.
Brian Valentine
Posts: 574
Joined: Fri Apr 03, 2009 1:30 pm

Re: Another way of looking at the ecf grading method(?)

Post by Brian Valentine » Tue Aug 24, 2010 9:53 pm

Both these sources are of some interest. However I'm not too sure that players would be happy with a black box method (the iteration) to calculate gradings. I can just envisage the thread on this forum if the Trueskills method was used. It is impractical to audit and would still be adjusting ones skill level in 2005 (say) based on the latest game this year (played by two other players). On the other hand I like the idea that I have a probability of being the world's strongest player :D

I used something very similar to the Brent approach many years ago for a club handicapping event. The obscurity of the calculations was a big turn off for members, despite the increased statistical reliability. Something simpler replaced it the next season.

Paul McKeown
Posts: 3732
Joined: Thu Apr 12, 2007 3:01 pm
Location: Hayes (Middx)

Re: Another way of looking at the ecf grading method(?)

Post by Paul McKeown » Fri Aug 27, 2010 12:17 pm

Adam,

Very interesting link; very interesting conclusion that Bayesian methods converge much faster. Couple of asides: don't fancy white's response to Alekhine's Defence (2. f3!) and did you notice how the Prof. spoke of an Elo rating of 120? Perhaps he meant ECF grade?


I agree with Brian that a blackbox approach would initially be viewed with some suspicion. However, I have experience of blackbox systems in real life. The KNSB (Dutch) national rating system is essentially blackbox, but people like it, especially as they get to be good at anticipating the outcome with rules of thumb, anyway.

Regards,
Paul McKeown

Post Reply