Questions about extensibility

[vnmabus]

Hi, I was looking at this project as a possible candidate to implement generalized functional linear models in the future, in combination with my own functional data package scikit-fda. The objective would be to implement our own Term classes, that accept functional data. In order for this to be feasible, I have a few questions to ask.

• Is this project actively maintained? The last code commit is very old, and there are a lot of unsolved issues and PRs. Before coding against this API I want to know that if we found bugs or improvements we could fix them and they will be reviewed. We could always fork the project as a last option, but I would prefer not to.
• Does Gam accept Pandas dataframes with ExtensionArray columns as an input? Our input data are functions, not numpy arrays, so this is a must. My current understanding is that if we provide a Term that understand our input, Gam will work with Pandas dataframes.
• In the future we will probably need to implement also function on function regression. I think that it can be done in a wrapper from our side using discretized functions, but maybe it is possible to extend PyGam terms to return arbitrary Python objects (that support addition and link functions using numpy ufuncs).