Abstract: In the classic machine learning framework, models are trained on historical data and used to predict future values. It is assumed that the data distribution does not change over time ...