• python 3
• see requirements.txt

Download the traffic data files for Los Angeles (METR-LA) and Bay Area (PEMS-BAY) from Google Drive or Baidu Yun links provided by DCRNN. Put into the data/{METR-LA,PEMS-BAY} folder.

Download Solar-Energy, Traffic, Electricity, Exchange-rate datasets from https://github.com/laiguokun/multivariate-time-series-data. Put into the data/{solar_AL,traffic,electricity,exchange_rate} folder.

Run the following commands to generate train/validation/test dataset at data/{METR-LA,PEMS-BAY,solar_AL,traffic,electricity,exchange_rate}/{train,val,test}.npz.

python generate_data.py 

• METR-LA

# Use METR-LA dataset
python train.py --dataset_dir=data/METR-LA --input_dim=2

• PEMS-BAY

# Use PEMS-BAY dataset
python train.py --dataset_dir=data/PEMS-BAY --input_dim=2

• Solar-Energy

# Use Solar-Energy dataset
python train.py --dataset_dir=data/solar_AL --input_dim=1

• Traffic

# Use Traffic dataset
python train.py --dataset_dir=data/traffic --input_dim=1

• Electricity

# Use Electricity dataset
python train.py --dataset_dir=data/electricity --input_dim=1

• Exchange-rate

# Use Exchange-rate dataset
python train.py --dataset_dir=data/exchange_rate --input_dim=1

If you use our model LSCGF in research, please cite this paper:

@inproceedings{chen2023balanced,
  title={Balanced spatial-temporal graph structure learning for multivariate time series forecasting: a trade-off between efficiency and flexibility},
  author={Chen, Weijun and Wang, Yanze and Du, Chengshuo and Jia, Zhenglong and Liu, Feng and Chen, Ran},
  booktitle={Asian Conference on Machine Learning},
  pages={185--200},
  year={2023},
  organization={PMLR}
}