I’m trying to get my head round PyMC3 in order to port some existing code written (and working) for PyMC. I need to reproduce the functionality of PyMC’s draw_from_prior(). I see PyMC3 has functions called sample_prior_predictive() and sample_prior(). However these do not seem to behave in the same way. On inspecting the returned node values, I noticed that new values are being generated for any observed model nodes. Whereas for PyMC, draw_from_prior() always returned the observed value. Is there anyway I can reproduce the original functionality?