Combine InferenceData object

Tell us about it

We should have the possibility to combine/merge inference data objects.

Thoughts on implementation

We could support both __add__ and specific function.

posterior = az.convert_to_inference_data({'mu' : np.array([...])}, "posterior")
posterior_predictive = az.convert_to_inference_data({'y_hat' : np.array([...])}, "posterior_predictive")
inference_data = posterior + posterior_predictive

inference_data = az.concat(posterior, posterior_predictive)

Easiest this would be if we by default support only combining individual groups (combine only if no groups overlap). We could also have a possibility to add chains/draws or both.

[ahartikainen]

Maybe. I don’t think we have not decided which structure is simple enough to be useful.

[ahartikainen]

Partially closed by https://github.com/arviz-devs/arviz/pull/709

Other cases are “too hard” to have general solution.

For example, how should variables behave if they are missing the selected axis.

In future, if there is demand, we can create a convenience function to handle concat, where user supplied function is applied against all groups ::: not sure if this is really doable.