It is common practice in machine learning to separate out a training set and a test set. Often, we then split the training data again, to get a validation set.

Which is false?