Feature request: support for sdmTMB

[jebyrnes]

sdmTMB is a really interesting package for spatiotemporal modeling using TMB as the underlying engine. Some modelbased functions already work, some do not. Would be great to have it incorporated! Especially if there was a smooth way to pull space and time in ultimately for the by statements. Note, it already works well with ggeffects and its vignettes use ggeffects in a few places.

[strengejacke]

insight::get_predicted() (and hence, modelbased::estimate_relation() or modelbased::estimate_link() etc.) should work. Maybe it's not supported by marginaleffects, and thus estimate_means() doesn't work.

I just saw that re_form and re_form_iid are always set to NA. Maybe it's better to allow different options here?

[jebyrnes]

I think the part that is tricksy is the spatial component - in addition to most effects, there is an X and Y component from the spatial random field. Similarly, I'm noticing estimate_relation() also does not seem to work for random effects. include_random = TRUE does not seem to work. But the spatial or temporal varying fields are likely the most interesting thing to work with here and could be very cool if folded in!

[strengejacke]

Which arguments for predict() are relevant here, and which related ?options