STDN (Spatial-Temporal Dynamic Network)

About

Source code of the paper Revisiting Spatial-Temporal Similarity: A Deep Learning Framework for Traffic Prediction

If you find this repository useful in your research, please cite the following paper:

@inproceedings{yao2019revisiting,
  title={Revisiting Spatial-Temporal Similarity: A Deep Learning Framework for Traffic Prediction},
  author={Yao, Huaxiu and Tang, Xianfeng and Wei, Hua and Zheng, Guanjie and Li, Zhenhui},
  booktitle={2019 AAAI Conference on Artificial Intelligence (AAAI'19)},
  year={2019} 
}

Installation

Requirements

• Python 3.6 (Recommend Anaconda)
• Ubuntu 16.04.3 LTS
• Keras >= 2.0.8
• tensorflow-gpu (or tensorflow) == 1.3.0 (install guide)

Usage

• Download all codes (*.py) and put them in the same folder (let's name it "stdn") (stdn/*.py)
• Create "data" folder in the same folder (stdn/data/)
• Create "hdf5s" folder for logs (if not exist) (stdn/hdf5s/)
• Download and extract all data files (*.npz) from data.zip and put them in "data" folder (stdn/data/*.npz)
• Open terminal in the same folder (stdn/)
• Run with "python main.py" for NYC taxi dataset, or "python main.py --dataset=bike" for NYC bike dataset

python main.py

python main.py --dataset=bike

• Check the output results (RMSE and MAPE). Models are saved to "hdf5s" folder for further use.

Hyperparameters:

Please check the hyperparameters defined in main.py