Release notes for LensKit 0.1
The first beta release! What this means for the future is that our core APIs (such as the recommenders and predictors) are pretty stable, and future changes to them will be documented and called out in the release notes. In general, future release notes will be more detailed, as we’re working on adding more features and stabilizing LensKit to a 1.0 release.
This release contains many changes. Highlights include:
Restructuring of recommendation and prediction interfaces to keep rating-based recommendation cleanly separated from more generic interfaces and facilitate reuse of predictor-based recommendation logic.
ScoredIdin favor of
ScoredLongListfor the recommendation API.
Added Slope-One recommenders.
Lots of bugfixes.
Increased parallelism in the evaluators (evaluators are now parallized per-configuration rather than per-dataset, decreasing memory requirements for parallel evaluation, and evaluation tasks are aggregated from all data sets into a single work queue and thread pool to keep your CPU cores pegged).
Allow arbitrary recommender components to be extracted with
getComponentmethod. This allows client code to examine things like the item-item similarity matrix.
More test cases and lots more API documentation.
Keep watching LensKit for further developments and lots of exciting new work this summer.