Supervised learning algorithms need a training set and a test set. This snippet show two ways to generate train/test sets. First is by count, where it will take a random number of rows from the Frame and generate two new frame of the specified size. This is useful if you have a very large dataset, but you only want to explore your learning algorithm on a small subset. Second is by ratio, this will break the frame into two parts based on the ratio given. For example you might want to train on 25% of your data, then test on the other 75%.