[mlpack-svn] [MLPACK] #298: Enhancement of linear regression library
MLPACK Trac
trac at coffeetalk-1.cc.gatech.edu
Sun Aug 11 12:57:08 EDT 2013
#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: |
--------------------------------+-------------------------------------------
Changes (by rcurtin):
* owner: => rcurtin
* status: new => accepted
Comment:
> But except that everything is fine. I have checked my answers against
octave modules and they both giving the same results.
> Just check test.cpp and pass csv file in that.
So, the idea behind the test is that you've tested it by hand, and I trust
you that it's right, but two years from now, someone might change a line
in there and everything might break -- and they're not going to take the
time to run the tests again. So writing a test just makes the testing you
were doing automatic. That way, if someone makes a change, all they have
to do is 'make mlpack_test' and then run the test to know that everything
is still working right. If you send me the Octave numbers you were
comparing against, I can show you an example of what I mean. There is
already a linear regression test for standard linear regression you can
run; if you 'make mlpack_test' you can then run 'mlpack_test -t
LinearRegressionTest' and see if it still passes the test.
Thanks for changing the implementation to work with column-major matrices.
data::Load() automatically transposes data (usually, data is stored in
row-major format, but we need it in column-major format), so your code was
calling data::Load(), which transposes, and then transposing again.
arma::solve() should work in place to arma::inv(), to my knowledge. If it
still passes your tests then there is no problem. :)
I'll take a look into the return type for the predictions and let you know
what that ought to be.
--
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/298#comment:4>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.
More information about the mlpack-svn
mailing list