This chapter will dive deeper into some more cutting-edge machine learning algorithms. In the first part of this book, I had a chapter on machine learning classification, so why didn't I put this material there?
My rule was that the classifiers in the earlier chapter could be used pretty much as black boxes. Yes, it was certainly possible to dissect them, analyze why they worked well or poorly, use them to extract fancier features, and so on. But you could just use them out of the box, and they would probably work pretty well.
The classifiers in this chapter are something of a dark art and really designed more for machine learning specialists than normal data scientists. They have a lot of internal structure, which must be planned out by the user, there are many different parameters to tune (arbitrarily many, depending on how complex the structure is), and there is no good precanned answer about how to make those decisions. You will have to think critically about the problems you plan on solving, design a classifier that is appropriate for those problems, and then experiment with different parameters and layouts to find what works best. And if you screw it up somehow, your classifier is liable to perform horribly.
In exchange for this extra work, you get to use the most powerful types of classifiers in the world today. This is the kind of stuff going on the heart of Google and Microsoft. It is used for hard problems such as identifying the people ...