Model-agnostic variable importance for predictive uncertainty: an entropy-based approach

This repository contains the code for the experiments found in the paper "Model-agnostic variable importance for predictive uncertainty: an entropy-based approach".

Installation

This code is intended to be run using poetry. In order to set up the environment and install the requirements, run

poetry shell
poetry update
poetry install

All notebooks should run inside the environment. Notebooks to recreate specific experiments are in the relevant directories, with common containing Python files which define functions which are used across multiple notebooks.

Project Structure

Code for functions that are used in multiple experiments can be found in the common directory. Notebooks for each experiment are contained in directories named according to the kind of dataset used (real/synthetic, regression/classification). Notebook titles reference the section of the paper with which the experiments within the notebook correspond.