In order to choose the best value of the SVM parameter C:

A) we need to run cross-validation. There is no really a way to tell except for trying to minimize the test set error.

B) we need to consider how many variables are in the problem. When there are many variables, we need to use smaller C (inversely proportional to the number of variables).

C) we need to consider how many variables are in the problem. When there are many variables, we need to use larger C (proportional to the number of variables).

D) we need to evaluate the variance of the data. When variance is large, we need to use large C.