Hi,
I have a problem with SVM in matlab, I use fitcecoc for multi classification im matlab, when I use linear kernel it works good, but when I use 'rpf' kernel all data classified into one class.
Where is the problem?
Best
Dear Huthaifa
SVM classifier when is used with RBF kernel is very sensitive to the kernel parameters, i.e. the kernel parameters must be finely tuned.
When using Matlab built-in SVM model trainer (fitcecoc), it can set the parameters automatically if it is said to. For example:
- svmParams = templateSVM('KernelFunction','rbf', 'KernelScale', 'auto', 'Standardize', 1);
- svmModel = fitcecoc(trainData, trainLabels, 'Learners', svmParams, 'Coding', 'onevsall');
I have attached a full example, too.
I hope it helps.
Best
Fardin
Dear Fardin,
Thank you very much for your help, your code works fine and it is really helpful.
May I ask you how can I find the best C and sigma for my model??
And how can I give the fitcecoc this parameters (C and sigma).
Best,
Huthaifa
- Badges
- Science topic