Model never converges when events with multiples winners are observed

[liampwll]

I'm attempting to apply kickscore to a game where each team has 5 players, but I've found that when I try to add a single event with multiple winners and losers the model never converges. When I have a separate event for each winner winning against each loser the model does usually converge. Is there something I should be doing differently in this scenario?

[lucasmaystre]

Hi @liampwll , thanks for raising this issue, this is a very good observation. I should put this in the documentation eventually.

You can usually fix this by decreasing the learning rate. To do this, when calling the fit() function, you can set the keyword argument lr with a value between 0.0 and 1.0 (default).

Could you try with lr=0.5 or lr=0.2?

Best, Lucas

[liampwll]

Thanks, 0.2 works well with my dataset and kernels although 0.5 does not.

[lucasmaystre]

Great. I think lr = (1 / n_players) is a good rule of thumb.

When you decrease the learning rate, you might also want to decrease the convergence threshold (tolparameter of the fit function), maybe to 1e-4 or 1e-5.