Adaptation to account for draws

[Venners09]

Can this be adapted to include draws between opponents? In terms of a football application there can be a home win, away win or a draw

[mbhynes]

Hi @Venners09,

Thanks for your question. It could be; I'm afraid I wasn't looking at games with draws originally, but this is a good extension.

AFAIK the Elo and Glicko systems both treat draws as a 0.5 point to each player. So the necessary change for this is relatively straightforward:

Change List

• the base RatingModel.update() API needs to be updated;

  • probably best to take a flag for a draw in the signature

    def update(self, winner, loser, timestamp, is_draw=False):
    ....

  • The label value passed to evolve_rating() should just be set to 0.5 in this case here and here

• The RatingEstimator.fit() loop here needs to have some additional logic to check for a 0.5-valued floating point label, rather than an assumed binary/integer label, and use this to call rating_model.update(winner, loser, timestamp, is_draw=???)

• The additional draw case should have tests added to test_elo.py here

• The docstrings for the RatingModel.update and RatingEstimator.fit() methods need to be updated to explain the API change, and probably an example should be added to the main documentation as well

• The Elo and Glicko rating models don't need to be changed

• the package version needs to be up'd and uploaded to pypi as per the steps here

So all in all nothing too complicated, but a bit of work. I could do this, but just not sure when.

Would you please lmk if you'd be interested in contributing a PR for this?