[mlpack-svn] [MLPACK] #306: Use sparse AllkNN in collaborative filtering

[MLPACK Trac]

#306: Use sparse AllkNN in collaborative filtering
----------------------------------------+-----------------------------------
  Reporter:  rcurtin                    |        Owner:              
      Type:  defect                     |       Status:  closed      
  Priority:  major                      |    Milestone:  mlpack 1.0.9
 Component:  mlpack                     |   Resolution:  invalid     
  Keywords:  cf, allknn, sparse, spmat  |     Blocking:              
Blocked By:                             |  
----------------------------------------+-----------------------------------
Changes (by rcurtin):

  * status:  new => closed
  * resolution:  => invalid


Comment:

 So, after a lot of discussion with different people, and learning a bit
 about CF myself, this ticket doesn't actually make all that much sense.
 The reconstructed dense rating matrix is, in some sense, what "fills in"
 the missing ratings, so it is necessary.

 I still don't like the huge memory footprint that can incur, and I think
 it might be possible to do a nearest-neighbor search to find the best item
 (or user) to recommend without actually calculating the dense rating
 matrix, but either way, this ticket is invalid.

-- 
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/306#comment:1>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.