Any reason why the HPD in Pymc3 differs from what the code in Statistical Rethinking gets?

Any idea why the HPD between the Pymc3 version and the Rethinking package as far off as it is below? I get that they won’t be exact but is this level of difference expected?

I’m running into the same case as I’m doing the exercises in pymc3 when compared to his solutions.

Screenshot from Rethinking

Screenshot from pymc3 resources github repo

No problem here


PyMC3 version 3.3

Hmmmm, I’ll check the version I’m on. Thank you