Oh, yes, that's your problem. You need to use the DataModel passed to the buildRecommender method, not build your own!
Otherwise you are creating a DataModel with all of your data, which contains all of the "answers" -- it already knows the correct value of all test data points. The point is that the framework constructs a new DataModel without the test data points, to see how well the recommender guesses the value of those data points. On Sat, Nov 21, 2009 at 1:35 PM, jamborta <[email protected]> wrote: > > it works fine if I use this code: > > public Recommender buildRecommender(DataModel dataModel) throws > TasteException { > > > UserSimilarity userSimilarity = new > PearsonCorrelationSimilarity(dataModel); > UserNeighborhood neighborhood = > new NearestNUserNeighborhood(10, userSimilarity, dataModel); > Recommender recommender = > new GenericUserBasedRecommender(dataModel, neighborhood, > userSimilarity); > return recommender; > > } > > but if I create a dataModel objects here that it doesn't work. I think the > reason is that I have two dataModel objects created and the one that is > passed to the evalautor is not separated. > > > > > > > > I get a result of 1.04. Are you sure your data file is right? Should be > using ua.base from that data set. > > > -- > View this message in context: > http://old.nabble.com/evaluating-recommender-systems-tp26421408p26456606.html > Sent from the Mahout User List mailing list archive at Nabble.com. > >
