AI4HealthUOL / preddiff-interactions

Public code repository accompanying "PredDiff: Explanations and Interactions from Conditional Expectations"

PredDiff: Explanations and Interactions from Conditional Expectations

This repository provides resources to reproduce results from the paper: PredDiff: Explanations and Interactions from Conditional Expectations

@article{bluecher2022,
title = {PredDiff: Explanations and Interactions from Conditional Expectations},
journal = {Artificial Intelligence},
volume = {312},
pages = {103774},
year = {2022},
doi = {https://doi.org/10.1016/j.artint.2022.103774},
author = {Stefan Blücher and Johanna Vielhaben and Nils Strodthoff},
}

We provide ready-to-run jupyter notebooks, which apply PredDiff on different datasets

• Synthetic regression (synthetic_dataset.ipynb): (Interaction) relevances for a regressor on the synthetic dataset discussed in the paper
• MNIST (mnist.ipynb): (Interaction) relevances for a classifier trained on MNIST seen as a tabular dataset
• NHANES (nhanes.ipynb): (Interaction) relevances for a classifier trained on the NHANES (mortality regression) dataset

Requirements

Install dependencies from pred_diff.yml by running conda env create -f pred_diff.yml and activate the environment via conda activate pred_diff