The optimum decision rules for the oddity task