Home

Awesome

AGE

Source code and datasets for KDD 2020 paper "Adaptive Graph Encoder for Attributed Graph Embedding"


Requirements

Please make sure your environment includes:

python (tested on 3.7.4)
pytorch (tested on 1.2.1)

Then, run the command:

pip install -r requirements.txt

Run

Run AGE on Cora dataset:

python train.py --dataset cora --gnnlayers 8 --upth_st 0.011 --lowth_st 0.1 --upth_ed 0.001 --lowth_ed 0.5

To reproduce the node clustering experiment results, please follow our hyper-parameter settings:

Datasetgnnlayersupth_stupth_edlowth_stlowth_ed
Cora80.01100.00100.10.5
Citeseer30.00150.00100.10.5
Wiki10.00110.00100.10.5
Pubmed350.00130.00100.70.8

For link prediction, please run link_pred.py. We did not tune hyper-parameters for link prediction, so you can tune all kinds of hyper-parameters to get better performance.

Cite

If you use the code, please cite our paper:

@inproceedings{cui2020adaptive,
  title={Adaptive Graph Encoder for Attributed Graph Embedding},
  author={Cui, Ganqu and Zhou, Jie and Yang, Cheng and Liu, Zhiyuan},
  booktitle={Proceedings of SIGKDD 2020},
  year={2020}
}