Release notes for LensKit 2.2

The Git changelog and the list of closed tickets and pull requests provide more information on what has happened, including bugs that have been fixed.

General Changes

Fast iteration is now a no-op, and some algorithms have been refactored to maintain performance.

Add KnownRatingRatingPredictor, a rating predictor that just returns the user’s existing ratings (#563).
The item-item CF implementation’s model representation has been changed to be more efficient without fast iteration.

Added ItemNameDAO to provide a standard means of presenting names or short descriptions of items, along with a memory-backed implementation and a provider to read it from a file.
Added support to TextEventDAO for skipping header lines. The number of header lines to skip are specified by the EventFormat; DelimitedColumnEventFormat has a setter to specify to skip some header lines (e.g. 1 line for a CSV file with headers).
Added pre-defined format Formats.movieLensFormat() for the new MovieLens data sets.
The Like and LikeBatch event types provide a standard type for unary and count data in LensKit.

Added global-recommend command to do global (non-personalized) recommendation with reference items.
The recommend, global-recommend, and predict commands use the ItemNameDAO, if available, to print item names (e.g. movie titles).
The recommend, global-recommend, and predict commands accept a --item-names command to specify a CSV file mapping item IDs to item names.
Fixed configuration loading to properly handle custom classpaths.
The --help output is now more thorough.

Incompatible change: the default holdout for addDataset methods on SimpleEvaluator is now the same as the default on CrossfoldTask (fixed holdout of 10, instead of the 20% previously used). This change was made to make SimpleEvaluator and CrossfoldTask have consistent defaults.
Fixed a bug in the component cache that caused the recommender build to be deferred to test time, making timings irrelevant, if the component cache was disabled (an important piece of making timings relevant). Fixed in PR 692.
Outputting recommendations and predictions is now handled by the recommendations and predictions metrics, rather than the old predictFile and related settings. This allows the recommendation output to be configured (candidate items, exclude set, and # of items to recommend).
Added a Mean Average Precision (MAP) metric (topNMAP).
Fixed broken top-N metric files.

LenskitRecommenderEngine now has a getComponent() method to allow a component to be extracted from the model without needing to first create a recommender object.