Bossanova

Unified mixed-effects modeling in Python. JAX + NumPy backends. R/lme4 parity.

Installation

uv add bossanova

or

pip install bossanova

Features

• One unified entry point — model() handles linear, generalized, and mixed-effects models
• R-style formulas — y ~ x + (1|group), factor(), center(), scale()
• Multiple inference strategies — asymptotic, bootstrap, permutation, cross-validation
• Marginal effects — .explore() for estimated marginal means, slopes, and contrasts
• JAX acceleration — automatic acceleration when JAX is available, NumPy fallback
• R parity — matches lme4/emmeans within f64 tolerance

Quick Example

from bossanova import model, load_dataset

df = load_dataset("sleepstudy")

# Fit a mixed model
m = model("Reaction ~ Days + (1 + Days | Subject)", data=df).fit()
m.params

# Perform bootstrapped parameter inference
m.infer(how='boot')

# Explore marginal effects
m.explore("Days")
m.effects

Development

# Install dependencies
pixi install

# Install the IPython kernel (required for notebooks)
pixi run setup-kernel

# Run the full lint + type-check chain
pixi run lint

# Run fast unit tests
pixi run test