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

[MLPACK Trac]

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

Comment (by rcurtin):

 Hello Sumedh,

 Sorry for the slow response.  I have been busy.  I would suggest this as a
 check for the cost:

 {{{
 const double cost = lr.ComputeCost(points, responses);

 // 0.01% tolerance.
 BOOST_REQUIRE_CLOSE(cost, 9.07497e-5, 0.001)
 }}}

 It might be easy to write tests for some other simple corner cases for
 ComputeCost().  For instance, if the responses fit the model perfectly,
 then the cost should be zero.  It shouldn't be too hard to make a dataset
 like that... here's one in two dimensions, with the responses as the one-
 dimensional column vector to the right:

 {{{
 [[0 0]  [[0]
  [1 1]   [2]
  [2 2]   [4]
  [1 2]   [3]
  [6 2]   [8]
  [2 6]]  [8]]
 }}}

 That's the relation y = x_1 + x_2, which the linear regression model
 should fit perfectly (giving a cost of 0).  You could write a test
 function to run linear regression on that dataset then ensure the cost of
 that dataset is 0.

-- 
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/298#comment:6>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.