Holdout validation, data taken randomly? 3 questions

In classification learner, I got this accuracy of 97% using gaussian SVM technique. I used holdout validation (125 set of data) with 25% data as test set.

Q1: These 25% data taken randomly? Q2: How do I know which data are taken for testing? I have two classes defined. Q3: Does it mean it will take half of the 25% data from class 1 and other half from class 2?

Answer 1

[Sal]

When you are doing the partition, what variable are you supplying to the function? This should be your class labels. That way, you can ensure that you have a "balanced" training and testing set e.g. they will contain roughly the same percentage of data from each class as in the original data. Yes, I believe this 25% data are taken randomly.