ST-GAT

This is an implementation of Spatio-Temporal Graph Attention Network.

Datasets

The processed datasets are available at Google Drive.

https://drive.google.com/drive/folders/13uLQ14vehR0ztrlbNwXelE7fioejuEcb?usp=sharing

All datasets should be in './data'.

Environment

python 3.8.10
pytorch 1.9.0
numpy 1.19.5

Commands

# Train
python ST-GAT.py --mode=train --data=METR-LA --conf=./config/metr-la.conf

# Test
python ST-GAT.py --mode=test --data=METR-LA --saved_model=./out/metr-la/best_model --conf=./config/metr-la.conf

Experiments

Table: Training and inference time comparison.

Authors

Junho Song (junonjuno@bigdas.hanyang.ac.kr)
Jiwon Son (tinybeing@bigdas.hanyang.ac.kr)
Dong-hyuk Seo (hyuk125@bigdas.hanyang.ac.kr)
Kyungsik Han (kyungsikhan@hanyang.ac.kr)
Namhyuk Kim (namhyuk.kim@hyundai.com)
Sang-Wook Kim (wook@hanyang.ac.kr)

Cite

We encourage you to cite our paper if you have used the code in your work. You can use the following BibTex citation:

@inproceedings{song2022st,
  title={ST-GAT: A Spatio-Temporal Graph Attention Network for Accurate Traffic Speed Prediction},
  author={Song, Junho and Son, Jiwon and Seo, Dong-hyuk and Han, Kyungsik and Kim, Namhyuk and Kim, Sang-Wook},
  booktitle={Proceedings of the 31st ACM International Conference on Information \& Knowledge Management},
  pages={4500--4504},
  year={2022}
}