🚩 Updates (2025-1-22) Initial upload to arXiv PDF.

TimeFilter is a cutting-edge solution for time series forecasting, incorporating three main components: the Spatial-Temporal Construction Module, the Patch-Specific Filtration Module, and the Adaptive Graph Learning Module.

Ensure you are using Python 3.10.16 and install the necessary dependencies by running:

pip install -r requirements.txt

Begin by downloading the required datasets. All datasets are conveniently available at iTransformer. Create a separate folder named ./data and neatly organize all the csv files as shown below:

data
└── electricity.csv
└── ETTh1.csv
└── ETTh2.csv
└── ETTm1.csv
└── ETTm2.csv
└── traffic.csv
└── weather.csv
└── solar_AL.txt
└── PEMS03.npz
└── PEMS04.npz
└── PEMS07.npz
└── PEMS08.npz

All scripts are located in ./scripts. For instance, to train a model using the ETTh1 dataset with an input length of 96, simply run:

bash ./scripts/ETTh1.sh

After training:

• Your trained model will be safely stored in ./checkpoints.
• Numerical results in .npy format can be found in ./results.
• A comprehensive summary of quantitative metrics is accessible in ./result_long_term_forecast.txt.

If you find this repo useful, please consider citing our paper as follows:

@article{hu2025timefilter,
  title={TimeFilter: Patch-Specific Spatial-Temporal Graph Filtration for Time Series Forecasting},
  author={Hu, Yifan and Zhang, Guibin and Liu, Peiyuan and Lan, Disen and Li, Naiqi and Cheng, Dawei and Dai, Tao and Xia, Shu-Tao and Pan, Shirui},
  journal={arXiv preprint arXiv:2501.13041},
  year={2025}
}

Special thanks to the following repositories for their invaluable code and datasets:

• PatchTST
• DUET
• TimeSeriesCCM
• Time-Series-Library
• iTransformer

If you have any questions, please contact huyf0122@gmail.com or submit an issue.