MyMediaLite

Rating data with timestamps is now supported. There are two time-aware baselineline recommenders, TimeAwareBaseline and TimeAwareBaselineWithFrequencies (from the Netflix Grand Prize solution), and chronological splits. Some matrix functions are now implemented as extension methods, which makes them easier to use in many cases. The script download_movielens.sh has been updated because the download URLs have changed. Lots of improvements have been made under the hood.

MyMediaLite now needs Mono 2.8 or later (2.10.x is recommended, or .NET 4.0). Multi-core support has been added for several components: item recommendation evaluation, cross-validation for rating prediction and item recommendation, and distributed SGD training for rating prediction matrix factorization. Long int user and item IDs are now supported. Different modes for selecting candidate items now also work for cross-validation. Mean reciprocal rank (MRR) has been added as a new evaluation measure. Numerous improvements have been made to the documentation, API, commandline tools, and helper scripts.

API changes were made. In ISplit, IList is used instead of List for Train and Test, and NumberOfFolds is now a uint instead of an int. In Eval.Ratings and Eval.Items, DisplayResults was replaced by FormatResults. Recommendations for groups were added. New recommenders include WeightedBPRMF (which is used for KDD Cup 2011, track 2), and SoftMarginRankingMF (an item recommender inspired both by CofiRank, ranking loss, and BPR-MF, stochastic gradient descent learning). Numerous improvements were made in the documentation, command-line tools, and helper scripts.

A simplified API for item recommendation and matrix factorization, with nicer class names. MAE optimization has been merged as an option into the BiasedMatrixFactorization recommender. The rating prediction tool allows user-defined formatting of prediction output. The item prediction tool allows specification of the set of items to consider for evaluation: the items in the training set (default), a given set (via --relevant-items=FILE), the ones in the test set (--test-items), or only those items in both the training and the test set (--overlap-items). Save time by evaluation on a random subset of the users.

This release includes a prototype SOAP Web service for rating prediction. It is planned that most of MyMediaLite's functionality will be offered using RESTful services in the future. New recommenders were added: FactorWiseMatrixFactorization and BiasMatrixFactorizationMAE. The library (and the command-line tools) now offer an online evaluation protocol that uses incremental updates. This may be the last version to support Mono 2.6.