Aki Vehtari

You can use the R code as a reference implementation https://github.com/stan-dev/loo/blob/master/R/gpdfit.R It follows Zhang and Stephens , but uses numerically more stable computations plus weakly informative prior.

For Neff see the variance formula which is last equation on page 4 in https://arxiv.org/abs/1507.02646 and for the reference code see https://github.com/stan-dev/rstan/blob/develop/rstan/rstan/R/stanmodel-class.R#L288 lines 288-304

> - no Jacobian: MAP for the constrained model > - Jacobian: MAP for the unconstrained model Good summary, which would good to have in the case study, too. >...

> I can see why some ML people would expect this behaviour, but I still think it's the wrong behaviour for most "normal" (i.e., non-ML) users so we want to...

> It would be interesting to compare the differences from Andrew's perspective of taking MAP plus Hessian to be a kind of approximate Bayes. With the new distributional approximation diagnostics...

The shuffling of the mcmc results by default is evil, and as discussed elsewhere there are many who think it should be removed, which would solve also this problem.

See specifically this comment https://github.com/stan-dev/stan/pull/2618#issuecomment-473595318 which contains the justification for the changes I made and why certain things are not possible.

> We can do this, but we need to do it today / tomorrow to make it into the next rstanarm release That might be too hasty. It's enough that...

> How can the baseline have zero intercept if the only thing in the model is the intercept? I think if the user fits a model without an intercept the...

> Are you still looking for a hand on this? Sure! > Do you want classification accuracy for binomial? Ben asked, Aki wasn't sure. I'm fine adding it, and I...