> It would be nice to provide a non-DL approach as a baseline, but I would expect it to lag behind the DL methods. The M# competitions have usually shown very

> It would be nice to provide a non-DL approach as a baseline, but I would expect it to lag behind the DL methods. The M# competitions have usually shown very old forecasting algorithms work quite well, with frankly, way less training overhead and data. Ensemble models usually do best, but for a lot of use cases, DL is probably overkill versus ARIMA or triple exponential smoothing.