Awesome

Directed Graph Contrastive Learning

Paper | Poster | Supplementary

The PyTorch implementation of Directed Graph Contrastive Learning (DiGCL).

In this paper, we present the first contrastive learning framework for learning directed graph representation.

<center>Illustration of Laplacian perturbation</center> <center><img src="imgs/lp.png" alt="augmentation" style="zoom:50%;" /></center>

<center>Illustration of DiGCN model using Laplacian perturbation</center> <center><img src="imgs/framework.png" alt="framework" style="zoom:100%;" /></center>  

If you find our work useful, please considering citing

@article{tong2021directed,
  title={Directed Graph Contrastive Learning},
  author={Tong, Zekun and Liang, Yuxuan and Ding, Henghui and Dai, Yongxing and Li, Xinke and Wang, Changhu},
  journal={Advances in Neural Information Processing Systems},
  volume={34},
  year={2021}
}

Requirements

Our project is developed using Python 3.7, PyTorch 1.7.0 with CUDA10.2. We recommend you to use anaconda for dependency configuration.

First create an anaconda environment called DiGCL by

conda create -n DiGCL python=3.7
conda activate DiGCL

Then, you need to install torch manually to fit in with your server environment (e.g. CUDA version). For the torch and torchvision used in my project, run

conda install pytorch==1.7.0 torchvision==0.6.0 cudatoolkit=10.2 -c pytorch

Besides, torch-scatter and torch-sparse are required for dealing with sparse graph. For these two packages, please follow their official instruction torch-scatter and torch-sparse.

Other requirements can be set up through:

cd DiGCL
pip install -e .

Usage

cd code
python train_digcl.py --gpu_id 0 --dataset cora_ml --curr-type log
python train_digcl.py --gpu_id 0 --dataset citeseer

The --dataset argument can be one of [cora_ml, citeseer] and the --curr-type argument can be one of [linear, log, exp, fixed].

License

DiGCL is released under the MIT License. See the LICENSE file for more details.

Useful Links

We are grateful for the following enlightening works, which are also of great use in our work.

Graph Contrastive Learning Library for PyTorch: PyGCL
Graph Contrastive Learning with Adaptive Augmentation: GRACE and GCA
Graph Contrastive Learning with Augmentations: GraphCL
Our another supervised approach to process directed graphs: DiGCN
MagNet: A Neural Network for Directed Graphs: MagNet

Acknowledgements

The template is borrowed from Pytorch-Geometric benchmark suite. We thank the authors of following works for opening source their excellent codes. Pytorch-Geometric, Graph2Gauss, and GNN-benchmark.