Choose correct number of knots for GAMs

[Aariq]

Talked with @diazrenata and came up with the following ideas for getting the number of knots right:

• (re-) read ?choose.k and do the thing where you look for patterns in the residuals
• work with even more aggregated data to start with (for faster iteration)
• Follow through to ultimate analysis (map of marginal slopes with non-significant areas greyed out) because if changes in k don't produce qualitative statistical changes, they probably don't matter much
• possibly increase k more in the interaction term than in the main effects (maybe should ask someone if this is a legit thing to do though)

Originally posted by @Aariq in #23 (comment)

[Aariq]

Going to bump this to the next sprint in hopes that someone will share example code for the NCV paper (cct-datascience/organization#2091).

https://fosstodon.org/deck/@millerdl@mathstodon.xyz/112498626465593674

[Aariq]

So not sure what the "neighborhoods" are when NCV is applied to a raster dataset. Is it some area around every pixel? If so, that feels impractical. Going to do some reading about spatial CV:

Brenning, A., 2012. Spatial cross-validation and bootstrap for the assessment of prediction rules in remote sensing: The R package sperrorest, in: 2012 IEEE International Geoscience and Remote Sensing Symposium. Presented at the 2012 IEEE International Geoscience and Remote Sensing Symposium, pp. 5372–5375. https://doi.org/10.1109/IGARSS.2012.6352393

Le Rest, K., Pinaud, D., Monestiez, P., Chadoeuf, J., Bretagnolle, V., 2014. Spatial leave-one-out cross-validation for variable selection in the presence of spatial autocorrelation. Global Ecology and Biogeography 23, 811–820. https://doi.org/10.1111/geb.12161

[Aariq]

Going to close this for now as it has become more of a quest to get NCV working. Follow up in #30 includes "choose correct number of knots"