Adding GAMData struct

[yahrMason]

Working on refactoring the model functions to use struct arguments.

I reviewed the text and got a much better understanding of how to build the penalty matrix for handling multiple smoothing terms. I now have a working example off of the trees data set that converges to a solution for the smoothing penalties.

I will look to extend this to handle generalized additive models with different link functions.

[yahrMason]

Still need to clean up the functions a bit by adding doc strings and perhaps better typing. Also need to add "smooth" indicators to the struct so that it can handle both smooth and non-smooth terms.

[hendersontrent]

Thanks for this! Just tested it all -- looking real nice! Turns out that having a copy of the textbook at hand is useful! Super keen to see where you get this to :) I'm happy to help with docstrings and things too!

[yahrMason]

Thanks! Your previous comments about GCV not being appropriate for models where the dependent variable follows a non-normal distribution were correct. He outlines two additional approaches in his text that handle the generalized cases. One is the random effects approach, which I think MixedModels.jl could be a good candidate for. The other is the Penalized Iteratively Re-weighted Least Squares (PIRLS) algorithm. That looks pretty straight forward so will try my hand at implementing that.

[hendersontrent]

Yep, I am aware of the random effects/spline stuff (see https://www.tjmahr.com/random-effects-penalized-splines-same-thing/ for a great overview). If you feel confident trying PIRLS then I think we start there. May as well get something working with one less dependency for now!

[hendersontrent]

Sorry, I would contribute more but I am a bit slammed across PhD and work at the moment. Hopefully can help out more substantively soon!