Challenging the negative learning bias hypothesis of depression: reversal learning in a naturalistic psychiatric sample