I have a single classifier which produces probability vectors (a probability for each category). I know ahead of time that certain sets of input samples should result in the same, unknown label. With no other knowledge, how should I choose the "correct" label for these samples when I get conflicting predictions from each one? A friend suggested I take the geometric mean of the probability vectors, if that's the case can anyone explain the motivation behind this?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.