Posterior predictive sampling with data variance

Yes, thanks @simon_o for checking. In fact, the standard deviation match was already confirmed in the original file, and I was glad that the test is positive.
This brings us back to my original questions:

  • How to reduce the extra dimension in the sampling output? (averaging is no option)
  • How about the use of find_MAP(), when to take it, comparison to using trace?
  • How does the prediction variability relate to model residual?

:wink: No offense, I really appreciate your time and help!