This repository contains code for ranking according to Mark E. Glickmans "Glicko2" ranking system. See
There are are couple of exports, which are the main interface to the code base. Suppose we have
Rating = 1500,
RatingDeviation = 200,
Sigma = 0.06
Opponents = [{1400, 30, 1},
{1550, 100, 0},
{1700, 300, 0}].
Where each opponent is given as a triple {R, RD, S}
of respectively
the rating of the opponent, the rating deviation and the score. A
score of 1
means we won and a score of 0
means we lost. A score of
0.5
is a draw, but I am not using that in the code.
Calling
glicko2:rate(Rating, RatingDeviation, Sigma, Opponents).
Returns a triple {R1, RD1, Sigma1}
of the players new rating, rating
deviation and sigma/volatility value.
Alternatively, you can generate a configuration configuration(350, 0.06, 0.5)
which defines, respectively, the initial rating deviation,
the standard volatility base and the value of tau in Glicko2. The
tau value says something about how much volatility means as a
factor. I suggest tuning these for optimal values.
To use your configuration, call:
Conf = glicko2:configuration(360, 0.06, 0.5),
glicko2:rate(R, RD, Sigma, Opponents, Conf).
Another export is phi_star/2
which is an internal step in the
Glicko2 scheme. This step is useful for tuning, hence it is exported.
MIT