View Single Post
  #11  
Old 05-19-2013, 02:53 PM
warren warren is offline
Junior Member
 
Join Date: May 2013
Location: Tulsa, OK
Posts: 4
Default Re: *ANSWER* questions w linear regression & weight decay

I re-did #2 using Octave instead of Clojure. I'm a big Clojure fanatic; it's always the first tool I reach for, but I have to say when doing a lot of matrix stuff Octave is a lot more fun. Strangely, I got exactly the same weights for linear regression using the pseudo-inverse with Octave that I had gotten with Clojure, but I seem to have gotten the right answer with Octave for the in-sample and out-of-sample classification errors. So I'll just take this as showing that Octave is a lot less error-prone for me for this type of problem than Clojure is. (I still don't see what's wrong with my Clojure. If it were problems with Clatrix (bugs or numerical instability) I would think it would have shown up when doing the linear regression to get the weights, not just when calculating in-sample and out-of-sample errors. Oh, well, I guess I'll stick with Octave since that seems to be helping me the most at the moment. (I also had problems with my current evaluation copy of Mathematica.)
Reply With Quote