Runtime Exception with NUTS

It’s the starting point. This seems to work:

with model:
    idata = pm.sample(init='adapt_diag',
                      chains=4,
                      return_inferencedata=True)

Not sure why I didn’t think about that earlier.