Throughout this book, we have built models involving many manual steps. For declinations of the SARIMAX models, for example, we had to develop a function to select the best model according to the Akaike information criterion (AIC) and a function to perform rolling forecasts. In the deep learning portion of the book, we had to build a class to create windows of data, as well as define all the deep learning models, although this was greatly facilitated by the use of Keras.
While manually building and tweaking our models allows for great flexibility and total control over our forecasting techniques, it is also useful to automate most of the forecasting process, making it easier to forecast time series and accelerating experiments. Therefore, it is important to understand the automation tools, as they are a fast way to obtain predictions, and they often facilitate the use of state-of-the-art models.
In this chapter, we’ll first look at the various libraries that automate the process of time series forecasting. Then we’ll focus specifically on the Prophet library, which is arguably the most well-known and widely used forecasting library. We’ll explore its functionality using a real-life dataset. Finally, we’ll conclude this chapter with a forecasting project so we can see Prophet in action.