Matthew Kay

Yeah the reason I'm interested specifically in comparisons of SPI to HDCI (not HDI) is that HDCI uses the CDF approach from HDInterval, so if those two methods are similar...

I should probably state more clearly what hdci is in the docs. It uses the algorithm for a continuous interval from HDInterval, which basically uses the ECDF to find the...

> Unless it already exists, it would be quite interesting to create a gist/vignette/paper that would 1) describe and explain the advantages/caveats of each interval method, 2) provide clean, community-contributed,...

Alright I started playing with this for my own curiosity. Some comparisons here: https://github.com/mjskay/interval-estimators/blob/main/interval-estimators.md (source Rmd in same repo). I can see three broad classes worth investigating: - equi-tailed intervals,...

> Would it be interesting to compare these methods with [ggdist::curve_interval()](https://mjskay.github.io/ggdist/reference/curve_interval.html) in the context of interval-bands? Or is that too different of an approach I think the comparison could make...

Yeah I think something like this is definitely doable. I was already thinking about making curve_interval generic so that posterior::rvar objects could be passed to it. Internally since rvars just...

Yeah something like this would be useful. Not sure the best way to do it, as the wilkinson algorithm does not have regularly spaced bins, so just setting a boundary...

Uh yes actually! Because the draws formats from {posterior} have been integrated into the stan packages and are supported in {tidybayes}, this is a lot easier now. If you're using...

Yup, see the ggplot34 branch of ggdist for a mostly complete fix (still needs a bit of polish): https://github.com/mjskay/ggdist/tree/ggplot3.4

more thoughts: the gradient pdf/cdf thing isn't really worth it atm. This should be doable by making sure geom_lineribbon works with infinite bounds. So I'd say the simple version of...