do all svm models have soft margin? how about if I don't put it in my model? from academic point of view , is it so unexeptable? I don't want to put epsilon on my constraints because I want to solve prime SVM with quadratic programming. if I remove epsilon then it is very easy to implement.