Hi,
The problem is not wheter to use Elo ratings or not, but rather how to
compute them. The intro of my WHR paper gives an overview of different
possible approaches:
http://www.remi-coulom.fr/WHR/WHR.pdf
The algorithms I compare in my paper all have a major flaw: they
consider the variability of ratings is the same for the whole
population. In practice, the ratings of beginners tend to vary much
faster than the rating of experts. A good rating system must take this
into consideration if it has to be applied to a population that contains
beginners and experts.
If you population has strongly connected groups that are sparsely
connected to each other, then you should avoid incremental rating systems.
If you want academic papers, Mark Glickman's web page has many more:
http://www.glicko.net/index.html
Rémi
On 08/03/2015 02:22 AM, Aguido Davis wrote:
Good morning.
We're looking at replacing the Australian national ranking system, and
the question has come up: how many players and how many recent
games/player are needed for ELO to generate good strength ratings?
(Questions begged: what does a "good" set of ratings even mean? does
it matter if the play graph (edges = games, vertices = players) is
well-connected or quite cliquey? is ELO the last word in rating
algorithms? do humans behave differently from bots when they know
they're being rated?)
Does anybody know of a good academic paper, or ideally, someone's thesis?
My apologies if this is off-topic, but it's an interesting computation
related to go...
Cheers,
Horatio
_______________________________________________
Computer-go mailing list
[email protected]
http://computer-go.org/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://computer-go.org/mailman/listinfo/computer-go
