http://arcdog.com/architects/a....ttackidea63/activity
Football dataset) had optimistic accuracy improvements on 63 lessons, no change in accuracy in 21 classes, destructive change in accuracy in 17 lessons (drop by 7%). With a purpose to validate the hypothesis (Part 3.2.3) that multi-label mannequin might have higher efficiency on numbers with less photographs, we examine its results with best multi-class mannequin on numbers with less than 50 images in training set. Although football players typically fulfill several different roles through the course of a season and even within