java,plugins,prediction,collaborative-filtering,lenskit
First, the traditional way to do this is via cross-validation, where you do a robust randomized splitting of the data into training data and test data. The LensKit Evaluator supports doing this. The Quick Start descries how to get started; also there is a quick start that includes current best...
LensKit lead developer here. Sparse vectors and scored IDs are used to associate values with IDs (usually users or items). Side channels are a mechanism to associate additional values with each ID. They are extra data (an additional 'channel' of data) that goes with the 'primary' data. 'Typed' side channels...