[mlpack-svn] [MLPACK] #305: Logistic Regression

[MLPACK Trac]

#305: Logistic Regression
----------------------------------+-----------------------------------------
  Reporter:  sumedhghaisas        |        Owner:  rcurtin     
      Type:  enhancement          |       Status:  accepted    
  Priority:  major                |    Milestone:  mlpack 1.0.8
 Component:  mlpack               |   Resolution:              
  Keywords:  logistic_regression  |     Blocking:              
Blocked By:                       |  
----------------------------------+-----------------------------------------

Comment (by rcurtin):

 Hey,

 Thanks for double-checking my work.  :)

 In MATLAB writing things as vectorized can often result in much faster
 evaluation, but that doesn't apply so much in this case.  Internal
 Armadillo code often uses for loops in the exact same way I did.  What I'm
 trying to avoid is the allocation and filling of a big vector of ones,
 when that isn't actually necessary to get the correct result.  When I have
 a main executable running (I'm working on this now, albeit a little
 slowly), I'll test both implementations and report my results back, but
 I'm nearly certain that in this case the non-vectorized implementation
 will be faster.

 You are correct about the regularization, though.  Thank you for pointing
 it out, because the tests were also written incorrectly.  I've fixed the
 issues in r16065 and r16066.

-- 
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/305#comment:6>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.