ST-ReP

August 29, 2025 ยท View on GitHub

This the official implementation of ST-ReP: Learning Predictive Representations Efficiently for Spatial-Temporal Forecasting (AAAI 2025). [Paper]

Requirements

The code is built based on Python 3.9, PyTorch 1.13.1. You can install PyTorch following the instruction in PyTorch. For example:

pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117

After ensuring that PyTorch is installed correctly, you can install other dependencies via:

pip install -r requirements.txt

Data Preparation

Following the detail descriptions here.

Pre-Training and Downstream forecasting

The scripts for reproduction of ST-ReP and two simple baselines HL and Ridge are presented as follows.

PEMS04

python run.py --config_file PEMS04.yaml --modelid STReP --device cuda:0
python run.py --config_file PEMS04.yaml --modelid HL
python run.py --config_file PEMS04.yaml --modelid Ridge

PEMS08

python run.py --config_file PEMS08.yaml --modelid STReP --device cuda:0
python run.py --config_file PEMS08.yaml --modelid HL
python run.py --config_file PEMS08.yaml --modelid Ridge

Temperature

python run.py --config_file temperature2016.yaml --modelid STReP --device cuda:0
python run.py --config_file temperature2016.yaml --modelid HL
python run.py --config_file temperature2016.yaml --modelid Ridge

Humidity

python run.py --config_file humidity2016.yaml --modelid STReP --device cuda:0
python run.py --config_file humidity2016.yaml --modelid HL
python run.py --config_file humidity2016.yaml --modelid Ridge

SDWPF

python run.py --config_file SDWPF.yaml --modelid STReP --device cuda:0
python run.py --config_file SDWPF.yaml --modelid HL
python run.py --config_file SDWPF.yaml --modelid Ridge

CA

python run.py --config_file CA.yaml --modelid STReP --device cuda:0
python run.py --config_file CA.yaml --modelid HL
python run.py --config_file CA.yaml --modelid Ridge

Results will be saved in the EXP_results folder.

Acknowledgement

Thanks to the following inspiring research and their valuable codes.

Cite

If you find this project helpful, please cite us:

@article{STReP_2025,
title={ST-ReP: Learning Predictive Representations Efficiently for Spatial-Temporal Forecasting},
volume={39},
url={https://ojs.aaai.org/index.php/AAAI/article/view/33465},
DOI={10.1609/aaai.v39i12.33465},
number={12},
journal={Proceedings of the AAAI Conference on Artificial Intelligence},
author={Zheng, Qi and Yao, Zihao and Zhang, Yaying},
year={2025},
month={Apr.},
pages={13419-13427}
}