Micro-averaged F1 vs. Accuracy

[desilinguist]

Given that f1_score_micro is basically the same as accuracy for multi-class classification problems, is there a reason why we should continue to use that as the default tuning metric rather than just switching to accuracy? We can still continue to make f1_score_micro available as a metric in SKLL.