CAL-plus

We provide a detailed code for CAL-plus.

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 HIV, please use the following command:

CUDA_VISIBLE_DEVICES=$GPU nohup  python -u  train_real.py \
--dataset hiv \
--domain size \
--beta=1 \
--lr=0.001 \
--min_lr 1e-6 \
--weight_decay 0.001 \
--hidden 300 \
--epoch 100 \
--batch_size 256 \
--trails 10 \
--memory True \
--prototype True \
--me_batch_n 3 >> ./log/hiv_size.log 2>&1 &

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

CUDA_VISIBLE_DEVICES=$GPU nohup  python -u ./molbbbp/train_real.py \
--dataset ogbg-molbbbp \
--domain size \
--beta 0.05 \
--co 0.1 \
--c 0.1 \
--o 1 \
--lr=0.001 \
--min_lr 1e-8 \
--weight_decay 5e-6 \
--hidden 64 \
--layers 2 \
--epochs 100 \
--pretrain 50 \
--batch_size 128 \
--trails 10 \
--memory True \
--prototype True \
--me_batch_n 1 >> ./log/molbbbp_size.log 2>&1 &

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

CUDA_VISIBLE_DEVICES=$GPU nohup  python -u train_real.py \
--dataset motif \
--domain size \
--beta 0.5 \
--lr 0.001 \
--min_lr 1e-3 \
--weight_decay 0 \
--hidden 128 \
--epoch 400 \
--batch_size 256 \
--trails 10 \
--memory True \
--prototype True \
--me_batch_n 5 >> ./log/motif_size.log 2>&1 &

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

CUDA_VISIBLE_DEVICES=$GPU nohup  python -u train_real.py \
--dataset cmnist \
--domain color \
--beta 1 \
--lr=0.001 \
--layers=5 \
--min_lr 1e-6 \
--weight_decay 0 \
--hidden 128 \
--epoch 200 \
--batch_size 256 \
--memory True \
--prototype True \
--me_batch_n 6 >> ./log/cmnist_color.log 2>&1 &

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

lr=0.002
min=5e-6
b=0.5

CUDA_VISIBLE_DEVICES=$GPU python -u main.py --bias $b --lr $lr --min_lr $min --model CausalGIN \
--beta 0.05 \
--memory True \
--prototype True \
--me_batch_n 3