Chapter 16. Time Series Packages
In the past several years, there have been a number of packages and papers released by large tech companies related to how they deal with the massive number of time series they collect as digital organizations with enormous customer bases, sophisticated logging, cutting-edge business analytics, and numerous forecasting and data processing needs. In this chapter we will discuss some of the main areas of research and development related to these ever-expanding time series data sets, specifically: forecasting at scale and anomaly detection.
Forecasting at Scale
For many large tech companies, dealing with time series is an increasingly important problem and one that arose naturally within their organizations. Over time, several of these companies responded by developing smart, automated time series packages specifically targeted to “forecasting at scale” because so many forecasts were needed in a wide variety of domains. Here’s how two data scientists at Google who developed the company’s automated forecasting package described the circumstances that motivated their product in a 2017 blog post (emphasis added):
The demand for time series forecasting at Google grew rapidly along with the company over its first decade. Various business and engineering needs led to a multitude of forecasting approaches, most reliant on direct analyst support. The volume and variety of the approaches, and in some cases their inconsistency, called out for an attempt to ...