This repository contains the code used in the paper "Finding Counterfactually Optimal Action Sequences in Continuous State Spaces", published at NeurIPS 2023.

All the experiments were performed using Python 3.9. In order to create a virtual environment and install the project dependencies you can run the following commands:

python3 -m venv env
source env/bin/activate
pip install -r requirements.txt

The directory src contains the source code for the experiments.

The directory scripts contains bash scripts that use the aforementioned code and pass parameter values required for the various experiments.

• build_datasets.sh is preprocessing the sepsis management data
• build_scm.sh is used to train the final SCM
• cv_scm.sh is used to evaluate the goodness of fit of the SCM via cross-validation, under different values of the networks' Lipschitz constants
• grande_experiment_slurm.sh is the main script used to run experiments using a slurm scheduler and this is where each experiment's configuration is set
• single_experiment_slurm.sh is a helper script called by grande_experiment_slurm.sh
• solve_facility_location.sh is used to precompute anchor sets using the Facility-Location method

The directory notebooks contains jupyter notebooks producing the figures appearing in the paper. Each notebook reads output files generated by scripts. For details, see grande_experiment_slurm.sh and the description within each notebook.

The directory figures is used for saving the figures produced by the notebooks.

The directory outputs and its sub-directories are used for saving the outputs generated by the scripts.

If you use parts of the code in this repository for your own research, please consider citing:

@inproceedings{tsirtsis2023finding,
    title={Finding Counterfactually Optimal Action Sequences in Continuous State Spaces},
    author={Tsirtsis, Stratis and Gomez-Rodriguez, Manuel},
    booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
    year={2023}
}