In this repository, we implement the Graph Diffusion via the System of SDEs (GDSS) using Graph Transformer.

Paper: Score-based Generative Modeling of Graphs via the System of Stochastic Differential Equations (ICML 2022).

Original Code Repository: https://github.com/harryjo97/GDSS

Dependencies

Please create an environment with Python 3.9.15 and Pytorch 1.12.1, and run the following command to install the requirements:

pip install -r requirements.txt
conda install pyg -c pyg
conda install -c conda-forge graph-tool=2.45
conda install -c conda-forge rdkit=2022.03.2

Running Experiments

1. Preparations

We provide four general graph datasets (Planar and SBM) and two molecular graph datasets (QM9 and ZINC250k).

Download the datasets from the following links and move the dataset to data directory:

• Planar (planar_64_200.pt): https://drive.google.com/drive/folders/13esonTpioCzUAYBmPyeLSjXlDoemXXQB?usp=sharing

• SBM (sbm_200.pt): https://drive.google.com/drive/folders/1imzwi4a0cpVvE_Vyiwl7JCtkr13hv9Da?usp=sharing

We provide the commands for generating general graph datasets as follows:

python data/data_generators.py --dataset <dataset> --mmd

where <dataset> is one of the general graph datasets: planar and sbm. This will create the <dataset>.pkl file in the data directory.

To preprocess the molecular graph datasets for training models, run the following command:

python data/preprocess.py --dataset ${dataset_name}
python data/preprocess_for_nspdk.py --dataset ${dataset_name}

For the evaluation of generic graph generation tasks, run the following command to compile the ORCA program (see http://www.biolab.si/supp/orca/orca.html):

cd evaluation/orca 
g++ -O2 -std=c++11 -o orca orca.cpp

2. Training

We provide the commands for the following tasks: Generic Graph Generation and Molecule Generation.

To train the score models, first modify config/${dataset}.yaml accordingly, then run the following command.

CUDA_VISIBLE_DEVICES=${gpu_ids} python main.py --type train --config ${train_config} --seed ${seed}

for example,

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py --type train --config planar --seed 42

and

CUDA_VISIBLE_DEVICES=0 python main.py --type train --config qm9 --seed 42

3. Generation and Evaluation

To generate graphs using the trained score models, run the following command.

CUDA_VISIBLE_DEVICES=${gpu_ids} python main.py --type sample --config planar

or

CUDA_VISIBLE_DEVICES=${gpu_ids} python main.py --type sample --config sample_qm9

Pretrained checkpoints

We provide checkpoints of the pretrained models in the follwoing links:

• Planar: https://drive.google.com/drive/folders/18P_W6B-aBul_OFkIBsl9aPdT906CPDfZ?usp=drive_link
• SBM: https://drive.google.com/drive/folders/1LIUNf96IYefMfkospvbqmcvvPYBSukgP?usp=drive_link
• QM9: https://drive.google.com/drive/folders/1loFz_DIzt6JGAvUoB2zvTV9TvuX34A3G?usp=drive_link
• ZINC250k: https://drive.google.com/drive/folders/19WBDXDLph_QdA7T6MfEWkpmGPujpgPZ4?usp=drive_link

Citation

@article{jo2022GDSS,
  author    = {Jaehyeong Jo and
               Seul Lee and
               Sung Ju Hwang},
  title     = {Score-based Generative Modeling of Graphs via the System of Stochastic
               Differential Equations},
  journal   = {arXiv:2202.02514},
  year      = {2022},
  url       = {https://arxiv.org/abs/2202.02514}
}