Why cross validation on SVM tests gammas for all kernels ?
![marwen_sallem](https://us.v-cdn.net/6038231/uploads/Dataiku/nAvatar10.png)
marwen_sallem
Registered Posts: 1 ✭✭✭✭
In new version of DSS 4.1, SVM has been added to the list of machine learning models.
When doing a cross validation on this model trying, for instance, linear kernel and rbf kernel by specifying some Gamma's, the cross validation tests all the Gamma's for the not only for rbf kernel (that's normal) but also for the linear kernel. However, Gamma is supposed to be a particular hyper-parameter for the rbf kernel, so we can skip it for the linear kernel in order to gain time.
Thank you for your time
When doing a cross validation on this model trying, for instance, linear kernel and rbf kernel by specifying some Gamma's, the cross validation tests all the Gamma's for the not only for rbf kernel (that's normal) but also for the linear kernel. However, Gamma is supposed to be a particular hyper-parameter for the rbf kernel, so we can skip it for the linear kernel in order to gain time.
Thank you for your time
Tagged:
Best Answer
-
Thank you four your feedback. This is indeed a limitation of our current grid-search mechanism which assumes hyperparameter independence.