On the Complexity of Inductively Learning Guarded Rules