PGM-Causal-Reasoning

Experiments on structure learning of Bayesian Networks with emphasis on finding causal relationship

These experiments are done in context of the course Probabilistic Graphical Models (WiSe 18/19) held by Prof. Kristian Kersting at at the Technische Universität Darmstadt.

Datasets

Artifical generated data

LUCAS (LUng CAncer Simple set)

• http://www.causality.inf.ethz.ch/data/LUCAS.html

Real world data

CINA-dataset (Census Is Not Adult dataset)

• http://www.causality.inf.ethz.ch/challenge.php?page=datasets

---

Frameworks

SPFlow - An Easy and Extensible Library for Sum-Product Networks

• https://github.com/SPFlow/SPFlow

bnlearn - An R package for Bayesian network learning and inference

• http://www.bnlearn.com/

BDgraph: Bayesian Structure Learning in Graphical Models using Birth-Death MCMC (R)

• https://cran.r-project.org/web/packages/BDgraph/index.html

BNFinder - Tool for learning bayesian networks

• https://github.com/sysbio-vo/bnfinder
• http://bioputer.mimuw.edu.pl/software/bnf/
• https://academic.oup.com/bioinformatics/article/25/2/286/218091

Note: I needed to install a slightly older version for scipy in order to avoid: scipy/scipy#9606

SyntaxError: Non-ASCII character '\xe2'

using

sudo pip2 install scipy==1.1.0

libpgm - An endeavor to make Bayesian probability graphs easy to use

• https://pythonhosted.org/libpgm/

BayesSpy – Bayesian Python

• seems that it doesn't support structure learning functionality
• https://github.com/bayespy/bayespy

OpenGM - a C++ template library for discrete factor graph models and distributive operations

• seems that it doesn't support structure learning functionality
• http://hciweb2.iwr.uni-heidelberg.de/opengm/

Citation

If you find this project useful please consider citing:

@misc{queensgambit_experiments_2019,
	title = {Experiments on structure learning of {Bayesian} {Networks} with emphasis on finding causal relationship: {QueensGambit}/{PGM}-{Causal}-{Reasoning}},
	copyright = {MIT},
	shorttitle = {Experiments on structure learning of {Bayesian} {Networks} with emphasis on finding causal relationship},
	url = {https://github.com/QueensGambit/PGM-Causal-Reasoning},
	urldate = {2019-03-15},
	author = {Willig, Moritz and Czech, Johannes},
	month = feb,
	year = {2019},
	note = {original-date: 2019-01-20T13:56:16Z}
}

Our project report can be found here.

References

Graphical Models for Probabilistic and Causal Reasoning, Judea Pearl

• https://link.springer.com/chapter/10.1007%2F978-94-017-1735-9_12#authorsandaffiliations

Towards A Rigorous Science of Interpretable Machine Learning, Finale Doshi-Velez, Been Kim

• https://arxiv.org/abs/1702.08608

Partial orientation and local structural learning of causal networks for prediction, Jianxin Yin, You Zhou, Changzhang Wang, Ping He, Cheng Zheng, Zhi Geng

• http://proceedings.mlr.press/v3/yin08a/yin08a.pdf

An Exploration of Structure Learning in Bayesian Networks, Constantin Berzan

• https://thirld.com/files/senior_thesis.pdf

Learning Bayesian Networks with the bnlearn R Package

• https://arxiv.org/pdf/0908.3817.pdf