README.md

April 3, 2023 ยท View on GitHub

LightCTS: A Lightweight Framework for Correlated Time Series Forecasting

This is the repository of the paper entitled "LightCTS: A Lightweight Framework for Correlated Time Series Forecasting", encompassing the code, datasets, and supplemental material.

Supplemental Material

Detailed time and space complexity analysis and other groups of experimental results can be found at the Supplemental Material (downloading to local pdf viewer is recommended for better readability).



Code and Datasets

Requirements

To install requirements:

pip3 install -r requirements.txt

Multi-step Forecasting

Datasets

LightCTS is implemented on four public multi-step correlated time series forecasting datasets.

  • PEMS04, PEMS08, METR-LA, and PEMS-BAY can be downloaded in Google Drive. Please move them to the corresponding data folders.

Baselines

ModelConferenceYearLink
DCRNNICLR2018https://openreview.net/pdf?id=SJiHXGWAZ
GWNETIJCAI2019https://dl.acm.org/doi/10.5555/3367243.3367303
AGCRNNeurIPS2020https://dl.acm.org/doi/abs/10.5555/3495724.3497218
MTGNNKDD2020https://dl.acm.org/doi/abs/10.1145/3394486.3403118
AUTOCTSVLDB2021https://dl.acm.org/doi/10.14778/3503585.3503604
EnhanceNetICDE2021https://ieeexplore.ieee.org/abstract/document/9458855
FOGSIJCAI2022https://www.ijcai.org/proceedings/2022/545

Model Training and Testing

  • PEMS04/PEMS08
cd Multi-step/Traffic Flow/{dataset_name}/
python train_{dataset_name_in_lowercase}.py
python test_{dataset_name_in_lowercase}.py -checkpoint {path_to_the_checkpoint_file}
  • METR-LA/PEMS-BAY
cd Multi-step/Traffic Speed/{dataset_name}/
python train_{dataset_name_in_lowercase}.py
python test_{dataset_name_in_lowercase}.py -checkpoint {path_to_the_checkpoint_file}

Single-step Forecasting

Datasets

LightCTS is implemented on two public single-step correlated time series forecasting datasets.

  • Solar and Electricity datasets can be downloaded in Google Drive. Please move them to the corresponding data folders.

Baselines

ModelConferenceYearLink
DSANetCIKM2019https://dl.acm.org/doi/abs/10.1145/3357384.3358132
MTGNNKDD2020https://dl.acm.org/doi/abs/10.1145/3394486.3403118
AUTOCTSVLDB2021https://dl.acm.org/doi/10.14778/3503585.3503604
MAGNNarXiv2022https://arxiv.org/abs/2201.04828

Model Training and Testing

  • Solar-Energy/Electricity
cd Single-step/{dataset_name}/
python train_{dataset_name_in_lowercase}.py
python test_{dataset_name_in_lowercase}.py -checkpoint {path_to_the_checkpoint_file}

Pre-trained checkpoint files

Pre-trained checkpoint files can be download from Google Drive. Please replace "args.checkpoint" with the corresponding path in the test code file.

Contact

The code might be a little messy. For any question, feel free to contact: ``` Zhichen Lai: zhla@cs.aau.dk ```