I find in the paper that ProBis is used to avoid highly-similar target appear both in TRAIN and TEST files. But what is the point of totally avoiding the test-set has similar target with train-set? I argue that it would make the evaluation on test more difficult than it should be.
I mean, does other scoring functions use the same methods to train model? If not, how could you contrast gnina with them?
I find in the paper that ProBis is used to avoid highly-similar target appear both in TRAIN and TEST files. But what is the point of totally avoiding the test-set has similar target with train-set? I argue that it would make the evaluation on test more difficult than it should be.
I mean, does other scoring functions use the same methods to train model? If not, how could you contrast gnina with them?