This repository implements an approach to curating naturally adversarial datasets via adversarial ordering presented in this paper accepted to ICCPS 2024.

Clone this repository.

git clone https://github.com/sfpugh/Naturally-Adversarial-Datasets
cd Naturally-Adversarial-Datasets

The paper results were generated on a Linux machine with Ubuntu 20.04 and Python 3.8. We provide a Dockerfile to construct an image to this specification. To build the docker image, use the following command:

docker build -t nad .

To create and run a container from this image nad, use the following command:

docker run -it nad

Prior to running the code, decompress the data data.tar.gz in the data directory using the following commands:

cd data
tar -xzvf data.tar.gz
cd ..

To apply the approach, run from main.py with appropriate arguments. All arguments are listed in main.py. For example, to run the approach on the spo2_hr_lo dataset using probabilistic labeler majority vote, you can use the following command:

python main.py --dataset spo2_hr_lo --default_pred 1 --pl majorityvote --evaluate

Finally, to reproduce Table III from the paper run generate_results.sh in the scripts/ directory.

cd scripts
bash generate_results.sh

Please cite this paper in your publications if this code helps your research.

@misc{pugh2023curating,
      title={Curating Naturally Adversarial Datasets for Learning-Enabled Medical Cyber-Physical Systems}, 
      author={Sydney Pugh and Ivan Ruchkin and Insup Lee and James Weimer},
      year={2023},
      eprint={2309.00543},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}