I’ve released lme-python 0.2.6, the Python bindings for my Rust statistical library,
lme-rs. It supports linear, generalized linear, and nonlinear mixed-effects models
with lme4-style formulas and Polars, pandas, or Arrow inputs.
This release adds explicit treatment/sum factor coding, configurable marginal-mean
grids, model-based ANOVA/ANCOVA tools, and null-model likelihood-ratio bootstrap
support. Binary wheels cover CPython 3.10–3.14 on Windows x64, Linux
x86_64/aarch64, and macOS x86_64/aarch64.
Try it with python -m pip install "lme-python==0.2.6"; the import is lme_python.
The quick start uses synthetic data
and needs no R installation when installing a compatible wheel.
I’m looking for reproducible examples from real analysis workflows: installation
problems, unsupported formulas, convergence issues, surprising inference, and
differences from reference packages. The library is still under validation.
Our retained comparisons include numerical failures, especially near covariance
boundaries; it is not a drop-in-equivalence claim.