How would your feature engineering strategy change if this data was streaming in real-time instead of arriving in a static batch?
