TBF, going by the problem statement in the article, the objective function could be pretty wacky. For example, do you weight accuracy by location? Is it more im

TBF, going by the problem statement in the article, the objective function could be pretty wacky. For example, do you weight accuracy by location? Is it more important to get your predictions right over sports stadiums than over residential areas? Would it be useful to weight accuracy by time, like, it's more important to get predictions right at 7 AM when people are driving to work than it is to get it right at 1 AM when everyone is asleep? I don't think that that's what's actually happening, but I do think that weather is a sufficiently ugly problem that comparing to human performance is useful.