2025 Releases#

2025 brought breaking changes to across the LensKit APIs to improve ergonomics, correctness-by-default, and flexibility. It also adopts SPEC0, a standard for supported versions of scientific Python libraries, and changes the LensKit version number scheme to “SemCalVer”. See Migrating from LensKit 0.x for information on how to upgrade your code.

2025.3.1#

LensKit 2025.3.1 is a significant feature and bugfix upgrade to the LensKit 2025 series. It maintains backwards compatibility with 2025.2 for stable interfaces (see Stability Levels), except as noted below. See the GitHub milestone for a full list of changes in this release; key changes are noted below.

Highlights#

  • Added Rust acceleration for several data processing operations and kNN models.

  • Added a new, experimental PyTorch matrix factorization model (see Flexible Matrix Factorization).

  • lenskit.basic now exports the configuration classes for basic algorithms (⛙ 672).

  • The pipeline runner now supports Pipeline Hooks to inspect or modify pipeline operations.

  • The pipeline runner type checking logic has been refactored and simplified. As a consequence, when None is provided to a component input that does not accept None, the runner now raises TypeError instead of PipelineError, as this is a type error. Details on a type error’s input wiring are now provided as a note on the exception, instead of in the main exception message.

  • Pipelines can now be specified in configuration files, and used from the command line.

  • Datasets can now have repeated or duplicate interactions (although such data sets are typically slow, see 🐞 869).

  • Experimental and mostly undocumented hyperparameter tuning support.

  • Added a configuration facility in lenskit.config to configure random numbers, power measurement, etc. Scripts using LensKit should call lenskit.configure().

  • Added support for querying power consumption from Prometheus. Documentation on how to set this up to be useful is TBD.

Compatibility Changes#

Important

These changes should not break most programs, but do introduce and document stricter requirements on certain names.

  • Pipeline component input names are not allowed to be prefixed with _, as such names are reserved for LensKit internal operation. This is not yet enforced, but will be enforced beginning in LensKit 2026.

  • Entity and attribute names are not allowed to be prefixed with _, and attribute names cannot end with _id or _num.

Component Changes#

  • ItemKNNScorer and UserKNNScorer are rewritten to use Rust acceleration, along with changes to its internal data representation to use Arrow instead of SciPy. This also fixes a segfault with very large similarity matrices.

    Note

    The model parameters of the KNN scorers have changed. They are no longer suffixed with _, and the similarity matrix is a PyArrow list array. Code that was directly examining internal elements will need to change.

  • The lenskit.als scorers have been similarly refactored, and had their learned parameters renamed for better consistency.

  • Replaced the broken SoftmaxRanker with a proper stochastic sampler (⛙ 667, StochasticTopNRanker). The old ranker will be removed in LensKit 2026.

  • Added lenskit.training.UsesTrainer for more sophisticated iterative training support.

  • Added lenskit.data.ItemList.top_n() to get the top-N values of an item list efficiently.

  • lenskit.data.Vocabulary is now backed by a Rust hashtable instead of a Pandas Index. An index view is still available.

Data Handling#

  • Added versioning to the native data format, documented data format compatibility, and added compatibility tests.

  • Added compressed sparse row extension types for Arrow, and use them in the LensKit native format (as well as Python/Rust data interchange) to more reliably handle CSR matrix data in Arrow (previously, we had to carry the matrix width or row dimension in side information; it is now embedded into the Arrow metadata).

  • Fix MovieLens import to detect movies without genres (🐞 727, ⛙ 738).

  • Parallel Processing now supports comma-separated lists for configuring parallelism within worker processes, and LK_NUM_CHILD_THREADS is now deprecated.

  • Added importers for UCSD Amazon data sets.

Evaluation#

  • Reworked the design of the Metric interface, along with metric accumulation for run measurement, to facilitate more types of metrics and more flexible use of the evaluation facilities. More breaking changes will come in LensKit 2026.

CLI#

Other Changes#

  • sample_negatives() now accepts "popular" as an alias for "popularity".

  • Several bug fixes for logging in niche setups (including ray clusters) (⛙ 673).

2025.2.0#

LensKit 2025.2.0 was released March 12, 2025.

Some small quality-of-life improvements (and removing invalid API compat).

2025.1.1#

LensKit 2025.1.1 was released March 7, 2025.

The changes in this release are too numerous and fundamental to fully document in traditional release notes. See the following for release update documentation:

  • Migrating from LensKit 0.x for conceptual changes and how to upgrade your code.

  • The notes below for behavior changes (e.g. new defaults, new metric capabilities), and small bits not covered in the migration guide.

  • The full changelog in the Git history and issue/PR milestone.

Breaking Changes#

LensKit 2025 has many breaking changes, with the migration guide (Migrating from LensKit 0.x) documenting the major ones. Below are some smaller ones not covered by that document:

  • Where Pandas data frames are still used, the standard user and item columns have been renamed to user_id and item_id respectively, with user_num and item_num for 0-based user and item numbers. This is to remove ambiguity about how users and items are being referenced.

  • The Popular recommender has been removed in favor of PopScore.

  • The DCG metric has been removed, as it is basically never used and was not useful as a part of the NDCG implementation.

New Features (incremental)#

  • Many LensKit components (batch running, model training, etc.) now report progress the progress API in lenskit.logging.progress, and can be connected to Jupyter or Rich.

  • Added RBP top-N metric (⛙ 334).

  • Added command-line tool to fetch datasets (⛙ 347).

Metric Behavior Changes#

Important

Some LensKit metric default has been changed; this results in values different from those computed by previous versions, either more correct or more consistent with common practice.

  • The NDCG metric now defaults to ignore rating values.

Model Behavior Changes#

Most models will exhibit some changes, hopefully mostly in performance, due to moving to PyTorch. There are some deliberate behavior changes in this new version, however, documented here.

  • ALS models only use Cholesky decomposition (previously selected with the erroneously-named method="lu" option); conjugate gradient and coordinate descent are no longer available. Cholesky decomposition is faster on PyTorch than it was with Numba, and is easier to maintain.

  • The default minimum similarity for UserUser is now \(10^{-6}\).

  • k-NN algorithms no longer support negative similarities; min_sim is clamped to be at least the smallest normal in 32-bit floating point (\(1.75 \times 10^{-38}\)).

  • The implicit bridge algorithms no longer look at rating values when they are present.

  • Bias is no longer optional for BiasedMFScorer and FunkSVD; both are inherently biased models, and FunkSVD is not commonly used.

  • lenskit.hpf.HPF no longer uses ratings as synthetic counts by default.

Bug Fixes#

Dependencies and Maintenance#

  • Bumped minimum supported dependencies as per SPEC0 (Python 3.11, NumPy 1.24, Pandas 2.0, SciPy 1.10).

  • Added support for Pandas 2 (⛙ 364) and Python 3.12.

  • Improved Apple testing to include vanilla Python and Apple Silicon (⛙ 366).

  • Updated build environment, dependency setup, taskrunning, and CI to more consistent and maintainable.

  • Removed legacy random code and SeedBank usage in favor of SPEC 7 (see Random Seeds).

  • Code is now auto-formatted with Ruff.