Arimaa Forum (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
Arimaa >> Off Topic Discussion >> Rating Uncertainty
(Message started by: mistre on Feb 4th, 2014, 3:10pm)

Title: Rating Uncertainty
Post by mistre on Feb 4th, 2014, 3:10pm
I combed the forum for an answer to this question, but couldn't find it, so I will ask it here.

Why is the RU minimum set to 30?  Is this an arbitrary value set by Omar or is an intrinsic value required by the rating system?

The reason I ask is that if you play very often, an RU of 30 seems a bit high as within the course of 10 games or so, you can still gain or lose 100 rating points easily if you get on a long winning or losing streak.

I know the gameroom ratings have many shortcomings, but reducing the minimum RU would seem to stabilize things a little bit.

Thoughts?

Title: Re: Rating Uncertainty
Post by aaaa on Feb 4th, 2014, 5:33pm
Setting a minimum rating uncertainty is done to prevent ratings from becoming too rigid, which has been a common complaint wherever rating systems have been in use. I doubt that there is something special about 30, but I note that this value was suggested as the minimum rating deviation in the Glicko paper.

I had this idea of the flawed gameroom rating system being replaced by the (computationally similar) Glicko and to have their parameters be empirically optimized based on applying the WHR system over the database of games.

Title: Re: Rating Uncertainty
Post by Janzert on Feb 4th, 2014, 8:18pm
On a sorta semi-related note I ran across another rating system recently. Purporting to outperform Glicko while being easier to implement and also having faster runtimes.

A Bayesian Approximation Method for Online Ranking (http://jmlr.org/papers/volume12/weng11a/weng11a.pdf)

Janzert

Title: Re: Rating Uncertainty
Post by rbarreira on Feb 5th, 2014, 4:41am

on 02/04/14 at 20:18:03, Janzert wrote:
On a sorta semi-related note I ran across another rating system recently. Purporting to outperform Glicko while being easier to implement and also having faster runtimes.

A Bayesian Approximation Method for Online Ranking (http://jmlr.org/papers/volume12/weng11a/weng11a.pdf)

Janzert


I read the abstract, it seems pretty interesting as it also supports team games. Will definitely read.



Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1!
YaBB © 2000-2003. All Rights Reserved.