2026 Releases#

2026 builds on the foundation of 2025 to improve the ergonomics of data access, querying, and metrics, and clean up some APIs that seemed good at the time, while making it even easier to use LensKit for recommendation scenarios besides ID-based personalized recommendation.

There are no new major paradigm shifts, though — pipelines, datasets, and components work as they do in the 2025 series, but with more features, some rough corners polished off the interfaces, and hopefully fewer bugs.

2026.1.0#

This is the first major feature and update release of the 2026 series! Key highlights:

  • The Trainable protocol now has a method to query whether a model has been trained, instead of relying on every trainable to correctly implement skip-on-retrain logic.

  • Batch inference inputs are simplified and more explicit.

  • Cleanup of several deprecated interfaces, and simplifying the public module hierarchy.

  • Dependency updates with SPEC0.

  • FlexMF is nearing stability, and will be stabilized in an upcoming 2026.x point release. It also supports LightGCN.

  • Updated default embedding sizes to a power of 2 (64) throughout LensKit.

  • Improved metric interface to better support complex metric aggregations. Some automatic aggregation and analysis is no longer supported or will soon be removed (RunAnalysis is deprecated), in favor of clear metric collection within a single run and letting the user aggregate and summarize in whatever tool they prefer.

Additional changes are described here. See the milestone GitHub release for further change-level details.

Breaking Changes#

Major Breaking Changes#

These are the major changes that will affect users who are writing code to generate recommendations with LensKit, or implementing recommendation components.

  • Many submodules (such as most modules under lenskit.pipeline) have been renamed to be private modules. Code importing from their original locations will need to be updated to import from the higher-level module (🐞 947).

  • The Trainable protocol now has a separate method is_trained() to query if a component has been trained, and responsibility for skipping retraining of already-trained components when retrain is False has been moved from individual components to lenskit.pipeline.Pipeline.train() (🐞 1042, ⛙ 1044). Components that implement Trainable need to implement the new method as well.

    The pipeline builder will issue a warning if a component implements train() but not is_trained().

  • Default embedding sizes for all embedding-based models (matrix factorizers, etc.) have changed to be 64 instead of 50, with the exception of lenskit.implicit, which continues to default to Implicit’s defaults (🐞 846).

  • RunAnalysis is deprecated, and no longer supports explicit defaults or global metrics. Use MeasurementCollector instead. Its global metric columns have also changed in some cases.

  • LensKit now requires Python 3.12.5 or newer, along with NumPy 2.x, Pandas 2.3 or newer, and SciPy 1.13 or newer (see Dependency Versioning, ⛙ 954). Earlier versions of Python 3.12 have a bug in the standard library that affects LensKit.

  • Stopped providing wheels for macOS on Intel. Users who still need to run LensKit on Intel-based Macs should use the Conda packages (available in conda-forge and prefix.dev).

  • Removed the deprecated unrated and all-items candidate selectors, in favor of TrainingItemsCandidateSelector (🐞 935).

  • Several input types to batch inference are no longer supported, to reduce ambiguity and implicit behavior (🐞 958, ⛙ 1074).

Other Breaking Changes#

These changes are more minor, and won’t directly affect most users who are using LensKit components and metrics as-is.

  • Pipeline components and inputs now have restrictions on their names, and cannot have names beginning with _. See Component Names for details.

  • Pipeline configurations serialized with previous versions cannot be re-loaded in LensKit 2026, due to moves of module paths. Import path canonicalization (🐞 948) reduces the risk of such breakage in future releases. Handwritten configurations will often still work.

  • The AttributeSet class has been renamed to EntityAttribute (🐞 946), to better clarify its design and role.

  • We no longer publish 32-bit binary wheels.

  • Removed DecomposedMetric, as the Metric interface is now decomposed. All metrics based on listwise measurements or intermediate results should directly extend from Metric or ListMetric. (⛙ 983)

  • GlobalMetric has been removed. Code needing to compute global metrics should directly compute over item lists.

  • MeasurementCollector no longer supports defaults when adding metrics.

  • ListGini and ExposureGini now require an item vocabulary or dataset as input.

  • Removed the deprecated lenskit.training.IterativeTraining base class in favor of lenskit.training.UsesTrainer.

  • Component class members have been renamed to no longer use the Scikit-Learn pattern of trailing _ in their names.

  • Removed the lenskit.stats.argtopn function (🐞 833). lenskit.data.ItemList.top_n() now uses a Rust-accelerated top-N implementation (⛙ 1049).

  • lenskit.data.ItemList.to_arrow() no longer accepts mappings for its optional columns argument, sequences of names are used instead (⛙ 1052).

  • Removed the n option to predict_pipeline() (🐞 835).

  • Revised RecQuery to only accept keyword arguments, and removed the legacy user_items attribute.

New Features#

  • Added environment variables to TrainingOptions, along with the env_var() method to query them. This allows for open-ended configuration of training processes.

  • Added support for free-threaded Python, including binary distributions for Python 3.14t on Linux and macOS (🐞 916, ⛙ 1022).

  • Added support for parallel batch inference using thread pools on free-threaded Python (🐞 921, ⛙ 1025).

  • Added LightGCN support to FlexMF (🐞 1019, ⛙ 1033).

  • Added SLIM and fsSLIM to lenskit.knn (🐞 894, ⛙ 896).

Performance Changes#

  • lenskit.data.RelationshipSet.co_occurrances() is much faster and uses parallel computation, at the expense of increased memory use in the symmetric (non-ordered) case (🐞 970, ⛙ 1007).

Minor Changes#

  • lenskit.metrics.call_metric() has been renamed to measure_list(), and the old name preserved as a deprecated alias.

  • Pipeline type-checking for ArrayLike component inputs no longer works, due to a breaking change in NumPy 2.4. No LensKit components used ArrayLike as an input or output data type.

  • Pipeline component inputs with default values can now have missing inputs (🐞 1000, ⛙ 1001).

  • Rust progress updates now use a background thread to simplify logic and keep locks out of the work path (⛙ 1008).

  • Fixed a bug saving item list collections of emtpy item lists to Parquet files (🐞 1051, ⛙ 1052).