On Dec 02, 2017; 1:58pmDevon McCormick wrote:
<http://jsoftware.2058.n7.nabble.com/template/NamlServlet.jtp?macro=user_nodes&user=14>

> Does anyone have J code to implement support vector machines that they would
like to share? 


I started a wrapper for sequential SVM using Leon Bottou's "lasvm."

http://leon.bottou.org/projects/lasvm

Never finished though. Probably never will, as my interests have changed.

For batch computations, libSVM has a reasonable API, though importing
data in their libSVM format will be messy. I also have a few conformal
prediction tools (C++ based on libSVM) in my githubs which you could
expose an API for.

SVM is pretty good for high dimensional problems like text
classification, but if that's your problem, give Naive Bayes a shot;
it's often close.

It should be possible to do SVM in native-J using the sparse type
without too much of a performance hit, but I wouldn't attempt it.

Oh yeah, you can do "instance learning" (aka kernel regression and KNN)
using 'math/flann' if your problem isn't too big in column-space.

-SL


----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to