As I mentioned in previous posts, I've been using SVMLight for quite some time now. It's been good while it lasted, but I'm now in the search for a stronger SVM library, especially one that handles multiple labels well. I've been introduced to LibSVM, which seems very promising. In theory, they have libraries for various systems and languages, including Matlab. However, I've had quite some trouble making it work with Matlab, so instead of breaking my head into solving the problem, I've decided to work around it in a quick-and-dirty manner (my favorite...).
What I did, is simply wrap the LibSVM's executables in a simple way, so it can be executed using Matlab functions (given the executables are located somewhere in your PATH).
The following two posts will include a wrapper for svmtrain.exe and svmpredict.exe
Please let me know if you encounter any problems using them.
Ah, of course, they are provided AS IS, no warranty or anything alike. I also don't take any credit for it, nor any responsibility as to how/who/when/where/etc. it can be used. For this - please check with the authors of LibSVM.
2 comments:
Hi IIan,
My name is Myra. I am using LIBSVM for my dissertation project. I am trying to reproduce an experiment that I did months ago. My problem is I am now getting different prediction results even with the same kernel parameters. Has this been your experience. My email add is custy15 at hotmail dot com.
cheers
I am trying to access the site of LIBSVM with no avail. I know you have nothing to do with that. Just wanted to discharge my disappointments.
thanks
Michael
Post a Comment