{moderator edit - remove quiz answers}
I agree that the wording in some of these questions is pretty subtle and hard to understand. But I think you’re making a mistake in how you interpret what it says on that slide. Just because the decisions of what the best metric is and then how to do well on that metric are separate concerns, it does not mean that you don’t need to solve both of those problems. Prof Ng’s point on the slide is not to let your concern about how hard it might be to meet a given metric or the fact that you’ve already got experience with some other metric prevent you from picking the correct metric. It’s either the best metric for what the customer actually wants or it’s not, right? That’s the first problem you need to solve correctly. Then once you’ve picked the best one, now you have to solve the problem of how to train your system to do really well on that metric.
Maybe a useful analogy is the old joke about somebody looking for their lost keys in the front yard. A friend comes by and offers to help, but after searching unsuccessfully for a while, then asks:
“Where were you when you last remember seeing your keys?”
“In the house.”
“Then why are you looking in the front yard?”
“Because the light is better here.”
If the goal is to find the keys, then you need to find them where they are, not where the light is good, right?
nice example thanks
