#298: Enhancement of linear regression library
--------------------------------+-------------------------------------------
Reporter: sumedhghaisas | Owner:
Type: enhancement | Status: new
Priority: major | Milestone: mlpack 1.0.7
Component: mlpack | Resolution:
Keywords: linear_regression | Blocking:
Blocked By: |
--------------------------------+-------------------------------------------
Comment (by rcurtin):
Hello Sumedh,
Thank you very much for the contribution! I will be taking a look through
it shortly and integrating your improvements. The improvement of
Predict() looks good; I will have to make sure it still passes the test.
I do think the equation for theta, as implemented, is not correct because
Armadillo uses column-major memory ordering. See
http://www.mlpack.org/doxygen.php?doc=matrices.html for a better
explanation (and let me know if that doesn't clarify).
Are you willing to write a test for the ridge regression and the cost
function calculation? Everything that mlpack provides is tested; you can
find the tests in src/mlpack/tests/. We already have one for linear
regression in src/mlpack/tests/linear_regression_test.cpp. A good test
for ridge regression might be to give a dataset where the inversion will
fail if lambda = 0 (i.e. pass in a data matrix X where (X * X^T) is not
invertible), but will give good results for some nonzero lambda. You
could use the current test as an example. A test for the cost should not
be hard; give an example dataset or two, and compare the cost with the
cost that is expected for that dataset.
Let me know if you have any questions. Again, thanks for the
contribution. I will be working at least the modifications to Predict()
in over the next day or two, and once I have a test for the new
functionality I am happy to add that too.
Thanks!
Ryan
--
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/298#comment:1>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.