This repository contains the source code for the paper GNNExplainer: Generating Explanations for Graph Neural Networks by Rex Ying, Dylan Bourgeois, Jiaxuan You, Marinka Zitnik & Jure Leskovec, presented at NeurIPS 2019.

[Arxiv] [BibTex] [Google Scholar]

@misc{ying2019gnnexplainer,
    title={GNNExplainer: Generating Explanations for Graph Neural Networks},
    author={Rex Ying and Dylan Bourgeois and Jiaxuan You and Marinka Zitnik and Jure Leskovec},
    year={2019},
    eprint={1903.03894},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

See INSTALLATION.md

This is the model that will be explained. We do provide pre-trained models for all of the experiments that are shown in the paper. To re-train these models, run the following:

python train.py --dataset=EXPERIMENT_NAME

where EXPERIMENT_NAME is the experiment you want to replicate.

For a complete list of options in training the GCN models:

python train.py --help

TODO: Explain outputs

To run the explainer, run the following:

python explainer_main.py --dataset=EXPERIMENT_NAME

where EXPERIMENT_NAME is the experiment you want to replicate.

For a complete list of options provided by the explainer:

python train.py --help

The result of the optimization can be visualized through Tensorboard.

tensorboard --logdir log

You should then have access to visualizations served from localhost.

We provide an example visualization through Jupyter Notebooks in the notebook folder. To try it:

jupyter notebook

The default visualizations are provided in notebook/GNN-Explainer-Viz.ipynb.

Note: For an interactive version, you must enable ipywidgets

jupyter nbextension enable --py widgetsnbextension

You can now play around with the mask threshold in the GNN-Explainer-Viz-interactive.ipynb.

TODO: Explain outputs + visualizations + baselines

We provide export functionality so the generated masks can be visualized in other data visualization frameworks, for example d3.js. We provide an example visualization in Observable.

Datasets with a * are passed with the --bmname parameter rather than --dataset as they require being downloaded manually.

TODO: Provide all data for experiments packaged so we don't have to split the two.

A graph attention model is provided. This repo is still being actively developed to support other GNN models in the future.

See CHANGELOG.md