There are a lot of recommenders here -- what specifically are you looking at?
I might not call the FileDataModel update files as a form of incremental training. They add more data to the model efficiently though it's up to the recommender implementation to update its computations or not as appropriate. In general they try to be efficiently updateable, and re-compute results internally in the presence of new data. The distributed computations are sort of incremental in this sense. You can re-use existing similarity computations across runs, and you can write your own code to tweak those similarities incrementally. The existing jobs however just recompute similarities from scratch and so aren't incremental in that sense. On Mon, Apr 11, 2011 at 3:37 PM, Mathieu sgard <[email protected]> wrote: > Hello, > > I'm working on a recommender feature in e-commerce. > Is it possible to train the mahout recommender in incremental way or the > only way is compute entire dataset when new items are added ? > > Thanks >
