statsmodels 0.15.0

Release 0.15.0

on Python PyPI

The statsmodels developers are happy to announce the release of 0.15.0. 358 issues were closed in this release and 655 pull requests were merged. Major new features include:

• Standardized on rng for controlling randomness across the package (SPEC 007), replacing seed/random_state
• Functions with variable-length tuple returns now return documented NamedTuple results instead
• A new abstracted formula engine that supports both patsy and formulaic as the backend
• Support for Polars DataFrames and Series as model input
• Switched the build backend from setuptools to meson-python
• New robust estimators: CovDetMCD, CovDetS, CovDetMM, and RLMDetSMM
• New tests: Diebold-Mariano, Pesaran-Timmermann, Jonckheere-Terpstra, Leybourne-McCabe, and a delete-k block jackknife estimator
• The Hamilton filter, local false discovery rate correction, and an L1-penalized GLM solver
• HurdleCountModel gained fit_regularized, and MICEData is now iterable

This release also raises the minimum supported versions of NumPy, SciPy, and pandas, and tightens input validation for many string-valued options across the package (invalid values that previously failed silently or with a confusing error now raise a clear ValueError). A handful of long-standing bugs in seldom-exercised code paths were also corrected as part of a systematic coverage audit this cycle, some of which change numerical output for affected models. See the release notes for the complete list of enhancements, breaking changes, and bug fixes.