[mlpack-svn] [MLPACK] #288: run GMM::Estimate with weighted data

[MLPACK Trac]

#288: run GMM::Estimate with weighted data
------------------------------------+---------------------------------------
  Reporter:  hanslovsky             |        Owner:            
      Type:  wishlist               |       Status:  closed    
  Priority:  major                  |    Milestone:            
 Component:  mlpack                 |   Resolution:  worksforme
  Keywords:  GMM Estimate weighted  |     Blocking:            
Blocked By:                         |  
------------------------------------+---------------------------------------
Changes (by rcurtin):

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


Comment:

 You could use the overload of GMM::Estimate() that takes probabilities of
 each point being in the distribution:

 http://www.mlpack.org/doxygen.php?doc=classmlpack_1_1gmm_1_1GMM.html#aa8088138d8de9d185afe9239ee3114e0

 So, along with your observations, you pass a vector of probabilities.  If
 the probability is high, then that point will affect the final results
 more than a point with small probability.  I think that is exactly what
 you're looking for.  If not, let me know (and feel free to reopen the
 ticket :)).

 I did have a concern back when I wrote it that the estimator is biased,
 but I never worked it out and in practice I doubt it matters much.

-- 
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/288#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.