search

Torvaney / regista

An R package for soccer modelling
https://torvaney.github.io/regista/
GNU General Public License v3.0
84 stars 8 forks source link

Add Ranked Probability Score (RPS) metric #7

Open ElSaico opened 6 years ago

ElSaico commented 6 years ago

As per Constantinou and Fenton (2012), this metric suits for win/draw/loss predictions quite well - in fact, it was used by the challenge behind a (still to be published offline) special issue of the Machine Learning journal.

Torvaney commented 6 years ago

Hey! This is a good suggestion; thanks for opening it.

Adding model evaluation metrics generally seems like a really good idea. RPS is, as you point out, particularly nice for evaluating 1X2 predictions. Off the top of my head, this and log loss seems like the key ones. Maybe even a Kelly staked betting-related metric could be worth adding at some point?

It seems like (tidy?) model evaluation could be a good idea for an R package in itself. I should probably see if one already exists. Although even if something like that does already exist, it seems like it'd be a good idea to export in regista's namespace so that models + evaluation come together.