[mlpack-svn] [MLPACK] #306: Use sparse AllkNN in collaborative filtering
MLPACK Trac
trac at coffeetalk-1.cc.gatech.edu
Tue Oct 1 11:52:41 EDT 2013
#306: Use sparse AllkNN in collaborative filtering
---------------------+------------------------------------------------------
Reporter: rcurtin | Owner:
Type: defect | Status: new
Priority: major | Milestone: mlpack 1.0.8
Component: mlpack | Keywords: cf, allknn, sparse, spmat
Blocking: | Blocked By:
---------------------+------------------------------------------------------
The basic idea in collaborative filtering is to decompose the sparse
rating matrix into a W and H matrix. The current code does this, but it
also reconstructs an approximate matrix 'rating = W * H'. Unfortunately
this can be very large, for situations with large numbers of users and
items.
I'd like to remove the dense 'rating' matrix and replace it with the
actual sparse rating matrix. To do this, the calls to AllkNN would have
to be modified so that sparse input matrices could be used.
Theoretically, it should work...
(Mudit, I CC'ed you because you're the author of the CF code :).)
--
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/306>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.
More information about the mlpack-svn
mailing list