As known, the one-class SVM has one major parameter i.e. Nu. This parameter is bound within (0,1]. For the OCSVM kernels, we have the parameters:
1. RBF : Gamma
2. Sigmoid and Polynomial: Gamma and Coef0
So like the Nu parameter, is there any given range for these kernel parameters (specifically for OCSVM) or are they simply chosen at random?
Abdelkader Dairi Thanks for your response. I do perform grid search and well, it is one of the best ways to find the optimal parameters. However, this is not my issue.
To perform said grid search, I would need a range for the values. For eg. Nu = (0 to 1]. So, we do have a range for Nu making it easy to perform grid search. Is it so for other parameters as well? If I choose to use the gamma value as, let's say 2-10 to 210, it would heavily increase the computational burden and make the method tedious.
- Badges
- Science topic