/DSN

[NeurIPS 22'] Dynamic Sparse Network for Time Series Classification: Learning What to “See”

Primary LanguagePython

Dynamic Sparse Network for Time Series Classification: Learning What to “See”

This is the offical implementation for the paper titled Dynamic Sparse Network for Time Series Classification: Learning What to “See”.

Requirements

  • PyTorch 1.4.0
  • torchvision 0.2.1
  • numpy
  • pandas

Dataset

We truly appreciate everyone who worked on making the datasets available and their contributions to the TSC community.

  • The univariate time series datasets (UCR 85 Archive) could be found here

  • The multivariate time series datasets (UEA 30 Archive) could be found here

  • Datasets from UCI could be found here

Training

To train models for UCR 85 Archive, change the value of --root (e.g., UCR_TS_Archive_2015) and run this command:

python trainer_DSN.py --sparse True --density 0.2 --sparse_init remain_sort --fix False --growth random --depth 4 --ch_size 47 --c_size 3 --k_size 39

To train models for UCI datasets, change the value of --root (e.g., UCI) and run this command:

python trainer_DSN.py --sparse True --density 0.2 --sparse_init remain_sort --fix False --growth random --depth 4 --ch_size 47 --c_size 3 --k_size 39

To train models for UEA 30 Archive, change the value of --root (e.g., UEA_TS_Archive_2018) and run this command:

python trainer_DSN.py --sparse True --density 0.1 --sparse_init remain_sort --fix False --growth random --depth 4 --ch_size 59 --c_size 3 --k_size 39

Acknowledgements

We appreciate the following github repos a lot for their valuable code.

Citation

@inproceedings{
xiao2022dynamic,
title={Dynamic Sparse Network for Time Series Classification: Learning What to {\textquotedblleft}See{\textquotedblright}},
author={Qiao Xiao and Boqian Wu and Yu Zhang and Shiwei Liu and Mykola Pechenizkiy and Elena Mocanu and Decebal Constantin Mocanu},
booktitle={Advances in Neural Information Processing Systems},
year={2022},
url={https://openreview.net/forum?id=ZxOO5jfqSYw}
}