Back door adjustments

[sotomsa]

First version of exaustive search minimal admissable sets that satisfy the back door criterion.

It is based on a d-separation function from pgmpy package (https://github.com/pgmpy/pgmpy). It implements an algorithm found in "Probabilistic Graphical Models: Principles and Techniques" - Koller and Friedman. Is it possible to integrate that piece of code giving the proper credit to the coders? ( That packagee is under MIT License but I do not know much about licenses)

I have also found a polinomial time version of the minimal admissable sets algorithm described in https://arxiv.org/abs/1202.3764 by Johannes Textor and implementedin his cool project DAGitty (http://www.dagitty.net/dags.html) which is a javascript browser-based causal inference analyzer.

At the last minute I realized that you use different naming conventions for functions while I use camelCase naming convention. I can change that later if the rest of the code is ok.

[sotomsa]

Forgot to mention, it only works with DAGs at this moment

[sotomsa]

Hello Adam,

I have finally found some time to improve the PR code. Mainly, I have

• Changed the function names so that it has the same naming convention that you use.
• Converted the function into a generator function as you suggested for MVP
• Improved the test coverage which will be useful in order to check a polynomial time version of the same algorithm.
• Temporarly added all the code that is used from pgmpy (all is stored in utils.py) so that it is easier to check how the code works.

Best,