API question: Bind model spec info to data container or not about formulaic HOT 3 CLOSED

The ModelMatrix type allows you to proxy any type, and yet also offer a consistent API to downstream users. Having the model spec explicitly attached to the model matrix allows you to always be able to produce a new model matrix like the one you have. This seems much more elegant to me.

Re: not appearing in dir, good point. I'll fix that.

from formulaic.

Why not

y, X, model_spec = model_matrix("y ~ a + b + a:b", df)

This is also clean and avoids patching data containers.
If one attaches metadata like a model spec to a pandas dataframe, then this is most likely lost when performing operations on this dataframe (like copy).

I know that patsy does it similarly. It might be a matter of personal taste, but I would refrain from doing so. In my opinion, it would make things simpler and cleaner.

from formulaic.

Thanks for sharing your thoughts here!

I actually really like the existing pattern, and since it is just a stylistic/artistic choice, I'm going to use my prerogative as maintainer to keep with the status quo. However, the fact that it is confusing is something I'll try to do something about :).

from formulaic.