Exam data - inference of student performance

Questions
2

Hi,

  • Bad initial energy associated with Mean of empty slice usually signals missing values somewhere. You mentionned some tests do have missing values, and indeed the error message mentions nH, so I’d bet there are NaNs in nH_obs. PyMC3 handles and infers them automatically, but you have to wrap them into a numpy masked array – see this NB, Code 14.7, for an example.

  • For more details on this common error message, you can take a look at the FAQ.

  • Finally, I think I’d use more informative priors – Uniforms are usually not a good choice.

Hope this helps :vulcan_salute: