Implementation of ID algorithm?

[djinnome]

Hi there,

Back door and IV is nice, but for larger graphs, I am wondering if it would be worthwhile implementing Ilya Shpitsers ID algorithm, which is complete. Implementations of ID are currently available in R and in clojure.

By the way, if you haven't checked out Whittemore, it implements a causal programming paradigm similar in philosophy to doWhy

[amit-sharma]

Thanks @djinnome ! Definitely, implementing the ID algorithm is on our priority list, but being delayed with work on refutations. Would you be willing to contribute to the development?

Yes, I really the idea behind Whittemore--had a chance to talk to Joshua (who built Whittemore) a few weeks back.

[djinnome]

Hi Amit, Implementing the ID algorithm, and possibly [the ID* and IDC* algorithms](https://arxiv.org/abs/1206.5294) in DoWhy would be exactly the kind of contribution our causality reading group is looking to make. We are about 1/3 of the way through [Chapter 4](http://bayes.cs.ucla.edu/PRIMER/primer-ch4.pdf) of Judea Pearl's Causal Inference in Statistics: A Primer. When we are done with that, we will focus our effort on the implementation of these algorithms. Sincerely, Jeremy

[amit-sharma]

Sounds great. Let me know once you are ready. Happy to chat in case you need you need more details on the library's code structure.

[gianlucadetommaso]

Hi all, was there meanwhile any progress on this front?

[amit-sharma]

yes, the ID algorithm is now implemented for identification.
https://microsoft.github.io/dowhy/example_notebooks/identifying_effects_using_id_algorithm.html

However, estimation based on the probability expressions returned by ID algorithm is not yet supported.

[gianlucadetommaso]

Thank you. Any chance somebody also implemented the Adjustment Criterion by Shpitser (here, Definition 5)? This is a complete criterion for identification via covariate adjustment.

[amit-sharma]

That's not implemented, but it will a simple exercise to implement it by extending the backdoor criterion. Would you like to implement it @gianlucadetommaso ? I've added an issue #402