Unleashing the Power of Graph Data Augmentation on Covariate Distribution Shift

We provide a detailed code for "Unleashing the Power of Graph Data Augmentation on Covariate Distribution Shift".

Yongduo Sui, Qitian Wu, Jiancan Wu, Qing Cui, Longfei Li, Jun Zhou, Xiang Wang, Xiangnan He.

In NeurIPS 2023: https://openreview.net/forum?id=hIGZujtOQv.

Installations

Main packages: PyTorch, Pytorch Geometric, OGB.

pytorch==1.10.1
torch-cluster==1.5.9
torch-geometric==2.0.3
torch-scatter==2.0.9
torch-sparse==0.6.12
torch-spline-conv==1.2.1
ogb==1.3.4

Preparations

Please download the graph OOD datasets and OGB datasets as described in the original paper. Create a folder dataset, and then put the datasets into dataset. Then modify the path by specifying --data_dir your/path/dataset.

Commands

We use the NVIDIA GeForce RTX 3090 (24GB GPU) to conduct all our experiments. To run the code on CMNIST, please use the following command:

CUDA_VISIBLE_DEVICES=$GPU python -u main_adv_syn_it.py \
--trails 10 \
--dataset cmnist \
--emb_dim 300 \
--epochs 100 \
--cau_gamma 0.5 \
--adv_gamma_node 1.0 \
--adv_gamma_edge 0.8 \
--adv_dis 0.2 \
--adv_reg 0.5 \
--cau_reg 1.0 \
--causaler_lr 0.001 \
--attacker_lr 0.005 \
--test_epoch 10 --data_dir $DATA_DIR

To run the code on Molbbbp, please use the following command:

CUDA_VISIBLE_DEVICES=$GPU python -u main_adv_mol_it.py \
--trails 10 \
--domain scaffold \
--dataset ogbg-molbbbp \
--epochs 100 \
--emb_dim 64 \
--cau_gamma 0.5 \
--adv_dis 0.5 \
--adv_reg 0.5 \
--cau_reg 0.5 \
--causaler_lr 0.001 \
--attacker_lr 0.001 --data_dir $DATA_DIR

To run the code on Motif, please use the following command:

CUDA_VISIBLE_DEVICES=$GPU python -u main_adv_syn_it.py \
--trails 10 \
--domain basis \
--dataset motif \
--epochs 100 \
--cau_gamma 0.5 \
--adv_gamma 1.0 \
--adv_gamma_edge 0.8 \
--adv_dis 0.2 \
--adv_reg 0.5 \
--cau_reg 1.0 \
--causaler_lr 0.001 \
--attacker_lr 0.005 \
--data_dir $DATA_DIR

To run the code on Molhiv, please use the following command:

CUDA_VISIBLE_DEVICES=$GPU python -u main_adv_mol_it.py \
--trails 10 \
--domain size \
--dataset hiv \
--epochs 100 \
--emb_dim 128 \
--cau_gamma 0.1 \
--adv_gamma_node 1.0 \
--adv_gamma_edge 1.0 \
--adv_dis 1.5 \
--adv_reg 0.5 \
--cau_reg 0.5 \
--causaler_lr 0.01 \
--attacker_lr 0.01 \
--data_dir $DATA_DIR

Citation

If you use our codes or checkpoints, please cite our paper:

@inproceedings{sui2023unleashing,
    title={Unleashing the Power of Graph Data Augmentation on Covariate Distribution Shift},
    author={Sui, Yongduo and Wu, Qitian and Wu, Jiancan and Cui, Qing and Li, Longfei and Zhou, Jun and Wang, Xiang and He, Xiangnan},
    booktitle={NeurIPS},
    year={2023},
    url={https://openreview.net/pdf?id=hIGZujtOQv}
}