Geek asks and answers

Optimizing other svm-train options

libsvm "grid.py" try to optimize only two parameters "c" and "g" svm-train. I would like to extend "grid.py" to optimize other parameters (for example, "r" or "d"), running "grid.py" again and again for different parameters. I have some questions.
1. Is there another script that can optimize parameters other than "c" and "g"?
2. Which parameters are more important and what is the maximum / minimum range. Sometimes changing / optimizing one parameter automatically optimizes another parameter. Is this the case with svm-train options?

+5
source share
1 answer

As far as I know, there is no script for this, however I do not understand why grid.py cannot be easily extended for this. However, I do not think it is worth the effort.

First of all, you need to choose your kernel. This is a parameter in itself. Each core has a different set of parameters and will work differently, so to compare the kernels you will have to optimize each kernel parameter.

C, the cost parameter is a general parameter that applies to the SVM itself. Other parameters are all inputs to the kernel function. C controls the trade-off between wide margin and other training points that have been misclassified (but a model that can better generalize future data), and a narrow margin that is better for training points, but can be reset to training data.

, ( ) RBF.

RBF . , .

​​Polynomial, d , . , - , . , ​​ , , , RBF .

, - , coef0 , , .

, grid.py , , , , . : , , . . C. . , , .

+11

More articles:


All Articles