Release 0.6

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

• Fixed a bug where FunkSVD crashed when predicting for a user not in the model.
• Added history summarizers to compute vector summaries of user histories. This paves the way for non-rating-based scorers.
• Updated to gl-common 0.4, including the new citation JavaDoc taglet. More citations in the JavaDoc to come.
• Added version of MutableSparseVector.freeze() that strips NaN values in the frozen vector.
• Fixed a couple bugs in the user-user rating predictor with how baselines were added and what happens when the total similarity for an item is 0.
• Build our first basket recommender (see #3). It isn't tested in basket configuration yet, but item-item CF is now sufficiently modular to operate in unary contexts.
  • Added NeighborhoodScorer abstract scoring items based on neighborhoods. WeightedAverageNeighborhoodScorer implements the traditional weighted average algorithm that has been in item-item CF since we implemented it, and SimilaritySumNeighborhoodScorer just sums similarities (useful for unary contexts).
  • Split ItemItemRatingPredictor apart into ItemItemScorer, which uses arbitrary histories and NeighborhoodScorers to score items, and ItemItemRatingPredictor which configures it to be a rating predictor.