Finally figure it out! ArviZ’s `from_pyro`

only works with `pyro.infer.MCMC`

fitted object.

However, there is a workaround:

```
import arviz as az
predictive = Predictive(model, guide=guide, num_samples=500)
preds = predictive(x_test)
sanitized_preds = {k: v.unsqueeze(0).detach().numpy() for k, v in preds.items() if k != 'obs'}
pyro_data = az.convert_to_inference_data(sanitized_preds)
az.plot_trace(pyro_data, compact=True);
```

Then, it works!

Question: there is no mention of a solution like that (at least I couldn’t find) neither in ArviZ nor in Pyro documentation; shouldn’t that be communicated somewhere? If yes, where?

Cheers!