Multiarmed bandits problem under the mean-variance setting