Code for L-Shapley and C-Shapley in the paper L-Shapley and C-Shapley: Efficient Model Interpretation for Structured Data by Jianbo Chen, Le Song, Martin J. Wainwright, Michael I. Jordan.

Dependencies

The code runs with Python 2.7 and requires Tensorflow of version 1.1 or higher. Please pip install the following packages:

• numpy
• pandas
• keras
• tensorflow
• csv

Running in Docker, MacOS or Ubuntu

We provide as an example the source code to run CCM on the three synthetic datasets in the paper. Run the following commands in shell:

###############################################
# Omit if already git cloned.
git clone https://github.com/Jianbo-Lab/LCShapley
cd LCShapley/texts/
############################################### 
# L-Shapley
python explain.py --method localshapley

# C-Shapley
python explain.py --method connectedshapley

The importance scores for each word will be saved for the first 100 test samples in IMDB movie review.

Citation

If you use this code for your research, please cite our paper:

@inproceedings{
chen2018lshapley,
title={L-Shapley and C-Shapley: Efficient Model Interpretation for Structured Data},
author={Jianbo Chen and Le Song and Martin J. Wainwright and Michael I. Jordan},
booktitle={International Conference on Learning Representations},
year={2019},
}