TGCRN

Time-aware Graph Structure Learning for Spatiao-temporal Forecasting

Requirements

• numpy >= 1.19.5
• pytorch == 1.2.0
• scipy == 1.4.1

Dependency environment can be installed using the following command:

pip install -r requirements.txt

Data preparation

The traffic data files for the Shanghai Metro, Hangzhou Metro, NYC-bike, and NYC-taxi are available at Google Drive and Baidu Drive. They should be put into the data/ corresponding folders.

Modeling training

cd ./model 

# HZMetro 
python run.py --dataset ../data/HZMetro --data HZ --lag 4 --horizon 4 --num_nodes 80

# SHMetro
python run.py --dataset ../data/SHMetro --data SH --lag 4 --horizon 4 --num_nodes 288

# Taxi
python run.py --dataset ../data/taxi --data taxi --lag 12 --horizon 12 --num_nodes 266

# Bike
python run.py --dataset ../data/bike --data bike --lag 12 --horizon 12 --num_nodes 250

Reference

Please cite our work if you find it useful.

@inproceedings{ma2024tgcrn,
  title={Learning time-aware graph structures for spatially correlated time series forecasting},
  author={Ma, Minbo and Hu, Jilin and Jensen, Christian S and Teng, Fei and Han, Peng and Xu, Zhiqiang and Li, Tianrui},
  booktitle={2024 IEEE 40th International Conference on Data Engineering (ICDE)},
  pages={4435--4448},
  year={2024}
}