"Choosing C and epsilon of a linear c-svm"

Question

Hey there, I currently doing my bachelor's thesis using the libsvm c-svm with a linear kernel. Now the time has come to optimize the parameters. What ranges should be checked for optimal values? I forgot where, but somewhere I read, that epsilon should stay at the default value of 0.001. Is that true?

FBT · Answer

No. γ (gamma) ≠ y. I read the guide quite a while ago, so I don't remember exactly. However, I believe they focus only on gamma and C and leave the default value for ε, which does not appear to have a big impact with rbf kernels.

clarabss · Answer

Hi!!

I have a question about  "A Practical Guide to Support Vector Classification" by Chih-Wei Hsu et al. Is in this article y = epsilon (ε), are they mining exactly the same parameter?

Thanks for answering, I'm little bit puzzled about the use of y and ε

dali · Answer

Hi,
in "A Practical Guide to Support Vector Classification" by Chih-Wei Hsu et al. it says "We recommend a grid-search on C and using cross-validation. Various pairs of (C; γ) values are tried and the one with the best cross-validation accuracy is picked. We found that trying exponentially growing sequences of C and γ is a practical method to identify good parameters (for example, C = 2^-5, 2^-3, . . . , 2^15; γ = 2^-15, 2^-13, . . . , 2^3).“

Further on Horváth and Suykens say "For an SVM the value of ε in the ε-insensitive loss function should also be selected. ε has an effect on the smoothness of the SVM’s response and it affects the number of support vectors, so both the complexity and the generalization capability of the network depend on its value. There is also some connection between observation noise in the training data and the value of ε. Fixing the parameter ε can be useful if the desired accuracy of the approximation can be specified in advance."

I hope this helps a little, otherwise there is a whole bunch of studies on this subject available online.
Greetings, dali.