Awesome
[ICDE'2023] STWave
📖 Introduction
This is a official PyTorch implementation of the paper: When Spatio-Temporal Meet Wavelets: Disentangled Traffic Forecasting via Efficient Spectral Graph Attention Networks.
<img src="./stwave.png" align="middle" width="95%">⚡ Environment
- PyTorch
- fastdtw
- PyWavelets
🔧 Data Preparation
PeMSD3 & PeMSD4 & PeMSD7 & PeMSD8
- Download the data PeMSD* with code: p72z.
- Unzip them to corresponding folders.
PeMSD7(M) & PeMSD7(L)
-
Download the data PeMSD7(M).
-
Email authors of STGCN to get the data PeMSD7(L).
Tips
- The name of downloaded datasets should be consistent with the name in config files.
📂 Folder Structure
└── code-and-data
├── config # Including detail configurations
├── cpt # Storing pre-trained weight files (should be created)
├── data # Including adj files and the main data should be downloaded
├── lib
│ |── utils.py # Codes of preprocessing datasets and calculating metrics
│ |── graph_utils.py # Codes of calculating eigens and deriving the temporal graph
├── log # Storing log files (should be created)
├── model
│ |── models.py # The core source code of our STWave
├── mian.py # This is the main file for training and testing
└── README.md # This document
🚀 Run
Given the example of PeMSD8
mkdir ./cpt/PeMSD8
mkdir ./log/PeMSD8
python main.py --config config/PeMSD8.conf
💬 Citation
If you find our work is helpful, please cite as:
@inproceedings{fang2023spatio,
title={When spatio-temporal meet wavelets: Disentangled traffic forecasting via efficient spectral graph attention networks},
author={Fang, Yuchen and Qin, Yanjun and Luo, Haiyong and Zhao, Fang and Xu, Bingbing and Zeng, Liang and Wang, Chenxing},
booktitle={2023 IEEE 39th International Conference on Data Engineering (ICDE)},
pages={517--529},
year={2023},
organization={IEEE}
}
👍 Contributing
We welcome contributions and suggestions!