The tree-based ones are very old and not fast, and were more of an experiment. I recall a few questions about them but it seemed like people were really just trying to do clustering, and this is a bad way to do clustering.
knn is old too, and in a sense spiritually quite similar to ALS. I don't mind removing it either. It would seal it if there were even a nominal argument that this improves the rest of the code base -- less to maintain, removes duplication, inconsistency, etc. I could imagine that argument here. On Thu, Dec 6, 2012 at 3:06 PM, Sebastian Schelter <[email protected]> wrote: > Hi there, > > I'm currently thinking whether we should do a little cleanup in the > non-distributed recommenders package and throw out recommenders that > have not been used/asked about on the mailinglist or that have been > replaced by a superior implementation. > > If anyone reads this and sees a recommender, he/she wants to be kept, > please shout! > > /s > > Here's a list of suggested stuff to remove, let me know what you think: > > org.apache.mahout.cf.taste.impl.recommender.svd.FunkSVDFactorizer > > RatingSGDFactorizer should be learning faster and has a nicer model as > it includes user/item biases > > > org.apache.mahout.cf.taste.impl.recommender.svd.ImplicitLinearRegressionFactorizer > > Seems to be using the same model as ALSWRFactorizer, however there are > no tests and ALSWR can handle more explicit and implicit feedback > > > org.apache.mahout.cf.taste.impl.recommender.TreeClusteringRecommender > org.apache.mahout.cf.taste.impl.recommender.TreeClusteringRecommender2 > org.apache.mahout.cf.taste.impl.recommender.knn > > I don't recall anybody using those or asking about them the last years. > > >
