CreateML with body landmarks

How do I directly input landmarks to the activity classifier rather than inputting an image/video?

CreateML with body landmarks
 
 
Q