Write conversion guides

[OriolAbril]

Tell us about it

Convert the cookbook into a collection of notebooks (one per PPL) with extended documentation on the respective converter function from_xyz. PPL tracker:

• PyMC3 (if the examples below are good enough it's only a matter of putting the notebook together)
• PyStan
• CmdStanPy
• Pyro
• NumPyro
• emcee (there are extensive examples in the from_emcee api docs that can help write this)
• TensorFlow Probability

Thoughts on implementation

I think it would be great to define a set of tasks of increasing difficulty and then (if possible) show how to tackle them in each of the PPLs and how to convert the data to InferenceData. The current example in the cookbook is a possible starting point. The examples (click on the QR codes) for my ProbProg poster are a more complete example of InferenceData creation. And my pymc3-arviz blogpost could be used as intermediate examples.

Note: the cookbook would not be deleted, it would keep the numpy array, dict, dataframe... conversions

[OriolAbril]

Tagging as Discussion for now until we agree on what kind of conversion tasks and models we want to use for our ArviZ rosetta stone

[OriolAbril]

Extra pro of doing this (which can be done once the examples in from_emcee docstring are moved to a notebook). Doc generation won't need any inference library so we can have the doc job depend only on basetests and remove the externaltests dependency to speed-up CI time.