This repository contains the code supplement for the paper "Exact Computation of Any-Order Shapley Interactions on Graph Neural Networks"

The code is written and tested with Python 3.10 and the requirements are listed in requirements_graphshapiq.txt. You can install the requirements using the following command:

pip install -r requirements_graphshapiq.txt

The implementation is organized in the shapiq package. For implementation details regarding the GraphSHAP-IQ implementation we refer to shapiq/graph_shap_iq.py. The shapiq package also contains the implementation of the GraphSHAP algorithm in shapiq/explainer/graph. To validate the experiments we refer to graph_shapiq_experiments. To see all plots and data please refer to the plots_in_paper folder.

The .interaction_values files created for the approximation experiments exceed 250MB and, thus, are not included in the repository. If you want to reproduce the results, please run the experiments in graph_shapiq_experiments and the files will be created.