multiversum 0.7.2

A helper package to conduct multiverse analyses in Python

multiversum

multiversum is a package designed to make it easy to conduct multiverse analyses in Python. The package is intended to seemlessly integrate into a normal analysis or ML workflow and can also be added to an existing pipeline.

Features

• Simple 🧩: Built with the goal of being as simple as possible to integrate into existing workflows.
• Parallel 👯: Different universes are automatically evaluated in parallel.
• Notebooks 📓: Analyses can be written as Jupyter notebooks or Python scripts.
• Play/Pause ⏯️: Interrupt and then continue a multiverse analysis where you left it.

Installation

Install this library using pip:

pip install multiversum

Usage

The package always works with two different files: The multiverse.toml (or json / py) ✨️, specifying the different dimensions (and their options) and the universe.ipynb (or py) ⭐️ containing the actual analysis code. The universe file is then evaluated (in parallel) using different dimension-combinations, by running python -m multiversum.

An example using a machine learning workflow in scikit-learn can be found here.

Getting Started

To quickly get started with running your own multiverse analysis, you can use the multiversum starter template.

Examples

This section contains real-world examples of multiverse analyses conducted using this library.

• Preventing Harmful Data Practices by using Participatory Input to Navigate the Machine Learning Multiverse [Code] [Paper]
• Bias Begins with Data: The FairGround Corpus for Robust and Reproducible Research on Algorithmic Fairness [Code] [Paper]

Development

To contribute to this library, first checkout the code. Then create a new virtual environment:

cd multiversum
python -m venv venv
source venv/bin/activate

Now install the dependencies and test dependencies:

python -m pip install -e '.[test]'

To run the tests:

python -m pytest

Formatting

Ruff is used for formatting and linting. Formatting can be automatically checked / applied wherever possible via ruff check . --fix && ruff format.

Docs

The documentation is generated using mkdocs and can be locally run using mkdocs serve after installing its dependencies using python -m pip install -e '.[docs]'.