Hi, I am using roboflow in my thesis. May I know what is the calculation for a class being called “over representation” and “under representation”?
These are just estimates. Our UI calls a class “balanced” if its label count is between 75% and 200% of the average class in the dataset’s count.
It’s best if they’re roughly equal, but that’s not always possible (eg imagine a chess dataset; you’re always going to have more pawns than queens and that’s probably Ok because that will be the case in the real world too).