Learning Actionable Counterfactual Explanations in Large State Spaces

Code to accompany the paper "Learning Actionable Counterfactual Explanations in Large State Spaces".

Basic Usage

• Packages are mentioned in requirements.txt

• Fully synthetic data generation. The scripts and .ipynbs files for creating the instances of agents and their hl-discrete CFEs are detailed in the hl-discrete-CFEGen folder

  • Edit the parameters based on what you need. E.g., the number of actionable features, $p_a$, $p_f$, e.t.c.

• The real-world datasets are added in the folder other datasets

  • Note that BMI, WHR data are from here, BRFSS/health risk data is from here, caloric prices is from here, and Foods nutritional composition is from here

• In this repository, there are 4 data-driven CFE generators added: hamming distance (baseline), hl-continuous (high-level continuous), hl-discrete (high-level discrete), and hl-id (high-level-identifier) data-driven CFE generators.

  • Each generator has scripts and sample notebooks
  • There is also code on varied actions access and varied features satisfiability CFE generation.

• If you need access to the semi-synthetic datasets or any other files, please email us and we will send it to you.

Citation

If you find this work useful in your research, please cite our paper:

@article{naggita2025learningactionablecounterfactualexplanations,
	title        = {{Learning Actionable Counterfactual Explanations in Large State Spaces}},
	author       = {Keziah Naggita and Matthew R. Walter and Avrim Blum},
	year         = 2025,
	journal      = {Transactions on Machine Learning Research (TMLR) (To Appear)}
}

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For questions or collaborations, please contact us here.