On the Robustness of Global Feature Effect Explanations

This repository is a supplement to the following paper:

Hubert Baniecki, Giuseppe Casalicchio, Bernd Bischl, Przemyslaw Biecek. On the Robustness of Global Feature Effect Explanations. ECML PKDD 2024 https://arxiv.org/abs/2406.09069

@inproceedings{baniecki2024robustness,
    title     = {On the Robustness of Global Feature Effect Explanations},
    author    = {Hubert Baniecki and 
                 Giuseppe Casalicchio and 
                 Bernd Bischl and 
                 Przemyslaw Biecek},
    booktitle = {ECML PKDD},
    year      = {2024}
}

Install the environment

1. mamba env create -f env.yml
2. install OpenXAI:
   • download https://github.com/AI4LIFE-GROUP/OpenXAI
   • remove version of torch
   • mamba activate robustfe
   • pip install .

Run the experiments

• experiment1.ipynb uses the algorithm (Baniecki et al., 2022) implemented in src to perform experiments reported in Section 5.1
• experiment2.ipynb, experiment2_plot.ipynb perform experiments reported in Section 5.2
• results directory contains metadata of results from running experiment1.ipynb and experiment2.ipynb

Check out also

Adebayo et al. Sanity Checks for Saliency Maps. NeurIPS 2018

Baniecki et al. Fooling Partial Dependence via Data Poisoning. ECML PKDD 2022

Gkolemis et al. RHALE: Robust and Heterogeneity-aware Accumulated Local Effects. ECAI 2023

Lin et al. On the Robustness of Removal-Based Feature Attributions. NeurIPS 2023