Consider a two-class problem with 5 examples from each class, sorted so the 5 H0's are first.
Now consider a classifier run on that, that gives outputs like:
[1 2 3 4 5 101 102 103 104 105]
Right now, our prtDecisionMinPe chooses 101 as the threshold. This is fine for here, but is no good in cross-validation. The right threshold is probably mean([5, 101])
Consider a two-class problem with 5 examples from each class, sorted so the 5 H0's are first.
Now consider a classifier run on that, that gives outputs like:
[1 2 3 4 5 101 102 103 104 105]Right now, our prtDecisionMinPe chooses 101 as the threshold. This is fine for here, but is no good in cross-validation. The right threshold is probably mean([5, 101])