This project is focused on developing a recurrent GAN architecture that can imitate and generate real-time biomedical time series.
In terms of methodologies, we are inspired by the RGAN and RCGAN (conditional RGAN) architecture proposed by Esteban, Hyland and Rätsch 2017. We aim to modify and further develop existing frameworks. The end goal of this project is to generate realistic biomedical time series which could enrich/mix salient medical features.
In terms of biomedical data, we aim to work with the existing MIMIC-III benchmarks which are documented in Harutyunyan, Khachatrian, Kale, Ver Steeg and Galstyan 2019. The MIMIC-III benchmark workflows can be found in the following public GitHub repository.
This repository's source code was tested with Python versions 3.7.*
and R versions 3.6.*
.
-
Install python dependencies located in
requirements.txt
:$ pip install -r requirements.txt
-
Install R-based dependencies:
> install.packages(c("ggplot2","tools","extrafont","reshape2","optparse","plyr"))
-
Optional: Install binary for adding progress bar to produced gif's.
-
Optional: To develop this repository, it is recommended to initialize a pre-commit hook for automatic updates of python dependencies:
$ ./init.sh
Our workflow and source code can be found in the src
directory of this repository. Additionally, the readme in the src
directory documents our functions, scripts and results.
A thorough development log for our ideas/progress can be found here.
Relevant BibTeX citations for the above-mentioned papers can be found below:
Harutyunyan et al. 2019
@article{Harutyunyan_2019,
title={Multitask learning and benchmarking with clinical time series data},
volume={6},
ISSN={2052-4463},
url={http://dx.doi.org/10.1038/s41597-019-0103-9},
DOI={10.1038/s41597-019-0103-9},
number={1},
journal={Scientific Data},
publisher={Springer Science and Business Media LLC},
author={Harutyunyan, Hrayr and Khachatrian, Hrant and Kale, David C.
and Ver Steeg, Greg and Galstyan, Aram},
year={2019},
month={Jun}
}
Esteban et al. 2017
@misc{esteban2017realvalued,
title={Real-valued (Medical) Time Series Generation with Recurrent Conditional GANs},
author={Cristóbal Esteban and Stephanie L. Hyland and Gunnar Rätsch},
year={2017},
eprint={1706.02633},
archivePrefix={arXiv},
primaryClass={stat.ML}
}
Atreya Shankar, German Research Centre for Artificial Intelligence (DFKI)