I have a dataset of feature/label pairs. My labels are probabilities of each feature vector to belong to the K classes. Here is an example for K = 3:

D1 = { (V0, [0.33,0.33,0.33]), (V1, [0.9,0.07,0.03]), (V2, [0.5,0.25,0.25])... }

The probabilities are normalized for a given data point. Yet the task is more a multilabel one, and it would make more sense to have independent Bernoulli distributions e.g.

D2 = { (V0, [0.9,0.9,0.9]), (V1, [0.99,0.0,0.0]), (V2, [0.9,0.2,0.5])... }

Is there a trick (smart heuristic) out there which would allow me to transform D1 into D2 based on the way the probability weights are distributed in D1?

  • $\begingroup$ Can you be more specific? How did you exactly got from (V0, [0.33,0.33,0.33]) to (V0, [0.9,0.9,0.9]) or from (V0, [0.9,0.9,0.9]) to (V1, [0.99,0.0,0.0])? $\endgroup$ – Antonio Jurić Feb 25 '19 at 13:05
  • $\begingroup$ That's the question :) $\endgroup$ – user3091275 Feb 26 '19 at 16:27

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.