WanyuGroup / CVPR2022-OrphicX

Official code for the CVPR 2022 (oral) paper "OrphicX: A Causality-Inspired Latent Variable Model for Interpreting Graph Neural Networks."

OrphicX

This repo covers the implementation for our paper OrphicX.

Wanyu Lin, Hao Lan, Hao Wang and Baochun Li. "OrphicX: A Causality-Inspired Latent Variable Model for Interpreting Graph Neural Networks," in the Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2022), Oral Presentation, New Orleans, Louisiana, June 19-24, 2022.

Prepare Python environment

conda create -n orphicx python=3.8.8
conda activate orphicx
conda install pytorch cudatoolkit=10.2 -c pytorch
conda install opencv scikit-learn networkx pandas tqdm matplotlib seaborn
pip install tensorboardx
pip install torch-scatter     -f https://pytorch-geometric.com/whl/torch-1.8.0+cu102.html
pip install torch-sparse      -f https://pytorch-geometric.com/whl/torch-1.8.0+cu102.html
pip install torch-cluster     -f https://pytorch-geometric.com/whl/torch-1.8.0+cu102.html
pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-1.8.0+cu102.html
pip install torch-geometric

Extract dataset and checkpoint from zip file

unzip ckpt.zip

Run experiments with checkpoints

python orphicx_node.py --gpu --dataset syn1 --plot_info_flow
python orphicx_node.py --gpu --dataset syn4 --plot_info_flow
python orphicx_graph.py --gpu --dataset Mutagenicity --plot_info_flow
python orphicx_graph.py --gpu --dataset NCI1 --plot_info_flow

Calcualte information flow is slow on CPU. If you don't have a GPU, please run following commands:

python orphicx_node.py --dataset syn1
python orphicx_node.py --dataset syn4
python orphicx_graph.py --dataset Mutagenicity
python orphicx_graph.py --dataset NCI1

Retrain OrphicX from scratch

python orphicx_node.py --gpu --dataset syn1 --output syn1_retrain --retrain
python orphicx_node.py --gpu --dataset syn4 --output syn4_retrain --retrain
python orphicx_graph.py --gpu --dataset Mutagenicity --output mutag_retrain --retrain
python orphicx_graph.py --gpu --dataset NCI1 --output nci1_retrain --retrain