WinGNN: Dynamic Graph Neural Networks with Random Gradient Aggregation Window

This repository is our PyTorch implementation of WinGNN.

Requirements

pip install -r requirements.txt

How to run

You can run the WinGNN with the following commands:

# UCI
python main.py --dataset uci-msg --lr 0.01 --maml_lr 0.008 --drop_rate 0.16 --window_num 8
# DBLP
python main.py --dataset dblp --lr 0.007 --maml_lr 0.003 --drop_rate 0.09 --window_num 8
# BitcoinAlpha
python main.py --dataset bitcoinalpha --lr 0.2 --maml_lr 0.003 --drop_rate 0.1 --window_num 8
# BitcoinOTC
python main.py --dataset bitcoinotc --lr 0.003 --maml_lr 0.006 --drop_rate 0.4 --window_num 7
# Reddit-Title
python main.py --dataset reddit_title --lr 0.07 --maml_lr 0.0009 --drop_rate 0.16 --window_num 10
# stackoverflow
python main.py --dataset stackoverflow_M --lr 0.03 --maml_lr 0.001 --drop_rate 0.1 --window_num 8 --num_layers 1 --num_hidden 32 --out_dim 16

Acknowledgement

Our source code and data processing are built heavily based on the code of Roland (https://github.com/snap-stanford/roland).

We modified loader.py in graphgym, and the modified one is loader.py in this project.

The data set download address is provided in the paper.

Reference

If you find this work is helpful to your research, please consider citing our paper:

@inproceedings{WinGNN,
  title={WinGNN: Dynamic Graph Neural Networks with Random Gradient Aggregation Window},
  author={Zhu, Yifan and Cong, Fangpeng and Zhang, Dan and Gong, Wenwen and Lin, Qika and Feng, Wenzheng and Dong, Yuxiao and Tang, Jie},
  booktitle={Proceedings of 29th {ACM} {SIGKDD} Conference on Knowledge Discovery and Data Mining},
  year={2023}
}