Profile memorization is the best any model can do on fitting. It represents the highest percent correct any model can possibly get in fitting which would be useful to know in validation tests. That also tells us how "hard" a data set is in general.
Of course, it usually won't be best in cross-validation, but it might be an interesting reference point in that case, too. The amount it drops from fitting to cross-validation is particularly informative about a data set.
Profile memorization is the best any model can do on fitting. It represents the highest percent correct any model can possibly get in fitting which would be useful to know in validation tests. That also tells us how "hard" a data set is in general.
Of course, it usually won't be best in cross-validation, but it might be an interesting reference point in that case, too. The amount it drops from fitting to cross-validation is particularly informative about a data set.