# BRMS to INLA bridge

Martin's idea for a student project

Call 2020-12-05: Martin Modrák, Dan Simpson, Aki

- `mgcv` now has (something like) INLA in it.
- Not sure if it has samples
- Dan thinks `inlabru` is more suitable
- Correlated random effects are a problem (for INLA, not sure if `mgcv`)
- INLA has limits for number of hyperparameters (roughly < 15 dimensions)
- Side idea: make posterior_predict faster (for very large prediction tasks)
- Use samples to build brmsfit?
- Alternatively develop specializations of some functions for the INLA object...
- Maybe better to build an mcmclist (or something similar from posterior_draws) - but that wouldn't let us posterior_predict
- Use later inlabru

$$
a + b
$$