I am trying to implement a recommender system uisng Mahout framework. As I do not have Linux machine, I cannot use Hadoop to deal with large martrix. (The tutorial for installing Hadoop on Windows does not work for me.)
My Users have three types of features, each type contains 5 to 9 features. I wonder whether I can build these features into one FileDataModel, or whether I can process each group of features separately and combine the results.
If the latter one works, then I need to have the UserIds from the previous processing result to create a new FileDataModel for the next group of features. Is it doable?
Here I still have questions for which I have not found answer, hope anyone can help: how many features can Mahout within Windows environment, without Hadoop, handle actually? And how often do we need to re-optimize a system after one algorithm being put in production? Does it work automatically? Thanks.
