Using brms

[seabbs]

Have you considered either using brms directly or extending brms i.e. like https://epidist.epinowcast.org/ rather than writing a custom model? That would get you a lot more flexibility for free. I can't see a clear reason why this wouldn't work but I may be missing something

[dchodge]

Heyy, so what is happening here in the stan code : https://github.com/seroanalytics/seroCOP/blob/main/inst/stan/logistic_model.stan is fitting a generalised logistic curve with hierarchical effects on some parameters (I'm slowly adding complexity, which is why stan feels natural). I've used brms for regression models but not sure the syntax for fitting curves like this (I'll ask Claude later maybe).

Can you briefly explain how brms gives more flexibility than writing stan code ? stan code obviously is more flexible, but perhaps using brms is easier in regards to adding optional features (like hierarchical effects) instead of writing flags for this stuff in stan (which is a bit of faff)?

[seabbs]

I've used brms for regression models but not sure the syntax for fitting curves like this (I'll ask Claude later maybe).

Yes I think it can do this directly but if not indirectly using its support for insertion. See here for the kind of reasonably complicated brms models you can construct: https://github.com/epiforecasts/covid19.sgene.utla.rt/blob/main/R/variant_rt.r and epidist for an example of stan code insertion

is easier in regards to adding optional features (like hierarchical effects) instead of writing flags for this stuff in stan (which is a bit of faff)?

Yes exactly and avoids the issue of i.e. getting it wrong. There is also the ecosystem integration i.e. tidybayes, loo etc which you then get for free. The epidist docs so some of this which might answer some questions. It also means the users have a standard interface that is well supported for things like priors vs having to relearn that across each package.

[dchodge]

OK cool, will look into this, thanks Sam!

[seabbs]

Note its more a for interest/for the future comment as you reasonably may not want to reimplement.