johnhostetter / xsvmc-lib

An open source library that implements algorithms proposed to make support vector machine (SVM) predictions interpretable

Introduction

XSVMC-Lib is an open source library that implements algorithms proposed to make support vector machine (SVM) predictions interpretable.

Requirements

XSVMC-Lib requires Python 3.8+. Since XSVMC-Lib has been implemented as an extension of sklearn.svm.SVC, it also requires the SciKit-Learn package (python3 -m pip install sklearn).

Although they are not required by XSVMC-Lib, the following packages are necessary for running the examples:

• Numpy (python3 -m pip install numpy)
• Matploplib (python3 -m pip install matploplib)
• OpenCV (python3 -m pip install opencv-python)

Examples

To run an example, say digits_explanation.py, you may use the following commands:

cd /path/to/xsvmc-lib
python3 examples/digits_explanation.py

Datasets

Parts of the following datasets are used in several examples that illustrate the use of XSVMC-Lib.

• The Quick, Draw! Dataset
• The MNIST database of handwritten digits

Technical Information

The mathematical foundation of XSVMC-Lib can be found in

M. Loor and G. De Tré, "Contextualizing Support Vector Machine Predictions," International Journal of Computational Intelligence Systems, Volume 13, Issue 1, 2020, Pages 1483 - 1497, doi: 10.2991/ijcis.d.200910.002.

License

XSVMC-Lib is released under the Apache License, Version 2.0.

Citing

If you use XSVMC-Lib, please cite the following article:

M. Loor, A. Tapia-Rosero and G. De Tré, "An Open-Source Software Library for Explainable Support Vector Machine Classification," 2022 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), 2022, pp. 1-7, doi: 10.1109/FUZZ-IEEE55066.2022.9882731.

BibTeX

@INPROCEEDINGS{
    xsvmlib,  
    author={Loor, Marcelo and Tapia-Rosero, Ana and {De Tr\'{e}}, Guy},  
    booktitle={2022 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE)},   
    title={An Open-Source Software Library for Explainable Support Vector Machine Classification},   
    year={2022}, pages={1-7},  
    doi={10.1109/FUZZ-IEEE55066.2022.9882731}
}