Selects the samples with the smallest predicted probability for top class.
Selects the samples with smallest difference between the predicted probability of the top two classes.
Selects the samples with greatest predicted probability entropy.