Is It a Cod Monkey or a Scientist?

In the latest m five competition, which dealt with retail datin. The approach that dominated let i beamis stands for light read in boosting machine learning method. But if we see the results of the m five competition albit more carefully, then we will realize that this out performance of macine learning over salistical methods is not consistent across ferent hieratical levels. How long can we keep the same masin learning model before you need to actually up date again? This is an argument, is a question that we don't have a clear indication. So do we net of date id every day, every ten days, every week, every month, every year? We don’t