結果 : regularized q-learning with linear function approximation