What’s new in 0.28

scikit-survival 0.28.0 (2026-07-05)

This release adds support for polars DataFrames and scikit-learn 1.9. The updated minimum supported version of dependencies are:

Package	Minimum Version
narwhals	2.0.1
scikit-learn	1.9.0

Enhancements

• Add support for polars DataFrame via narwhals to all estimators (#592).

• Add support for scikit-learn 1.9 (#593).

• Upgrade to Sphinx 8.2, pydata-sphinx-theme 0.18, and sphinx-design 0.7 (#595).

• Add output_type to sksurv.io.loadarff() to return pandas.DataFrame (default) or polars.DataFrame (#592).

• Add support to write polars.DataFrame with sksurv.io.writearff() (#592).

• Add support for polars.DataFrame to sksurv.kernels.clinical_kernel(), sksurv.kernels.ClinicalKernelTransform, and sksurv.preprocessing.OneHotEncoder (#592).

Bug Fixes

• Fix handling of pandas Boolean dtype in sksurv.kernels.ClinicalKernelTransform (#590).

Backwards incompatible changes

• Require at least narwhals 2.0.1.

• Require at least scikit-learn 1.9.0.

• Removed criterion from sksurv.ensemble.GradientBoostingSurvivalAnalysis (#593).