[mlpack-svn] [MLPACK] #298: Enhancement of linear regression library

[MLPACK Trac]

#298: Enhancement of linear regression library
--------------------------------+-------------------------------------------
  Reporter:  sumedhghaisas      |        Owner:  rcurtin     
      Type:  enhancement        |       Status:  closed      
  Priority:  major              |    Milestone:  mlpack 1.0.7
 Component:  mlpack             |   Resolution:  fixed       
  Keywords:  linear_regression  |     Blocking:              
Blocked By:                     |  
--------------------------------+-------------------------------------------
Changes (by rcurtin):

  * status:  accepted => closed
  * resolution:  => fixed


Comment:

 Clever!  I've written a test for that and another test to ensure ridge
 regression with a very small lambda is equivalent to linear regression.

 I also refactored the computation of the model to use the QR factorization
 even when ridge regression is used, and I've minimized copies by avoiding
 insert_rows() and shed_rows().  A call to insert_rows() and shed_rows() is
 two memory allocations and copies, but if we just copy the matrix once, it
 ends up being faster.

 Here's a link to more information on using the QR decomposition for ridge
 regression, if you are interested:
 http://math.stackexchange.com/questions/299481/

 I'm going to close this ticket since all of the functionality you
 originally submitted is now implemented.  You can use ridge regression
 with the 'linear_regression' executable by specifying a parameter to
 --lambda.

 Thanks again for your contribution.  The features you've designed will be
 available with the 1.0.7 release, which should happen by early October. :)

-- 
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/298#comment:13>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.