Yes, the rows of the user-item matrix exist in AggregateAndRecommendReducer, when they reach writeRecommendedItems(). This is where it's filtered.
I'm not sure if this will do what you want depend ing on what you mean by different dimensions. This is all still computing one user-item matrix for one set of 'prefs'. There aren't any other dimensions of anything hiding in here. On Tue, Dec 27, 2011 at 2:18 PM, Nick Jordan <n...@influen.se> wrote: > I've been playing with RecommenderJob for the last couple of days and have > it successfully running on my data set. > > The one question I have remaining is with the output. While it's possible > to define the min and max number of recommendations, what I'd really like > is the entire matrix of [users][items] with the recommendation score (if > any). The problem I'm trying to solve here is that there are various other > dimensions of my data that I'd like to be able to filter by. If I just > take the top n recommendations, none of them may meet the filter criteria. > > I suppose one option would be to run a separate job for each dimension, but > I lose the benefit of recommendations outside of that then limited data set. > > Is the entire recommendation matrix generated in one of the map-reduce jobs > prior to the final one that generates the output? Am I missing an easy > solution to this? > > Thanks in advance. > > ~N
