Which of the following methods can not achieve zero training error on any linearly separable dataset?

• A. decision tree
• B. 15-nearest neighbors
• C. hard-margin svm
• D. perceptron