There's no Data like More Data? Revisiting the Impact of Data Size on a Classification Task