[Question] Calculate loss given resampled data

[voloddia]

Hi,
I realize that sometimes adaptive generates significantly different losses (~2 orders of magnitude) given the same input.
As a sanity check, I want to recalculate losses of resampled data without calling learner and runner.
Is it possible to do so?