How to evaluate the normalized variational (approximate) log posterior?

After doing a variational .fit in PyMC, it’s straightforward to generates samples from the variational approximation. I would also like to evaluate the normalized variational log posterior function at an arbitrary point — is that possible?

There are tantilizing attributes like logq and logq_norm in the approximation classes like MeanField and FullRank but I don’t know whether or how I can evaluate them.