Asked about Data where order carries meaning
What is the first thing you do on a forecasting problem?
An answer that works
Compute the naive baseline: tomorrow equals today, or equals the same day last week. It is often hard to beat, and many forecasting projects never do without realising.
Reading an answer is not being able to give one. The section that teaches this is Data where order carries meaning, 5 hours of Time series.
Learn this properlyAlso asked about Time series
Why can you not use ordinary cross-validation on a time series?How do you build a rolling average feature safely?Would you use a neural network or a classical model for forecasting?Why evaluate at multiple forecast horizons?How would you evaluate a demand forecasting model?You need forecasts for 50,000 products, many with little history. How do you approach it?