Counterfactual Explanations with Probabilistic Guarantees on their Robustness to Model Change

This repository contains the code for BetaRCE method.

Setup

1. Setup Dependencies

Recursive clone:

git clone --recurse-submodules https://github.com/istepka/CARLA.git

Or

git clone https://github.com/istepka/CARLA.git 
git submodule update --init --recursive

2. Install Environment

Conda:

conda env create --name envname --file=environment.yml

or Pip (tested on Python 3.11.7):

python -m pip install -r requirements.txt

---

Usage

Directory structure:
experiments
├── notebooks
├── visualizations
images
data
configs
src
├── explainers
├── datasets
├── classifiers
├── experiment.py
experiment_runner.py

*Make sure that your python is calling scripts from the root directory level to avoid issues with wrong paths

---

Reproduce experiments

To run experiments for the paper we utilized our internal compute cluster running on slurm. To facilitate efficient use of resources we created a script that runs experments on slurm via hydra framework.

Scripts for running experiments are located in experiments folder.

Citation

If you find this work useful, please cite it as:

@inproceedings{stepka2025,
        author    = {Ignacy St\k{e}pka and Mateusz Lango and Jerzy Stefanowski},
        title     = {Counterfactual Explanations with Probabilistic Guarantees on their Robustness to Model Change},
        booktitle = {Proceedings of the 31st SIGKDD Conference on Knowledge Discovery and Data Mining},
        year      = {2025},
        month     = aug,
        address   = {Toronto, Canada},
      }