I've released lme-rs 0.2.6, a Rust library for linear, generalized linear, and
nonlinear mixed-effects models. It offers lme4-style formulas, reusable prepared
designs, prediction, and post-fit inference; PyO3 bindings expose the same fitting
engine to Python.
The new release expands factor coding, marginal means, ANOVA/ANCOVA, and
null-model bootstrap workflows, and includes an optional Basin optimizer.
You can add lme-rs = "0.2.6" to Cargo.toml; Rust DataFrame consumers should
use Polars 0.46. Native linear-algebra build requirements are documented in the
repository.
I'd welcome feedback from people integrating statistical models into Rust tools:
API ergonomics, build portability, and small reproducible numerical failures are
particularly useful. The library is under active validation and the release
notes disclose known optimizer/accuracy limitations. No general performance
parity is claimed.