Ok, I also wonder what you think about the type of that implementation. -Will it be an extesion to Taste? I mean Taste's current implementation will remain same, also that will be added as an extra option. -Or do you plan to migrate Taste to that style?
I am asking because I am also working on implementing that model creation with known similarity metrics, like log-likelihood similarity. On Thu, Oct 15, 2009 at 9:29 PM, Sean Owen <[email protected]> wrote: > Agree, and no that does not exist yet. I suppose I have temporarily paused > thinking about it until 0.20.x sorts out these issues. > > On Oct 15, 2009 11:21 AM, "Gökhan Çapan" <[email protected]> wrote: > > Sorry, i've sent last post accidently. > The question was not about which version of hadoop it uses. > Let me explain, > I think we all agree that item based recommendation is a 2 step process. > 1-compute item-item similarities > 2-create recommendations to a specific user for an item > > And, 1st step can be computed offline. > If we want to compute all item-item similarities, it would be very good if > that computation had a map-reduce implementation. > > I remember we had a discussion about making this computation with a matrix > based approach and its potential of being distributable. I think that will > be a very good extension for taste. > > On Thu, Oct 15, 2009 at 9:12 PM, Gökhan Çapan <[email protected]> wrote: > > I > mean, > > On Thu, Oct... > -- > Gökhan Çapan > -- Gökhan Çapan
