I use linear SVM and have a small dataset. Because of this I decided to so nestedCV for model checking and dir obtaining the penalty Parameter C. However, I am still confused on how to get to my final model. My understanding: I get more than one hyperparameter C from nestedCV. I just take the mean of it, however, If my model is stable it should not make much difference which of the C's I choose. Then I train the model with the found hyperparameters on my full dataset. Is this right? Do I still need a test Set at the end?