I’m looking to sample my priors and obtain both the canonical parameter values as well as their transformed counterparts. So far as I can tell, pm.sample_prior_predictive() omits the latter. Is there any easy way to trigger sample_prior_predictive() to include the transformed parameters?
Transformed parameters are not part of random sampling, only mcmc.
If you want you can either use mcmc to do prior sampling (remove observations from your model and call pm.sample) or compute them as deterministic.
But maybe it helps to first know why you want those?