InferenceData is missing details about divergent transitions?

Questions
2

I second this concern. I switched from return_inferencedata=False to return_inferencedata=True so I can do more analyses on the returned InferenceData object, but since I don’t have attribute report in the output of pm.sample, I can’t check for convergence issues explicitly. Is there a way to retrieve report by using return_inferencedata=True? This is an unfortunate omission.

My temporary (and not-so-elegant) workaround is to copy pm.sampling.sample to a “utilities” module in my project, rename the function as pm_sample, and replace its last lines with:

    if return_inferencedata:
        return idata, trace
    else:
        return None, trace

So in my code, I call:

inf_data, trace = utils.pm_sample(..., return_inferencedata=True, ...)