I am trying to plot a conditional posterior probability using ArviZ.
Imagine I have an InferenceData object of some sort, that I have obtained via an Markov Chain.
For instance, let us assume that I have two variables: a discrete variable $u$, that take values into $\left{-1, 1\right}$ and a continuous variable $x$ that, conditionally on $u$, is drawn from $\mathcal{N}\left(-10, 1\right)$ if $u = -1$ and from $\mathcal{N}\left(10, 1\right)$ if $u = 1$.
My goal is to plot the conditional posterior distribution for $x$ in the two cases. In other words, I would like to obtain two distinct plots for the two cases.
I have insofar tried to slice the ArviZ object, using the xarray syntax. But it is not working.
If we name the instance mcmc (which I have created from a dictionary), I obtain that it is an arviz.InferenceData object. It contains two dimensions: chain: 3 (for instance) and draw: 2000. Then, I have the Coordinates, the Data variables and the Attributes.
Selecting the Data variables (mcmc.posterior) and then slicing using the class method sel from xarray does not yiel the conditional posterior distribution.
How should I proceed?
