This repository provides all resources to reproduce the paper: PredDiff: Explanations and Interactions from Conditional Expectations
In particular, 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 - (tabular) MNIST (
mnist.ipynb
): (Interaction) relevances for a classifier trained on MNIST seen as a tabular dataset
Install dependencies from preddiff.yml
by running conda env create -f preddiff.yml
and activate the environment via conda activate preddiff