What’s new in 0.25#

scikit-survival 0.25.0 (2025-08-21)#

This release adds support for scikit-learn 1.7 and overhauls the API documentation to improve clarity and consistency.

Enhancements#

  • Add support for scikit-learn 1.7 (#532).

  • Move tox configuration to pyproject.toml.

  • Add PEP 735 dependency groups for optional dependencies.

  • Modernize C++ syntax in the coxnet model, improving code clarity and maintainability (#526).

  • Add license-files field to pyproject.toml (PEP 639).

  • Add artifact attestation for sdist and wheel files.

  • Update CI infrastructure to use the latest runners and tools, including check-jsonschema, ruff, and uv.

  • Update CI infrastructure to use miniforge to avoid licensing issues related to Anaconda’s default channels (#542).

  • Add running doctest to CI.

  • Bump versions of dependencies on Binder.

Documentation#

  • Overhaul the entire API documentation for improved clarity, consistency, and user experience. This includes updated docstrings for all major modules, including ensemble, linear_model, svm, tree, metrics, and nonparametric (#539).

  • For examples with matplotlib plots, include the plot as a static image in the documentation (#543).

  • Clarify what inputs each metric expects and add a graphical overview to Evaluating Survival Models (#535).

  • Clarify the calculation of the deviance_ratio_ in sksurv.linear_model.CoxnetSurvivalAnalysis with a detailed mathematical definition (#541).

  • Standardize the description of the structured survival array y across the library.

  • Clarify that an exception is raised for out-of-range test times when the censoring distribution cannot be estimated (#524).

  • Explain how the alphas sequence is automatically generated in sksurv.linear_model.CoxnetSurvivalAnalysis.

  • Fix pandas warnings in example code.

  • Update links to external documentation, including scikit-learn and numpy.