Thanks for the explanation. Everything makes more sense now.
Indeed, it took me a while to figure out how to pass the mu_home parameter, since it is not truly observed. I figure it out based on this topic DensityDist from scipy.stats distribution - #3 by balarsen . However, in the very same topic, the DensityDist is sampled without passing any extra argument to the sampler. I guess that was correct when the topic was written, but the same code would not work now, if I understand correctly. This can be very odd for someone who does not know how PyMC3 is evolving. So, I reiterate that it would be nice to mention this issue with sampling in the documentation – and to have an up-to-date working notebook on the different uses of DensityDist. Anyway, thank you again for your availability and help.