Formulae is a Python library that implements Wilkinson’s formulas for mixed-effects models. It was written to make group-specific effects convenient in Bambi, but it can also be used independently as a backend for another library. Formulae extends classical statistical formulas in a way inspired by R’s lme4.
Dependencies
Formulae requires Python 3.8 or later, plus the versions of NumPy, SciPy, and Pandas specified in pyproject.toml.
Installation
Install the latest release with pip:
pip install formulae
Or install the development version directly from GitHub:
Formulae was built to provide Bambi with a concise syntax for mixed-effects models. Before Formulae, Bambi used Patsy to parse formulas and build design matrices. Patsy is flexible and robust, but does not support mixed model effects. Formulae therefore adopted the familiar | syntax from lme4 and builds the common and group-specific design matrices Bambi needs.
Its main function, design_matrices(), returns a wrapper around the response, common-effects, and group-effects matrices, along with methods and metadata useful to downstream libraries.
Formulae draws on the following projects and references:
R, a widely used implementation of Wilkinson’s formulas.
lme4, for the | operator and mixed effects matrix construction.
Patsy, especially its evaluation-environment implementation.
Formulaic, whose backtick and quote operators inspired Formulae’s equivalents.
If you only need design matrices for linear models with fixed effects, Patsy or Formulaic may be a more direct fit. Formulae is designed primarily for developers and libraries that need to construct design matrices automatically for mixed-effects models.
