Less is More: Efficient Time Series Dataset Condensation via Two-fold Modal Matching (PVLDB 2025)

Citation

Please cite the following paper if this paper/repository is useful for your research.

@article{miao2024less,
  title={Less is more: Efficient time series dataset condensation via two-fold modal matching},
  author={Miao, Hao and Liu, Ziqiao and Zhao, Yan and Guo, Chenjuan and Yang, Bin and Zheng, Kai and Jensen, Christian S},
  journal={PVLDB},
  volume={18},
  number={2},
  pages={226--238},
  year={2024}
}

Running

• Data Preparation: Weather, Traffic, Electricity and ETT can be downloaded from Google Drive.

• Generating Expert Trajectories: Run each script in ./scripts_buffer/ to generate expert trajectories, for example

  sh ./scripts_buffer/weather.sh
  

• Time Series Dataset Condensation with TimeDC: After obtaining expert trajectories, run each script in ./scripts_distill/ to perform time series dataset condensation, for example

  sh ./scipts_distill/weather.sh
  

Requirements

python >= 3.8
Pytorch >= 1.11
numpy >=1.21.2
torchvision >=0.12