Introduction

In the previous chapter, we discussed the layers of a data-driven system and explained the important storage requirements for each layer. The storage containers in the data layers of AI solutions serve one main purpose: to build and train models that can run in a production environment. In this chapter, we will discuss how to transfer data between the layers in a pipeline so that the data is prepared to be used to train a model to create an actual forecast (called the execution or scoring of the model).

In an Artificial Intelligence (AI) system, data is continuously updated. Once data enters the system via an upload, application program interface (API), or data stream, it has to be stored securely and typically goes through a few ETL steps. In systems that handle streaming data, the incoming data has to be directed into a stable and usable data pipeline. Data transformations have to be managed, scheduled, and orchestrated. Further, the lineage of the data has to...