from_pymc3_predictions does not support arviz trace input

[kyleabeauchamp]

az.from_pymc3_predictions(predictions=ppc, posterior_trace=tr, model=model)                                                                                                                                                                                 

[...]/lib/python3.7/site-packages/arviz/data/io_pymc3.py in __init__(self, trace, prior, posterior_predictive, log_likelihood, predictions, coords, dims, model, save_warmup)
     99                 ].model
    100             self.nchains = trace.nchains if hasattr(trace, "nchains") else 1
--> 101             if hasattr(trace.report, "n_draws") and trace.report.n_draws is not None:
    102                 self.ndraws = trace.report.n_draws
    103                 self.attrs = {


AttributeError: 'InferenceData' object has no attribute 'report'

FWIW, I've also tried passing along the tr.posterior object rather than the full InferenceData object, but AFAIK the issue is that all from_pymc3_predictions code paths expect a pymc3 object, rather than an arviz trace.

[kyleabeauchamp]

Out[7]: '3.9.2'

In [8]: az.__version__                                                                                                                                                                                                                                                        
Out[8]: '0.9.0'

[kyleabeauchamp]

Actually, the idata_orig arg might do this, but I'm hitting an issue with dimensions, which I think is due to something I reported over in pymc3 (`ValueError: conflicting sizes for dimension) pymc-devs/pymc#4004