Hi all, (not sure if I’m the right category)
bayesplot hasn’t been updated in a while. There is a pull request of mine fixing a bug since a while now. And the killer feature of omitting y (Add new PPD module for plots without `y` variable by jgabry · Pull Request #222 · stan-dev/bayesplot · GitHub) is being there, finished but without a merge for more than a year.
Can I do something to help? Is the package alive and well? (Or should I start looking at tidybayes?)
2 Likes
It’s true we haven’t done a bayesplot release in quite a while, which is almost entirely my fault. It’s mostly just that I’m involved in too many R packages and other things now and there aren’t many other people actively working on bayesplot at the moment. help would be great!! I also just recently spoke with a student who is interested in helping out, but it might take a little while to get them up to speed.
tidybayes is great but I think it serves a different purpose, if I understand correctly (i might not!). bayesplot provides single functions that quickly produce plots but not necessarily tons of customizability, whereas tidybayes requires more coding but allows for a lot more customization. I would use bayesplot to get quick visual feedback after fitting models and tidybayes if you want more customizable graphics for publication.
5 Likes
just merged this! sorry for the delay.
2 Likes
ok, thanks!
Let me know how can I help then, I’m eager to see the killer feature of omitting y (Add new PPD module for plots without y variable by jgabry
What can I do to make it happen?
3 Likes
Yeah we really need to get that merged. I will add a few comments on the page for that PR over at the bayesplot repo
Thanks for offering to help. Have you tested out the branch recently? I just pushed some commits fixing the merge conflicts with master. If you haven’t tried it out recently then one way you could help (if you have time) is to try out the new ppd_ functions and make sure they work as expected.
I haven’t. I’ll do it.
2 Likes
It would be cool if bayesplot would support ggdist (a package for “Visualizations of distributions and uncertainty” that was separated from tidybayes). I understand that this would require refactoring bayesplot which would be a lot of work, but mention this wish here in case someone else than Jonah sees this and would have time and interest to work on it.
1 Like
Do you mean that the existing plots in bayesplot would be refactored to call ggdist internally instead of calling ggplot directly? That would definitely be a big project and I’m not sure all existing plots in bayesplot could be translated to ggdist, but some definitely could.
Another option would be to add additional plots to bayesplot in order to support the plots that ggdist can make (there are definitely some good ones that it supports that don’t have equivalents in bayesplot yet). So instead of refactoring the current functions we just add new ones, which I think would be less work actually. If eventually we wanted to deprecate any of the current plots we could eventually do that. @avehtari What do you think about this option?
It would be cool to be able to use the geoms/stats provided by ggdist.
I agree
I think it’s sensible.
1 Like
Hey,
I just wanted to thank you and @tjmahr!
Whenever all the changes are done, I’ll try again the dev version.
2 Likes
Thanks @bnicenboim. I think the PR for the PPD plots (the prior/posterior predictive distribution plots without y) is now ready if you want to try it again: