UnRAvEL

This repository is the official implementation of .

Installation

A local copy of this repository can be made using the following command:

git clone https://github.com/adityasaini70/BO-LIME.git

After that, the dependencies of the codebase(Python v3.8.5) can be installed using the following command:

cd UnRAvEL/
pip install -r requirements.txt

Source code

The source code of the API can be found in the unravel folder. The description of files are as follows:

unravel/tabular.py: Implementation of the UnRAvEL algorithm for tabular datasets
unravel/image.py: Implementation of the UnRAvEL algorithm for image datasets
unravel/acquisition_util.py: Implementation of all the used acquisition functions: FUR, UCB and UR
unravel/kernel_util.py: Wrapper module for the used GP kernels
unravel/plot_util.py: Wrapper module for plotting utilities

Evaluation

Apart from the main evaluation script, the code used in our experiments can be found in the evaluation folder. The description of files are as follows:

evaluation script: The main script used to call all the evaluation modules
evaluation/explanation_evaluator.py: Implementation of all the stability and fidelity metrics used for evaluating our method against LIME and BayLIME
evaluation/blackbox_util.py: Wrapper module for all the used datasets and corresponding prediction models
evaluation/settings.py: Implementation of all the evaluation settings

Results

Sample Efficiency	Stability

Tutorials and usage

We're yet to add the documentation of the final API, till then please refer the following tutorial for learning about the implemented calls and methods

Support Vector Classifier on Breast Cancer Classification dataset

If you face any problem in running any module of this codebase, please feel free to contact us at {aditya18125, ranjitha}@iiitd.ac.in.

License

The API is distributed under the MIT License