SVM Train
SVM train stands for a process in machine learning that creates a Support Vector Machine (SVM) model by learning from big data. There are several ways how such a model can be trained and we provide an overview below. This overview includes the training parameters in the de-facto standard SVM implementation called libsvm that offers a tool called svm-train. More detailed pieces of information can be obtained here. This overview explains a bit more in detail the different parameter options for the SVM training process..
C-Support Vector Classification (C-SVC)
This is a classification model based on SVMs including multi-class classification. The cost (C) in this case refers to a soft-margin specifying how much error is allowed and thus represents a regularization parameter that prevents overfitting. The range of the parameter C is from zero to infinity and the svm-train parameter is ‘-s svmtype’ whereby the svmtype in this case is 0 for C-SVC. Although some application domain have established some rules of thumb for the value of C it can be a bit hard to estimate it and use in practical situations. Often cross-validation is used to systematically determine the value. It is a regularization parameter that prevents overfitting. The range of the parameter C is from zero to infinity and the svm-train parameter is ‘-s svmtype’ whereby the svmtype in this case is 0 for C-SVC.
nu-Support Vector Classification (nu-SVC)
This is also a classification model based on SVMs including multi-class classification. The nu in this case refers to values between 0 and 1 and thus represents a lower and upper bound on the number of examples that are support vectors and that lie on the wrong side of the hyperplane.
Details on SVM train
The following video is interesting in context:
