This repo provides the official implementations for the experiments described in the following paper:

Subgroup Generalization and Fairness of Graph Neural Networks

Jiaqi Ma*, Junwei Deng*, and Qiaozhu Mei. NeurIPS 2021.

(*: equal constribution)

• torch 1.8.0 pytorch-1.8.1
• dgl 0.4.3 dgl-0.6.1
• networkx 2.4
• numpy 1.17.3 numpy-1.21.2

• torch 1.8.2 pytorch-1.8.1
• torch-geometric pyg-2.0.1
• networkx 2.4
• ogb 1.3.2
• numpy 1.17.3 numpy-1.21.2

If you choose to install the dependencies individually:

conda install pytorch=1.8 -c pytorch
conda install pyg -c pyg -c conda-forge
conda install networkx=2.4
conda install ogb==1.3.2 -c conda-forge
pip install dgl

For REPL (ipython), we install: conda install notebook.

Else, recreate the ma conda environment:

conda env create --file conda-envs/ma.yml

Activate with conda activate ma, and deactivate with conda deactivate.

NOTE: The specified version set is unsatisfiable: e.g., a) numpy==1.17.3 -> python[version='>=3.6,<3.7.0a0|>=3.7,<3.8.0a0|>=3.8,<3.9.0a0']),b) pytorch-sparse==0.6.12 -> pytorch=1.8, and, c) pytorch-sparse==0.6.12 -> python[version='>=3.9,<3.10.0a0'] -> libffi[version='>=3.3,<3.4.0a0'].

Please refer to the readme in each subfolder.

@article{DBLP:journals/corr/abs-2106-15535,
  author    = {Jiaqi Ma, Junwei Deng, Qiaozhu Mei},
  title     = {Subgroup Generalization and Fairness of Graph Neural Networks},
  journal   = {CoRR},
  volume    = {abs/2106.15535},
  year      = {2021},
  url       = {https://arxiv.org/abs/2106.15535},
  eprinttype = {arXiv},
  eprint    = {2106.15535},
  timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2106-15535.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}