-
Book Overview & Buying
-
Table Of Contents
Machine Learning for Time Series with Python - Second Edition
By :
For time series in particular, the quality of the features is what makes or breaks a model. Feature engineering turns raw data into the informative signals the model needs. This section covers the first layer of that work: preparing the data and creating the first features.
Our laboratory for this chapter is the M5 Forecasting Accuracy dataset, the data behind the influential M5 competition. It captures real daily unit sales for thousands of Walmart products across ten stores. Like most business data, it contains missing values (products not yet on sale at a given store), categorical hierarchies, calendar events, and price changes, so it makes for a realistic challenge.
Let's frame our mission. You are a forecasting analyst at a large retailer working from M5-style data: 30,490 product-store time series spanning about five years, plus a calendar of national events and SNAP food-benefit days, plus weekly sell prices per item. Your task is to produce daily unit...