Facebook’s motivation for building Prophet
As mentioned when introducing Prophet in Chapter 1, The History and Development of Time Series Forecasting, Facebook noticed that the internal demand for business forecasts was increasing. Its forecasting techniques did not scale well and its analysts were overwhelmed.
Facebook scoured the literature in search of a scalable forecasting methodology. At the time, Facebook’s forecasting was largely done with Rob Hyndman’s forecast package in R (https://github.com/robjhyndman/forecast, now superseded by his fable package: https://github.com/tidyverts/fable). Although powerful, the forecast package required R analysts with specialized data science skills in forecasting and substantial product experience. Further, as Python became more and more popular among new hires, Facebook found itself running short of analysts able to produce high-quality forecasts. Unfortunately, the completely automatic forecasting tools Facebook...
