Posterior Predictive Plots with multivariate Outcomes

[NathanielF]

Posterior Predictive Checks and Plots

I've been working with multivariate outcome data lately in a SEM/CFA case and I found it's generally useful to be able to plot the posterior predictive checks distinguishing the different outcomes rather than a total amalgamated outcome

See here: https://www.pymc.io/projects/examples/en/latest/case_studies/CFA_SEM.html#posterior-predictive-checks

Would it be useful or feasible to add this kind of check to the plot_ppc functionality?

[OriolAbril]

This should already be possible via the flatten keyword. I added an example to one of my blogposts: https://oriolabrilpla.cat/en/blog/posts/2023/plot_ppc.html#flatten. That being said, this should be easier and much more flexible with the refactored version https://arviz-plots.readthedocs.io/en/latest/api/generated/arviz_plots.plot_ppc_dist.html, and it is probably a good idea to add an example of this to the gallery which should also ensure both api and docs are better for doing this.

Note the grouby behaviour in plot_ppc for repeated coordinates is still wip in the refactored version though but it should not have any effect on the example you shared as coord values are never repeated

[NathanielF]

Perfect @OriolAbril I didn't know this!

Also Cc @fonnesbeck FYI