dmelis / interpretwoe

Machine Learning Interpretability based on the Weight-of-Evidence

Weight-of-Evidence Interpretability

Codebase accompanying the papers:

• Weight of Evidence as a Basis for Human-Oriented Explanations.
• From Human Explanation to Model Interpretability: A Framework Based on Weight of Evidence.

See the papers for technical details.

Dependencies

• python (>3.0)
• numpy
• scipy
• pandas
• scikit-learn
• matplotlib
• seaborn
• attrdict
• colored
• tqdm
• jupyterlab (optional - to run notebooks )
• pytorch (optional - to use with pytorch models: BETA)

Installation

It's highly recommended that the following steps be done inside a virtual environment (e.g., via virtualenv or anaconda).

Via Conda (recommended)

If you use [ana|mini]conda , you can simply do:

conda env create -f environment.yaml python=3.8
conda activate interpretwoe
conda install . # (optional: to import without needing to add path)

Via pip

pip install -r requirements.txt

Finally, install this package:

pip install .

How to use

For example usage, please see notebooks/WoE_UserStudy_Main.ipynb. That notebook has a self-contained full experimental setup, which we used for our user study.

Overall Code Structure

The main relevant code is in the following scripts in `src/':

• explainers.py - defines the Explanation, Explainer Classes, etc
• scoring.py - defines the explanation scoring function for choose class partitions
• woe.py - defines weight-of-evidence computation methods
• data.py - data loading functions
• classifiers.py - classification models
• woe_utils.py - misc utils used by woe models