S3 Method predict._elnet seems to take priority over predict.model_fit and registered predict modules for glmnet based models

[chillerb]

The problem

When registering a custom model which also returns a modified glmnet class object, it took me a while to notice that parsnip also exports the S3 Method predict._elnet. Since _elnet is the first listed class object in the list of class attributes, the method predict._elnet exported by parsnip seems to take priority over parsnip's own predict.model_fit, therefore never actually calling my registered predict module, but parsnips glmnet routine. I know that this must be some super rare edge case, but it just took me an hour to figure this out and it seems like a developer pitfall to be aware of.

class(my_model)
[1] "_elnet"    "model_fit"

[EmilHvitfeldt]

Hello @chillerb 👋

May I ask what specifically you want to do with the glmnet object that you can't do right now?

[chillerb]

Hi @EmilHvitfeldt
The glmnet object is returned by the fit-module of my custom parsnip model (which does some tricks with input scaling before passing the data down to glmnet). So after fitting the model, I called predict on it, expecting parsnip to call my registered predict module (which again does some rescaling). However, parsnip's predict._elnet took priority instead. Sure, there are some simple work arounds for that, like wrapping the glmnet object in a list, so that the parsnip model would not have the _elnet class. It was just unexpected, that parsnip disregarded the registered fit module in that specific case.