Machine learning is ultimately intelligent software, but it is not the magic wand that movies and literature (and recently also sales/marketing departments) love to depict. More importantly, machine learning is not a physical black box you can pick from the shelves of a drugstore, bring home, mount, and use.
In the real world, you can’t just “load data into the machine” and have the machine, in some way, just use it. In the real world, there are a few classes of approaches (mostly derived from statistics) such as regression, classification, and clustering and a bunch of concrete training algorithms. However, when to use which?
Determining which to use is a matter of experience and know-how, but it is also a matter of knowing data and how things actually work in the actual business domain. Does that mean that only experts can do machine learning? Yes, for the most part, that is just the point. However, nobody is born an expert, and everyone needs to get started in some way. This is the reason why automated tools for machine learning are emerging. In this chapter, we briefly looked at the Google Cloud AutoML and Visual Studio ML.NET Model Builder.
With the next chapter, we complete the preliminary path of machine learning, discussing the concept of a pipeline—namely, the sequence of steps that ultimately lead to the production of a deliverable model.