[AAAI 2025] Subgraph Aggregation for Out-of-Distribution Generalization on Graphs

Official PyTorch implementation of the paper Subgraph Aggregation for Out-of-Distribution Generalization on Graphs(SuGAr) in AAAI 2025.

Pipeline

Instructions

Installation and data preparation

run the following command to install environment:

pip install -r requirement.txt

plus the DrugOOD benchmark repo and config.

The data used in the paper can be obtained following these instructions.

Running example

Train

python main.py -c_in 'feat' -c_rep 'feat'  --seed '[1,2,3,4,5,6,7,8,9,10]' --num_layers 3 --dataset 'SUMotif' --bias 0.6 --r 0.25 --contrast 2 --spu_coe 0 --model 'gcn' --dropout 0 --epoch 40 --init 1 --penalty 1 --sample_rate 95 --save_model

Ensemble

python ens.py -c_in 'feat' -c_rep 'feat'   --num_layers 3 --dataset 'SUMotif' --bias 0.6 --r 0.25 --contrast 2 --spu_coe 0 --model 'gcn' --dropout 0 --init 1 --penalty 1 --sample_rate 95 --divp

Misc

If you find our paper and repo useful, please cite our paper:

@article{Liu_2025_sugar, 
title={Subgraph Aggregation for Out-of-Distribution Generalization on Graphs}, volume={39}, 
url={https://ojs.aaai.org/index.php/AAAI/article/view/34065}, 
DOI={10.1609/aaai.v39i18.34065},  
number={18}, 
journal={Proceedings of the AAAI Conference on Artificial Intelligence}, 
author={Liu, Bowen and Li, Haoyang and Wang, Shuning and Nie, Shuo and Zhang, Shanghang}, 
year={2025}, 
month={Apr.}, 
pages={18763-18771} 
}

We would like to acknowledge the contribution from CIGA to the base codes.