smerdov / pyglicko2 Goto Github PK

Hi, just tweaked to use the new version of glicko. Appears to be working 
(tested against the paper's test). You may want to try and merge in: 
http://code.google.com/r/lmeyerov-glicko2update/source/checkout .

Hi,

The player class has the rating stored in the Glicko2 scale. This forces to 
calculate the Glicko2 ratings when passing parameters to the update_player 
function as in your example.

In my opinion it would make a lot more sense to store the rating in 
standard format and convert it on the fly before and after updating the 
player's rating, and changing the definition.

Definition:

update_player(self, p_rating_list, p_RD_list, outcome_list)

Notice that "outcome_list" is not a list of "bools" as indicated in your 
documentation but a list of "floats", since matches can be drawn and 
therefore the outcome would be: 0.5.

At the beginning of update_player

self.rating = (self.rating - 1500) / 173.7178
self.rd = self.rd / 173.7178

rating_list = [ ( x - 1500) / 173.7178 for x in p_rating_list]
RD_list = [ x / 173.7178 for x in p_RD_list]

At the end of update_player

self.rating = self.rating * 173.7178 + 1500
self.rd = self.rd * 173.7178

The definition of __init__( self, rating = 0, rd = (200/173.7178), vol = 
0.06) would be changed to:

__init__( self, rating = 1500, rd = 350, vol = 0.06)

Notice the rd = 350, as you have 200 and the the Glicko2 paper proposes 350 
as the default value.

In your example, we would change the following line:

Ryan.update_player([(x - 1500) / 173.7178 for x in [1400, 1550, 1700]],
        [x / 173.7178 for x in [30, 100, 300]], [1, 0, 0])

To be the following:

Ryan.update_player([1400, 1550, 1700]], [30, 100, 300], [1, 0, 0])

This is a lot more readable for anyone making use of the code and later 
using "Player.rating" makes a lot more sense.

Best regards