/CSDI

Codes for "CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation"

Primary LanguageJupyter NotebookMIT LicenseMIT

CSDI

This is the github repository for the NeurIPS 2021 paper "CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation".

Requirement

Please install the packages in requirements.txt

Preparation

Download the healthcare dataset

python download.py physio

Download the air quality dataset

python download.py pm25

Experiments

training and imputation for the healthcare dataset

python exe_physio.py --testmissingratio [missing ratio] --nsample [number of samples]

imputation for the healthcare dataset with pretrained model

python exe_physio.py --modelfolder pretrained --testmissingratio [missing ratio] --nsample [number of samples]

training and imputation for the healthcare dataset

python exe_pm25.py --nsample [number of samples]

Visualize results

'visualize_examples.ipynb' is a notebook for visualizing results.

Acknowledgements

A part of the codes is based on BRITS and DiffWave

Citation

If you use this code for your research, please cite our paper:

@inproceedings{tashiro2021csdi,
  title={CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation},
  author={Tashiro, Yusuke and Song, Jiaming and Song, Yang and Ermon, Stefano},
  booktitle={Advances in Neural Information Processing Systems},
  year={2021}
}