This is the accompanying repository of our paper "Temporal Rule-Based Counterfactual Explanations for Multivariate Time Series" presented at IEEE ICMLA 22.
This code requires the installation of our RuleTransform package.
python3.6.13 is required. I suggest using conda to create the virtual environment:
conda create -n rt python=3.6.13
or alternatively, install python3.6.13 and:
python3.6.13 -m venv ./rt
source venv/bin/activate
Then:
pip install git+https://github.com/omarbahri/RuleTransform
The BasicMotions dataset is uploaded to the data directory. The other UEA datasets can be downloaded here.
terce.sh runs TeRCE on the BasicMotions dataset as described in the paper. Feel free to experiment with different datasets and parameters.
chmod +x terce.sh
./terce.sh
For large datasets, and depending on the time contract, parts of TeRCE might take longer to run. terce.sh keeps intermediary results to allow reusing them if needed.
@INPROCEEDINGS{10069254,
author={Bahri, Omar and Li, Peiyu and Boubrahimi, Soukaina Filali and Hamdi, Shah Muhammad},
booktitle={2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)},
title={Temporal Rule-Based Counterfactual Explanations for Multivariate Time Series},
year={2022},
volume={},
number={},
pages={1244-1249},
doi={10.1109/ICMLA55696.2022.00200}}
}