I’ve been busy, too, but today had time to check back to this.
It seems there’s still some confusion about the goals of projpred
- I think the goals of posteriordb are the Stan goals, that is the best possible Bayesian inference. Including support for other PPLs would not change that, which if I understood correctly was one of @betanalpha’s worries.
- We have talked (and still continue talking) this autumn with developers of PPLbench and Inference gym which are two other testing/benchmarking frameworks we are aware. With the developers of these other packages we agreed that the goals of these three packages are different and it’s sensible to continue as three separate projects. Thus there is no need to include support for goals that are not about best possible Bayesian inference, which if I understood correctly was one of @betanalpha’s worries.
- We also agreed with other package developers that there is enough overlap that it’s helpful keep talking and sharing information. The main collaboration would be sharing models and reference results, but also agreeing on common summaries and model sets for the common goals such as the best possible Bayesian inference. If I understood correctly @betanalpha agrees that it would be great if other PPL devs would share the this goal.
- Supporting other PPLs is planned to be lightweight, and with sharing models and refrence results with others the maintenance work is aimed to be kept minimal.
So if I’ve understood correctly there is not much disagreement after all?