[mlpack-svn] [MLPACK] #205: kernel usage

MLPACK Trac trac at coffeetalk-1.cc.gatech.edu
Sat Aug 4 22:59:04 EDT 2012


#205: kernel usage
--------------------------------------+-------------------------------------
  Reporter:  nslagle                  |        Owner:              
      Type:  defect                   |       Status:  closed      
  Priority:  major                    |    Milestone:  mlpack 1.0.2
 Component:  mlpack                   |   Resolution:  invalid     
  Keywords:  kernels, KDE, KCDE, SVM  |     Blocking:  199, 200    
Blocked By:                           |  
--------------------------------------+-------------------------------------
Changes (by rcurtin):

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


Comment:

 I haven't had the time to look into this until just now, unfortunately.

 At the moment we don't even have kernel density estimation code available.
 So even if we did discriminate kernels by 'valid density estimation kernel
 / invalid density estimation kernel' it's not quite that simple; one can
 use a valid density estimation kernel for PCA or some other methods which
 require a Mercer kernel.  I don't know the exact terminology for a valid
 density estimation kernel but I believe it to be a subclass of Mercer
 kernel.  If I took the time I could probably find or prove it but it's a
 bit of a moot point because at the moment we don't have KDE code.

 I think that if this does become an actual problem the best idea would be
 to have a separate directory for density-specific kernels under
 `core/kernels`, and then leave the larger class of Mercer kernels under
 just `core/kernels`.

 I'm going to close this ticket for now since it's not applicable to the
 current codebase, but if a KDE implementation appears then we can consider
 it then.  Or, reopen it if my analysis is flawed. :)

-- 
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/205#comment:1>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed by the FASTLAB at Georgia Tech under Dr. Alex Gray.


More information about the mlpack-svn mailing list