Implement __copy__ and __deepcopy__ model methods

[ricardoV94]

Description

A couple of users on discourse got some nasty bugs when trying to do copy.deepcopy(model). Whatever the default is doing, it ain't great.

We have a clone_model in model/fgraph that we can use for both copy and deepcopy!

It should work in all cases except for the known issue with GP variables #6883

[Dekermanjian]

Hey @ricardoV94 I'd like to try implementing this. I just need some guidance. I don't quite understand what a shallow copy of a PYMC Model would entail. From what I understand, you can't mutate nested structures in a PYMC Model after you define the model. I think what clone_model() does is a deepcopy where making changes to nested structures like random variables that get turned into fixed values by using pm.do() won't be propagated to the clone.

[ricardoV94]

We don't need a shallow copy, both methods can do a deep copy

pm.do itself also does a deepcopy before doing its business

[Dekermanjian]

Oh okay, I see what you are saying. So both __copy__ and __deepcopy__ should return the same result that is equivalent to the clone_model() method. Did I understand that correctly?

[ricardoV94]

Yup

[Dekermanjian]

Okay, cool. Thanks!