I would add that this kind of literal analysis is only useful to train _seeing_. More so, you are training to see in a very specific way. The lines drawn in the picture by the author to indicate composition, rhythm and stability are quite arbitrary and could be drawn a dozens other ways.

In a way, you train your neural networks to see by looking at images, then, while actively photographing you outsource decision making to your own trained neural networks. After a while you can predict the best camera position, settings and get a tingling sensation moments before an interesting situation (in street or event photography) happens.