In experiments with the latest polyssifier we've noticed that the mean accuracy values saved in files are different from the bar heights in the generated figure. After looking through the code I conclude that the values in the filenames are correct. It remains unclear why the labels get reordered.
Looks like a bug to me, either in polyssifier or seaborn/polyssifier connection