XGBoost: How Deep Learning Can Replace Gradient Boosting and Decision Trees — Part 2: Training