On the Role of Weight Sharing During Deep Option Learning