One thing I noticed in the tutorial and in my own experiments is that the r_hat and ESS's for latent variables indicate poor sampling (i.e., high and low, respectively). But your pp_check's show that they are doing just fine. How do we understand these typically reliable indicators of model performance in the context of a latent variable model? How should we best be assessing fit, if those are no longer working?
Thanks!
(I wouldn't typically post a question like this as an issue on github, but that's what you suggested we do at the bottom of Full-Luxury)
One thing I noticed in the tutorial and in my own experiments is that the r_hat and ESS's for latent variables indicate poor sampling (i.e., high and low, respectively). But your pp_check's show that they are doing just fine. How do we understand these typically reliable indicators of model performance in the context of a latent variable model? How should we best be assessing fit, if those are no longer working?
Thanks!
(I wouldn't typically post a question like this as an issue on github, but that's what you suggested we do at the bottom of Full-Luxury)