Probabilistic Prototype-Based Self-Explainable Neural Networks (Prob-PSENNs)

This repository contains the official implementation of the paper Uncertainty-Aware Explanations Through Probabilistic Self-Explainable Neural Networks. ArXiv Preprint

Setup

sh setup/prepare_setup.sh

The main requirements of the project are listed in the file requirements.txt, and can be installed using the following command:

pip install -r setup/requirements.txt

The remaining dependencies and resources are either included in this repository or downloaded using the source code.

Implementation of Prob-PSENNs

Training

The main code used to train the Prob-PSENNs introduced in the paper is contained in the file train.py. In order to assist the execution of this program and the configuration of the hyperparameters, we included a "launcher" script, which can be run as follows:

python3 launch/train_ProbPSENN.py

Evaluation

To evaluate the trained models, run:

python3 eval.py

by specifying the required configuration at the beginning of the script. This script will provide a summary of the effectiveness of the model, and reproduce the figures reported in the paper.