My classification problem has 4 classes.
If I use inbuilt LIBSVM library in which one-one approach is used, and check the dimension of support vector coefficient ie, model.sv_coef, it is 53 x 3. where 53 is the number of support vectors.
But, if I use one-rest approach for classification using LIBSVM model. then the dimension of support vector coefficient, i.e. model.sv_coef is 58 X 1.(58:nsv).
Why by one-one approach 3 columns and using one-rest, it is only 1.
Its very simple. Let the classes be 1,2,3,4. When learning one-one classifier, you are training classifiers for 1-2, 1-3,1-4 and likewise for the others. So you train n-1 one-one classifiers. For one-rest you train a classifier for 1-U(2,3,4) where U denotes union of the features of classes 2,3,4. So the dimension is one.
- Badges
- Science topic
- Similar topics
- Vectorization