Discrete RV and marginalisation

Questions v3
2

Yes the models should derive equivalent conclusions for w, and the non-marginalized seems biased.

I would suggest you try:

  1. Many more samples with the non-marginalized sampler

  2. pymc4 on google colab quickly (it’s installed by default) as there was a weird bug sometimes when the data was all ones or zeros in Theano, which was fixed after pymc3 I think?