Spark: With the original dataset split into 80% for training and 20% for test, how many of the from the test set were correctly classified?