Recursive Fix #187 & #174

[AlbertoAlmuinha]

Hi @mdancho84,

I have created a PR to resolve some of the open issues related to recursive models. Basically I have applied a condition that when in the test dataset there are more levels in the id than in the stored train_tail, then this object is filtered with the same levels available in the test dataset.

Anything we discuss.

Regards,

[AlbertoAlmuinha]

Hi @mdancho84 ,

I have added another commit to solve issue #188 . Basically what I have done is instead of detecting specifically the class "_elnet", detect any class that contains the word "net" to send it by recursive or panel recursive. It is very related with the issue #76 solved some time ago.

Perhaps the way to dispatch can be made even more sophisticated, but I think that for the moment it may be a good approximation.

Regards,

[mdancho84]

Awesome!! I appreciate the help. Let me take a look over the next day or so and I'll provide any comments.

[AlbertoAlmuinha]

I just added a new commit to give a solution to the request #160 creating the function drop_modeltime_model.

Regards,

[mdancho84]

I reviewed and am OK. Things we might need to plan for:

• The detect_net() I like this. But one thing I'm thinking is that if a model is not a glmnet style but has a class of "something_net", then it will get selected and might go through the glmnet prediction.

[AlbertoAlmuinha]

I thought the same yesterday, maybe it would be a better idea to look for the parsnip model classes with glmnet engines and encapsulate them under that function so there can be no confusion…

I'll take a look at it later