Exploit dependency DAG from Pyro models for nonparametric analysis

[eb8680]

While the design philosophy of this library is generally to frame causal computations as probabilistic ones, Pyro is a sufficiently rich foundation that other algorithmic approaches to causal inference can be incorporated into a larger workflow.

For example, Pyro includes utilities for computing the dependencies between random variables in a model. It should be straightforward to adapt the resulting data structures for use with existing packages like y0 that perform nonparametric analyses of causal graphs, e.g. computing nonparametric identifiability of a query.

[eb8680]

cc @djinnome

[qinqian]

Is there any literature for computing the nonparametric identifiability of a query?

[eb8680]

@qinqian see chapters 1-3 of Causality by Pearl for an introduction, especially chapter 3. This paper describes one algorithm for identification but it may be difficult to follow without reading the book chapters first.