Scriddie / Varsortability

Implementations of var-sortability, sortnregress, and chain-orientation as presented in the article "Beware of the Simulated DAG": https://arxiv.org/abs/2102.13647.

[Update 09/2023] Our new library CausalDisco provides the baseline algorithms and analytics tools presented here, as well as a new scale-invariant version in a single python package for causal discovery benchmarking.

Beware of the Simulated DAG! Causal Discovery Benchmarks May Be Easy To Game

This repository contains stand-alone implementations of varsortability, sortnregress, and chain-orientation as presented in

[1] Reisach, A. G., Seiler, C., & Weichwald, S. (2021). Beware of the Simulated DAG! Causal Discovery Benchmarks May Be Easy To Game.

For a basic experimental set-up for the comparison of causal structure learning algorithms as shown in the same work, see the VarsortabilityExperimentSuite repository.

If you find this code useful, please consider citing:

@article{reisach2021beware,
  title={Beware of the Simulated DAG! Causal Discovery Benchmarks May Be Easy to Game},
  author={Reisach, Alexander G. and Seiler, Christof and Weichwald, Sebastian},
  journal={Advances in Neural Information Processing Systems},
  volume={34},
  year={2021}
}

To run, perform the following actions within the /src directory:

1. Install dependencies by running ./install.sh in this directory.
2. For varsortability run source env/bin/activate; python varsortability.py in the current directory.
3. For sortnregress, run source env/bin/activate; python sortnregress.py in the current directory.
4. For chain-orientation
   • run source env/bin/activate; python chain_orientation.py in the current directory (may take some time).
   • run source env/bin/activate; python chain_orientation_three_vars_symbolic.py in the current directory (may take some time).