Aki Vehtari

I didn't understand the last message. I thougt we can two different function signatures, one with scalar and one with vector. Also I don't well enough the difference between std::vector...

Ok, then do it in a way which is easy for you. Maybe things are easier with stanc3?

> The idea was to have optimized Cholesky factorizations and all that good stuff for everything. I think we should reconsider that plan. When we discussed that, we forgot how...

Instead of ess_half1 and ess_half2 separately, I would return for this diagnostic ess_full/mean(ess_half1, ess_half2) The name of the diagnostic could be something like _empirical convergence rate_ After writing that name,...

The probability integral transformation for discrete data is discussed in Czado, Claudia, Tilmann Gneiting, and Leonhard Held (2009). “Predictive Model Assessment for Count Data”. In: Biometrics 65.4, pp. 1254–1261. Direct...

That reminded me that I should add LOO-PIT to CV-FAQ (but run out of time today, and try to do that tomorrow)

Which version of projpred you are using?

Thanks for working on this. Some comments Butterfly histogram is useful also for PITs and MCMC rank plots (see https://github.com/stan-dev/bayesplot/pull/179), so this should be modularized so that there would be...

> If there are 1000s of parameters, are you suppose to inspect every plot? I would not recommend that. > There's got to be a way to prioritize, even if...

Another modularization suggestion. We need independent draws also for MCMC rank ecdf difference plots, and maybe for other things. Provide a common helper function to do thinning with couple options:...