I'm self-studying some ML over the summer and ran into a question that I don't really understand. I don't really understand what (i) is asking. Is the minimum possible penalty not the trivial case in which you correctly predict the test example for a total penalty of 0? They mention multiple cases so it seems like they're asking for something else, but I guess I don't understand what the question is even asking. Would they not phrase the question as expected or average penalty if they were looking for something more? Any one know what I'm missing?
Hint: your dataset can contain contradictions. The same data point can occur multiple times with different labels.