ST-Norm

This is a implementation of ST-Norm. The implementations of backbone Wavenet is cited from the published resource.

Requirements

Python 3.7
Numpy >= 1.17.4
Pandas >= 1.0.3
Pytorch >= 1.4.0

Model Training

python main.py --mode train --snorm 1 --tnorm 1

Arguments

model: backbone architecture (wavenet / tcn / transformer).
snorm: whether use spatial normalization.
tnorm: whether use temporal normalization.
dataset: dataset name.
version: version number.
hidden_channels: number of hidden channels.
n_pred: number of output steps.
n_his: number of input steps.
n_layers: number of hidden layers.

Model Evaluation

python main.py --mode eval --snorm 1 --tnorm 1

Citation

@inproceedings{deng2021st,
  title={ST-Norm: Spatial and Temporal Normalization for Multi-variate Time Series Forecasting},
  author={Deng, Jinliang and Chen, Xiusi and Jiang, Renhe and Song, Xuan and Tsang, Ivor W},
  booktitle={Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery \& Data Mining},
  pages={269--278},
  year={2021}
}

@ARTICLE{9935292,
  author={Deng, Jinliang and Chen, Xiusi and Jiang, Renhe and Song, Xuan and Tsang, Ivor W.},
  journal={IEEE Transactions on Knowledge and Data Engineering}, 
  title={A Multi-View Multi-Task Learning Framework for Multi-Variate Time Series Forecasting}, 
  year={2023},
  volume={35},
  number={8},
  pages={7665-7680},
  keywords={Task analysis;Time series analysis;Forecasting;Correlation;Multitasking;Transformers;Predictive models;Time series forecasting;deep learning;normalization;multi-view multi-task learning},
  doi={10.1109/TKDE.2022.3218803}}