Notes / suggestions / bugs in ray tutorials

[shnayder]

Not sure if I'll get around to make a PR to fix some of these, so at least sharing notes:

Exercise 1: State up front that ray.get() works on arrays of object ids, not just individual object ids.

Exercise 4 and 7: make clear that you may need to add new lines, reorganize, etc.

Exercise 8:

• boring. Make me do something, e.g. use put() vs auto-creating multiple copies, show that it's faster

Exercise 9:

• num_gpus isn't actually set to 4.
• success misspelled in a couple of places
• also a bit boring. Not sure if there's something better to do without requiring actual GPUs. Maybe ask the student to print out the GPU ids in the actor and task.

Exercise 10:

• intro. a missing piece of context is how a machine specifies what resources it provides. If I understood correctly, Ray.init specifies what resources the cluster overall has, and each task says which resources are required, but the rest isn't clear.

Exercise 11:

• nit: telling me to use set/get_flat too late -- easy enough to fix, but better to suggest that up front.

Exercise 12:

• treefold! Yay :)
• I didn't notice the "you will need to refactor" comment at first and was confused. Maybe also mention in the instructions cell, not just in the code comment.

[iphysresearch]

I note all 12 exercises with my solutions at this post for your reference.