|
How multi-class classification works in LibSVM? I tried to find in the documentation and hadn't found, maybe I was looking to wrong place. Interested in their approach, one against all, converting to one optimization problem, something else. |
|
Libsvm supports different kinds of multi-class svms. Without additional options, the default is 1-vs-all. See the FAQ: http://www.csie.ntu.edu.tw/~cjlin/libsvm/faq.html#f419. I think there is some trick involved in optimizing these SVMs together. The details are in the paper: http://www.csie.ntu.edu.tw/~cjlin/papers/quadworkset.pdf Other possible methods are described here: http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/ |
