README.md

Spatio-Temporal Graph Few-Shot Learning with Cross-City Knowledge Transfer

Requirements

• torch >= 1.8.1
• numpy >= 1.20.3
• scikit-learn >= 0.24.2
• pytorch geometric >= 1.7.2
• pyaml
• scipy
• tqdm

Data

• Download the data from google drive

Model training & Testing

# Set the test dataset and model structure
CUDA_VISIBLE_DEVICES=[CUDA NUM] python main.py --test_dataset [FEW-SHOT DATASET] --model [ST-META MODEL]
# For example: Use GRU model to train a model, and test on PEMS-BAY datasets
CUDA_VISIBLE_DEVICES=0 python main.py --test_dataset pems-bay --model GRU

Citation

If you find this repository, e.g., the paper, code and the datasets, useful in your research, please cite the following paper:

@inproceedings{DBLP:conf/KDD/CrossCityTransfer22,
  author    = {Bin Lu and
               Xiaoying Gan and
               Weinan Zhang and
               Huaxiu Yao and
               Luoyi Fu and
               Xinbing Wang},
  title     = {Spatio-Temporal Graph Few-Shot Learning with Cross-City Knowledge Transfer},
  booktitle = {{KDD} '22: The 28th {ACM} SIGKDD Conference on Knowledge Discovery and Data Mining,
              Washington, DC, USA, August 14--18, 2022},
  publisher = {{ACM}},
  year      = {2022}
}