Dependencies

Option 1:

Please setup the environment following Requirements in this repository. Typically, you might need to run the following commands:

pip install torch==1.4.0
pip install torch-scatter==1.1.0 -f https://pytorch-geometric.com/whl/torch-1.4.0.html
pip install torch-sparse==0.4.4 -f https://pytorch-geometric.com/whl/torch-1.4.0.html
pip install torch-cluster==1.4.5 -f https://pytorch-geometric.com/whl/torch-1.4.0.html
pip install torch-spline-conv==1.1.0 -f https://pytorch-geometric.com/whl/torch-1.4.0.html
pip install torch-geometric==1.1.0
pip install torch-vision==0.5.0

Option 2:

You also can create a conda environment with https://github.com/Shen-Lab/GraphCL/blob/master/semisupervised_TU/environment.yml with the command:

conda env create -f environment.yml
conda activate graphcl

Then, you need to create two directories for pre-trained models and finetuned results to avoid errors:

cd ./pre-training
mkdir models
mkdir logs
cd ..
cd ./funetuning
mkdir logs
cd ..

Exploring the Role of Data Augmentation in GraphCL

Reproducing the results in the figure through executing the followings:

Pre-training:

cd ./pre-training
./run_all.sh $DATASET_NAME 0 $GPU_ID
./run_all.sh $DATASET_NAME 5 $GPU_ID
./run_all.sh $DATASET_NAME 10 $GPU_ID
./run_all.sh $DATASET_NAME 15 $GPU_ID
./run_all.sh $DATASET_NAME 20 $GPU_ID

Finetuning:

cd ./funetuning
./run_all.sh $DATASET_NAME 0 $EVALUATION_FOLD $GPU_ID
./run_all.sh $DATASET_NAME 5 $EVALUATION_FOLD $GPU_ID
./run_all.sh $DATASET_NAME 10 $EVALUATION_FOLD $GPU_ID
./run_all.sh $DATASET_NAME 15 $EVALUATION_FOLD $GPU_ID
./run_all.sh $DATASET_NAME 20 $EVALUATION_FOLD $GPU_ID

$DATASET_NAME is the dataset name (please refer to https://chrsmrrs.github.io/datasets/docs/datasets/), $EVALUATION_FOLD can be 10 or 100 for k-fold evaluation, and $GPU_ID is the lanched GPU ID.

The scripts will run 5*5=25 experiments (e.g. ./run_all.sh $DATASET_NAME 0 $GPU_ID will run 5 experiments marked with numbers 0~4 sequentially) and the final results will be recorded in ./funetuning/logs.

GraphCL with Sampled Augmentations

Take NCI1 as an example:

Pre-training:

cd ./pre-training
CUDA_VISIBLE_DEVICES=$GPU_ID python main.py --dataset NCI1 --aug1 random2 --aug2 random2 --lr 0.001 --suffix 0
CUDA_VISIBLE_DEVICES=$GPU_ID python main.py --dataset NCI1 --aug1 random2 --aug2 random2 --lr 0.001 --suffix 1
CUDA_VISIBLE_DEVICES=$GPU_ID python main.py --dataset NCI1 --aug1 random2 --aug2 random2 --lr 0.001 --suffix 2
CUDA_VISIBLE_DEVICES=$GPU_ID python main.py --dataset NCI1 --aug1 random2 --aug2 random2 --lr 0.001 --suffix 3
CUDA_VISIBLE_DEVICES=$GPU_ID python main.py --dataset NCI1 --aug1 random2 --aug2 random2 --lr 0.001 --suffix 4

Finetuning:

cd ./funetuning
CUDA_VISIBLE_DEVICES=$GPU_ID python main_cl.py --dataset NCI1 --aug1 random2 --aug2 random2 --semi_split 100 --model_epoch 100 --suffix 0
CUDA_VISIBLE_DEVICES=$GPU_ID python main_cl.py --dataset NCI1 --aug1 random2 --aug2 random2 --semi_split 100 --model_epoch 100 --suffix 1
CUDA_VISIBLE_DEVICES=$GPU_ID python main_cl.py --dataset NCI1 --aug1 random2 --aug2 random2 --semi_split 100 --model_epoch 100 --suffix 2
CUDA_VISIBLE_DEVICES=$GPU_ID python main_cl.py --dataset NCI1 --aug1 random2 --aug2 random2 --semi_split 100 --model_epoch 100 --suffix 3
CUDA_VISIBLE_DEVICES=$GPU_ID python main_cl.py --dataset NCI1 --aug1 random2 --aug2 random2 --semi_split 100 --model_epoch 100 --suffix 4

Five suffixes stand for five runs (with mean & std reported), and augmentations could be random2, random3, random4 that sampling from {NodeDrop, Subgraph}, {NodeDrop, Subgraph, EdgePert} and {NodeDrop, Subgraph, EdgePert, AttrMask}, seperately.

lr in pre-training should be tuned from {0.01, 0.001, 0.0001} and model_epoch in finetuning (this means the epoch checkpoint loaded from pre-trained model) from {20, 40, 60, 80, 100}.

Acknowledgements

The backbone implementation is reference to https://github.com/chentingpc/gfn.