Yep, that sounds the same as what I found a little while ago:
- Discourse post: Can't seem to `sample_prior_predictive` on model with missing value imputation
- My gist with MRE: MRE_missing_values · GitHub
- The currently open issue and dicussion on Github: Give NoDistribution a random method so that `sample_prior_predictive` can work · Issue #4598 · pymc-devs/pymc3 · GitHub
As per @ricardoV94 's note, the issue seems avoidable by using shape rather than dims