That is all 21 problems. From here on, it is your turn to do the work.
Multiply, add, bend, measure how far off you are, walk downhill. From Chapter 1 to Chapter 4, what you worked with your own hands was a combination of these five. And in Chapter 5 you saw that those tools do not work on everything.
Why it gets smarter the more parameters you give it is something the people who built it cannot account for either.
The table of 891 people you used in Problems 18-21 is published on a site called Kaggle. You put answers to the 418 people in the scoring set, send them, and they get marked. It is a place where people all over the world are working at the same time on problems whose answer is not settled yet.
All the numbers are measured from public data (the Titanic set on Kaggle). Problems 18-21 were counted on the 891 training passengers. The unseen score is the average of five rounds, with the 891 split into five parts.
Go back through any chapter you want to see again.