As part of: MSDT2526; Built by: Group 2.
While explanation methods are widely available for black-box models, explanation tools for Bayesian networks, which are interpretable white-box models, remain scarce. Our project's goal is to address this gap in XAI and make explanations more helpful for interpretation, using explanation methods and visualisation.
BNXplainer, lets users upload a Bayesian network, set evidence variables and select a target feature. From there, inference results are computed and visualised through an inference diagram and a prediction table. To support explainability, we implemented three explanation methods: Value of Information (VOI), Most Probable Explanation (MPE) and Scenario Analysis. The feedback function lets users rate and reflect on the explanations they receive, contributing to the ongoing improvement of BNXplainer.
We hope our efforts contribute to address this gap and improving how Bayesian network predictions are understood and interpreted.
To create your venv use:
python -m venv venv
To activate it use:
Linux/Mac:
source venv/bin/activate
Windows:
venv\Scripts\activate
OR
venv\Scripts\Activate.ps1
To install the correct packages use:
pip install -r requirements.txt
To start the app use:
python src/app.py
It should open on http://127.0.0.1:8050/
To leave your environment use:
deactivate
Further reference: https://docs.python.org/3/library/venv.html
This script exports the feedback table from the SQLite database to a CSV file.
Usage
Run the script from the command line:
python src/utils/extract_csv.py
This creates:
db_export.csv
Custom Output Filename
You can provide a filename as an argument:
python src/utils/extract_csv.py feedback_export
This creates:
feedback_export.csv
We are a team of nine Radboud University students taking the course Modern Software Development Techniques, organised by the Artificial Intelligence Department (Donders Institute), under the coordination of Dr. Bryan Souza. Our client Dr. Marcos Buenos, Assistant Professor at Radboud University, commissioned this project on explainable AI (XAI) for white-box models: implementation and visualisation.
Please consult it for further reference.
Make sure to read CONTRIBUTING.md when working in git.
Architecture partly adapted from: https://community.plotly.com/t/structuring-a-large-dash-application-best-practices-to-follow/62739