Open martinmodrak opened 9 months ago
I was looking into this some months ago and ended up uncovering an unfortunate interaction between cmdstanr
VI output and the way that brms
converts fits into rstan
format regardless of backend: https://github.com/paul-buerkner/brms/issues/1560
Since the error happens exclusively during summary printing of the $fit
object, I don't think this should impact anything that SBC itself cares about, but just a heads up in case this comes up while developing the code.
The current implementation of brms backends relies only on rename_pars
- then it uses the inner fit$fit
object to get draws. So as long as rename_pars
works for the converted object, we are goood.
Currently not supported, but should not be that hard to pipe through to the underlying Stan backend.