This is the original pytorch implementation of AutoCTS in the following paper: AutoCTS: Automated Correlated Time Series Forecasting.

This code is based on the implementation of PC-Darts.

Requirements

• python 3
• see requirements.txt

Data Preparation

AutoCTS is implemented on several public correlated time series forecasting datasets.

• METR-LA and PEMS-BAY from Google Drive or Baidu Yun links provided by DCRNN

• PEMS03, PEMS04, PEMS07 and PEMS08 from STSGCN (AAAI-20). Download the data STSGCN_data.tar.gz with password: p72z and uncompress data file usingtar -zxvf data.tar.gz To run AutoCTS on the PEMS03 dataset, you only need to download the PEMS03_data.csv and put it into the data/pems/PEMS03 folder.

• Solar-Energy and Electricity datasets from https://github.com/laiguokun/multivariate-time-series-data. Uncompress them and move them to the data folder.

Architecture Search

CUDA_VISIBLE_DEVICES=0 python3 train_search.py

Reference

If you use AutoCTS for your research, please cite the following paper.

     
@article{wu2022autocts,
  title={AutoCTS: Automated Correlated Time Series Forecasting},
  author={Xinle Wu and Dalin Zhang and Chenjuan Guo and Chaoyang He and Bin Yang and Christian S. Jensen},
  year={2022},
  pages={971--983}
  journal={Proceedings of the VLDB Endowment},
  volume={4}
}