On learning spatial sequences with the movement of attention