caus-am / sigmasep

Code for UAI 2018 paper by Forré & Mooij (causal discovery with mSCMs using sigma-separation)
BSD 2-Clause "Simplified" License
11 stars 2 forks source link

sigmasep

Code for UAI 2018 paper by Forré & Mooij (causal discovery with mSCMs using sigma-separation)

Version

v1.2

ChangeLog

License

This code is licensed under the BSD 2-clause license (see file LICENSE).

Citation

When making significant use of this code for a scientific publication, please cite:

@inproceedings{ForreMooij_UAI_18,
  author    = {Patrick Forr{\'e} and Joris M. Mooij},
  title     = {Constraint-based Causal Discovery for Non-Linear Structural Causal Models with Cycles and Latent Confounders},
  booktitle = {Proceedings of the 34th Annual Conference on {U}ncertainty in {A}rtificial {I}ntelligence ({UAI}-18)},
  year      = 2018
}

A considerable part of the code is based on the code accompanying the following paper:

@inproceedings{Hyttinen++2014,
  author    = {Hyttinen, A. and Eberhardt, F. and J{\"{a}}rvisalo, M.},
  title     = {Constraint-based Causal Discovery: Conflict Resolution with Answer Set Programming},
  booktitle = {Proceedings of the Thirtieth Conference on Uncertainty in Artificial Intelligence, ({UAI}-14)},
  address   = {Quebec City, Quebec, Canada},
  pages     = {340--349},
  year      = {2014}
}

Getting started

The code isn't designed to run with one keystroke or be user-friendly. It should however be helpful to reproduce the experiments reported in our paper, and as a starting point for a more user-friendly implementation.

To reproduce the experiments reported in the paper, look into the python notebook python/Experiments_and_Plotting.ipynb

Frequently Asked Questions

None so far. For questions, you can email the authors.