It looks like we’d need to undo Coerce coords into np.array in numpy_to_data_array by eidanch · Pull Request #1695 · arviz-devs/arviz · GitHub on ArviZ side, and change the test from testing a tuple to testing a multiindex. Or use a try except to use the coords as is and only if that fails conveet them to an array. That would fix the issue when using idata_kwargs, after that we’d need to look on how to support this better from pymc (I am not very familiar with internals of coord handling in v4 yet and can’t really see what unexpected consequences would each of the options have)
2 Likes