Loo on GitHub vs. CRAN



Dear Stan developers,
I have a quick question about the loo package.
I’m interested in some functions (eg, model_weights) that have been on GitHub for 6-10 months. Is there any specific reason why that version of the package has not been submitted to CRAN yet, eg, known bugs to be fixed or something like that? I just wanted to check with you before finalizing some analyses using the package on GitHub.

Thank you.


There is nothing wrong with the code. Specific reason is that there has to be tests in the package which have not yet been properly added. There are several non-specific reasons why these tests have been with “next week” status for couple months. So when I say that these functions will be in CRAN next week, you can assume some uncertainty in that prediction. Meanwhile you can use github version. It would be great if you can also report here if you have found the functions useful :)


thank you, Aki, that helps.
I’m now also looking at “Piironen & Vehtari, 2017, Comparison of Bayesian predictive methods for model selection” and the ‘projpred’ package. The “projection predictive method” looks like something that may be relevant for some of the stuff I’m working on too. It’s not easy to keep up with all your work :-) Hope ‘projpred’ will support more models in the future.


model_weights are very close to being merged to loo 2.0, shouldn’t take long time anymore

It has been now also officially released in CRAN

Me too. It’s a resource problem to get more features. Paul Bürkner is looking at brms support (but that does not mean more models).