Stochastic Message Passing

This repository is the official PyTorch implementation of "Permutation-equivariant and Proximity-aware Graph Neural Networks with Stochastic Message Passing".

IEEE TKDE

Arxiv

Dependencies

• PyTorch (tested on 1.12.0+cu113), please refer to PyTorch official site for installation

• PyTorch-geometric (tested on 2.0.4), please refer to PyTorch-geometric offical site for installation

• other dependencies are listed in requirements.txt, please install them with pip install -r requirements.txt

For datasets:

PPI and Email datasets are included in data folder. Please unzip ppi.zip first if you need to use PPI. The other datasets will automatically download and unzip when needed (thanks to the libraries networkx and obg )

Run

main.py is the entrance for a whole training-validation-testing process. Run python main.py -h for a full parameter list and information.

Alternatively, to run several tasks sequentially with more complex configures, please refer to run_all.py (to run all the tasks on all the dataset we adopt except PPA), and run_ppa.py (to run tasks on PPA). You can also refer to these files for suitable defualt parameter values.

Cite This

@ARTICLE{9721559,
  author={Zhang, Ziwei and Niu, Chenhao and Cui, Peng and Pei, Jian and Zhang, Bo and Zhu, Wenwu},
  journal={IEEE Transactions on Knowledge and Data Engineering}, 
  title={Permutation-equivariant and Proximity-aware Graph Neural Networks with Stochastic Message Passing}, 
  year={2022},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TKDE.2022.3154391}}