Aki Vehtari comments

Results 437 comments of


                                            Aki Vehtari

(loo_)(s)crps could ask for only one argument with predictions

@bnicenboim any update on this?

Use JuliaCall to Speed Up PSIS

> I forget why we used x[floor(n/4 + .5)]. @avehtari do you remember if we chose to use that instead of using quantile() for a particular reason? It's because I...

Use JuliaCall to Speed Up PSIS

> but I think this only requires changing one line in the documentation It's implemented and documented in more than one place, so there is more work. > and it...

loo_predictive_metric and loo_crps could accept psis objects to reduce computation

I prefer the behavior in different functions to be consistent, so if E_loo does that then it would be good to have in other functions, too

Threshold for R-hat (1.01 or 1.05)

> I think that in our latest Rhat paper we also moved to 1.05 as we realized 1.01 was too strict. No we didn't. Let me get back from my...

Allow CV in all posterior predictive checks

I don't think so

Support for handling censored data in PPC plots via imputation

Just added better figures to https://github.com/stan-dev/bayesplot/issues/319 The imputation part requires sampling from a truncated posterior predictive distribution. In many cases rejection sampling is likely to be fast enough, and simplest...

Fixes and reminders

I noticed this open PR with several unresolved conversations. @davkoh ?

Hessian NaN with ordered_logistic_lpmf

For the same case, Hessian works in BridgeStan. What is the difference to CmdStanR?

Hessian NaN with ordered_logistic_lpmf

Thanks. I missed `BRIDGESTAN_AD_HESSIAN=true` and the whole page with "Optional: Customizing BridgeStan" as I installed BridgeStan using the R package.