XdebugML

This repository contains the implementation source code of the following paper:

Explainable Debugger for Black-box Machine Learning Models

BibTeX:

@inproceedings{rasouli2021explainable,
               title={Explainable Debugger for Black-box Machine Learning Models},
               author={Rasouli, Peyman and Yu, Ingrid Chieh},
               booktitle={2021 International Joint Conference on Neural Networks (IJCNN)},
               pages={1--10},
               year={2021},
               organization={IEEE}
}

Setup

1- Clone the repository using HTTP/SSH:

git clone https://github.com/peymanrasouli/XdebugML

2- Create a conda virtual environment:

conda create -n XdebugML python=3.6

3- Activate the conda environment:

conda activate XdebugML

4- Standing in XdebugML directory, install the requirements:

pip install -r requirements.txt

5- Install the Accumulated Local Effects (ALE) package:

pip install git+https://github.com/MaximeJumelle/ALEPython.git@dev#egg=alepython

6- Install SHAP package:

pip install git+https://github.com/slundberg/shap.git

7- Run initial setup:

python setup.py

8- Install TBB library required by EXPLAN:

# Ubuntu/Debian
sudo apt-get update
sudo apt-get install libtbb2 

# CentOS
sudo yum update
sudo yum install tbb

Reproducing the results

1- To reproduce the neighborhood influence results run:

python neighborhood_influence.py

2- To reproduce the occurrence histogram results run:

python occurrence_histogram.py

3- To explain an instance using quasi-global explanation method run:

python quasi_global_explanation.py

4- To reproduce the diversity results of quasi-global explanations run:

python quasi_global_explanation_diversity.py

5- To explain an instance globally run:

python global_explanation.py

6- To visualize feature values vs. contribution values run:

python data_visualization.py