Machine Learning at Large Scale (Part 3): Best Practices for Train and Prediction