The only worthwhile version of this test involves previously unseen data that could not have been in the training set. Otherwise the results could be inaccurate to the point of harmful.