bn-benchmarks

This repository contains Bayesian networks used in my experiments. The networks are found in the bayesian-networks directory in BIF and UAI format. Some networks come from the bnlearn repository; if you use them please use the appropriate citation:

@article{scutari2009learning,
  title={Learning Bayesian networks with the bnlearn R package},
  author={Scutari, Marco},
  journal={arXiv preprint arXiv:0908.3817},
  year={2009}
}

In the scripts directory you can find some utilities to computes marginal probabilities from UAI files, either using a WMC encodings (references below) or with ProbLog

CNF encodings

Weighted Model Counting

I use various encoding based on weighted model counting, and I use this wonderful tool coded by the Cril to transform UAI-formatted bn into the appropriate CNF. Currently, this repository contains the following encodings; if you use any of these, please cite the associated papers.

• ENC3

@inproceedings{chaviraCompilingBayesianNetworks2005,
  title = {Compiling {{Bayesian}} Networks with Local Structure},
  booktitle = {{{IJCAI}}},
  author = {Chavira, Mark and Darwiche, Adnan},
  year = {2005},
  volume = {5}
}

• ENC4

@inproceedings{chaviraEncodingCNFsEmpower2006,
  title = {Encoding {{CNFs}} to Empower Component Analysis},
  booktitle = {Theory and {{Applications}} of {{Satisfiability Testing-SAT}} 2006: 9th {{International Conference}}, {{Seattle}}, {{WA}}, {{USA}}, {{August}} 12-15, 2006. {{Proceedings}} 9},
  author = {Chavira, Mark and Darwiche, Adnan},
  year = {2006},
  publisher = {{Springer}}
}

• ENC4LINP

@inproceedings{bartImprovedCNFEncoding2016,
  title = {An Improved {{CNF}} Encoding Scheme for Probabilistic Inference},
  booktitle = {Proceedings of the {{Twenty-second European Conference}} on {{Artificial Intelligence}}},
  author = {Bart, Anicet and Koriche, Fr{\'e}d{\'e}ric and Lagniez, Jean-Marie and Marquis, Pierre},
  year = {2016}
}

Projected Weighted Model Counting

For now, the only projected weighted model counting encoding is the one proposed in Schlandals.

• sch

@inproceedings{dubrayProbabilisticInferenceProjected2023a,
  title = {Probabilistic {{Inference}} by {{Projected Weighted Model Counting}} on {{Horn Clauses}}},
  author = {Dubray, Alexandre and Schaus, Pierre and Nijssen, Siegfried},
  year = {2023},
  publisher = {Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  doi = {10.4230/LIPIcs.CP.2023.15},
}