Colin Carroll

right now the behavior is equivalent to `data.posterior[var_names]` (and specifically, `data.posterior[var_names].sel(**coords)`, though, which I like: i.e., matching xarray's behavior.

Thanks for the input (and the volunteering!) I'd actually vote for the option to stay `True` because of: - backwards compatibility (we don't have strong guarantees, but it means something)...

Can you share exactly how you're installing stuff? This throws a more sane error, and has `jax` version 0.2.6, which looks like it matches yours. 

Wow! Can repro, and can confirm that the same behavior happens with PRNGKey 86074. More seriously, [the bug happens here](https://github.com/arviz-devs/arviz/blob/master/arviz/data/base.py#L265), when trying to grab the library version to stamp the...

Wow, here's something terrible, then. Restarting then running again fixes it, with the imports in either order.

that's a great suggestion -- i'm not sure how to _do_ that, though -- in theory, we just have an InferenceData object that is returned, but don't have any structured...

Hi - It looks like the colab is locked down, so I can not access it.

Note that the data is not saved with the colab, so I can not run this, but it looks as though the problem is with your use of `tfp.bijectors.CorrelationCholesky(ni)`. Note...

Fwiw, I've found some of the function transformations (inverse and ildj) here super useful, and the wheel is only 200kb, so I haven't minded that as a requirement for now.

Maybe just an adapter to map `draws -> num_steps` and `tune -> num_adapt_draws`? I think in the example above, bayeux will ignore the `adapt.run` kwarg. It is undocumented, but in...