If you are comparing several methods, whether dividing it
by the # of classes or not does not affect the result.
※ 引述《ydaniel00 (Daniel)》之銘言：
: Hello,
: I have a question about Brier Score. For a given instance, do we average
: its Brier Score over the total number classes? I get the impression that
: the Brier Score is averaged over the total number clases in Dr. Lin's paper.
: However, the textbook slide's formula for quadratic loss (Brier Score) does
: not take an average over the total number of classes.
: Also, when we want to calculate the probability error of multiple instances,
: do we always divide the sum of all instances' errors by the total number
: of instances regardless of the error criterion we use?
: Thanks.