TimeSeries_GAN_Review

A Survey and Taxonomy of recent GANs developments in time series.

Please refer to the details in our recent review paper “Generative Adversarial Networks in Time Series: A Survey and Taxonomy” available on ArXiv and submitted for peer-review. Authored by: Eoin Brophy, Zhengwei Wang, Qi She and Tomas E. Ward. We provide a list of papers related to GANs on time series.

While you're here, feel free to check out our main repository where we have compiled a master list of GAN developments across both Computer Vision and Time Series.

Citation

If you find this repo helpful in any way please consider citing our paper:

@misc{brophy2021generative,
  title={Generative adversarial networks in time series: A survey and taxonomy}, 
  author={Eoin Brophy and Zhengwei Wang and Qi She and Tomas Ward},
  year={2021},
  eprint={2107.11098},
  archivePrefix={arXiv},
  primaryClass={cs.LG}
}

We have classified the time series GAN-variants into two research lines based on recent GAN developments. We also provide a list of papers related to GANs on time series in the GAN_TS.csv file. Below, we provide a links to the code of novel methods listed in the review paper models.

Datasets

Unlike computer vision having lots of well-known and large-scale benchmarking datasets, time series benchmarking datasets are limited due to generalization and some privacy issues especially for clinical data. Below we provide some resources of well-known time series datasets. Hopefully it is useful. Feel free to suggest any well-known time series datasets to this repo by opening new issue. We will review it and add it to the list! We hope this can help push the time series research forward!

• Oxford-Man Institute realised library (Updated Daily). Real Multivariate Time Series dataset contains 2,689,487 instances and 5 attributes.
• EEG Motor Movement/Imagery Dataset (2004) . Real Multivariate Time Series contains 1,500 instances and 64 attributes.
• ECG 200 (2001). Real Univariate Time Series contains 200 instance and 1 attribute.
• Epileptic Seizure Recognition Dataset (2001). Real Multivariate Time Series dataset contains 11,500 instances and 179 attributes.
• TwoLeadECG (2015). Real Multivariate Time Series dataset contains 1,162 instances and 2 attributes.
• MIMIC-III Clinical Database (2016). Real, Integer & Categorical Multivariate Time Series.
• MIMIC-III Clinical Database Demo (2019). Real, Integer & Categorical Multivariate Time Series.
• EPILEPSIAE project database. Real Multivariate Time Series dataset contains 30 instances.
• PhysioNet/CinC. Lots of clinical data for challenging competition.
• Wrist PPG During Exercise (2017). Real Multivariate Time Series dataset contains 19 instances and 14 attributes.
• MIT-BIH Arrhythmia Database (2001). Real Multivariate Time Series dataset contains 201 instances and 2 attributes.
• KDD Cup Dataset. Lots of Real, Integer & Categorical Multivariate Time Series datasets.
• PeMS Database (updated daily). Real, Integer & Categorical Multivariate Time Series datasets.
• Nottingham Music Database. Special Text Format Time Series.

Discrete-variant GANs

• SeqGAN: TensorFlow; PyTorch
• Quant GAN: Code to be added

Continuous-variant GANs

• C-RNN-GAN: TensorFlow; PyTorch
• RCGAN: TensorFlow
• SC-GAN: Code to be added
• NR-GAN: Code to be added
• Time GAN: TensorFlow
• SigCWGAN: PyTorch
• DAT-CGAN: Code to be added
• SynSigGAN: Code to be added