[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 sumedhghaisas):

 In changeset 16020 you have changed the vector implementation of the
 logistic function into a for loop.. Wouldn't the vector implementation
 will give less computational overhead?? Your point of neglecting the
 dividing factor of number of features is correct... if we do remove it the
 vector implementation of that equation would be

 -(sum(arma::trans(responses) * arma::log(sigmoid)) +
                       sum(arma::trans(arma::ones<arma::vec>(nCols, 1) -
 responses) *
                       arma::log(arma::ones<arma::vec>(nCols,1) -
 sigmoid)))
                        + regularization;

 and u inverted both the result and the regularization for minimization
 purpose... but after deriving the regularized LR we get the sign of the
 regularization term as positive.. so it should be added to the term (-
 result).

-- 
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/305#comment:5>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.