Aki Vehtari

Could you 1) give a pointer to an example in `brms`, 2) include an example code how you would use it in `loo`, 2) and tell what is the case...

1. Thanks for the examples. I've never used `future`, so these were helpful to understand. 2. If `model` is `brms` model then this might be a `brms` package issue instead...

> To be clear, I would be able to run loo on a subsample of folds and still get a good approximation, but this is not yet implemented, right? Probably...

> Should we avoid NAs and return always Inf and NaN here? I would prefer khat=Inf when the failure is due to extremely long tail. At least in the cases...

Thinking for a moment, I think normalized densities would make more sense, and if someone would like to use normalized quadrature weights it's easier to obtain them from normalized densities...

I have tested. When using sample sizes that are typical gpd_fit use and when tail thicknesses are in the interesting region (

Thanks to a question by @sethaxen, I realized that what is above named as densities are not yet densities. Originally, the algorithm computes quadrature weights and for those most natural...

Here is an illustration  The yellow contours present the likelihood, the red dashed contours present the posterior, blue line is the profile line, the blue dots present the quadrature...

> I see; does that hold even though the samples aren't IID? That would also tend to make the posterior too narrow. True. I don't think this is an issue...

Sorry, this has been left hanging for such a long time. After considering, I think we shouldn't change the existing output variable names (ie k -> k_hat, sigma -> sigma_hat),...