Code for the EMNLP 2023 paper: MaNtLE: Model-agnostic Natural Language Explainer
Setup the environment and dependencies with the following command:
bash bin/init.sh
Next, each time you access this repository, make sure to run:
source bin/setup.sh
This allows the model to access the internal directories.
Link to download pre-trained model: Google Drive Link. Please place the contents at this link inside a folder named pretrained_mantle in the root of this repository.
Codes:
create_model_explanations.py parititons the dataset into train, val and test. It also trains a specified model and creates explanations for all the examples in the validation set.
create_subsets.py parititons the dataset into train, val and test. It also trains a specified model and creates explanations for all the examples in the validation set. It also converts the subset data into json files to be used with mantle.
mantle_explanations.py computes the mantle explanations.
evaluate_mantle.py evaluates mantle using the semantic parser. Note: parts of this code is still hard-coded for the adult dataset. But, this can be updated.
You can run MaNtLE for any of the datasets provided using the command: bash bin/run_dataset.sh {dataset_name}, where {dataset_name} is one of 'adult', 'recidivism', or 'travel_insurance'
For any doubts or questions regarding the work, please contact Rakesh (rrmenon@cs.unc.edu). For any bug or issues with the code, feel free to open a GitHub issue or pull request.
Please cite us if MaNtLE is useful in your work:
@inproceedings{menon2023mantle,
title={MaNtLE: Model-agnostic Natural Language Explainer},
author={Menon, Rakesh R and Zaman, Kerem and Srivastava, Shashank},
journal={Empirical Methods in Natural Language Processing (EMNLP)},
year={2023}
}