Value Generalization Among Policies: Improving Value Function with Policy Representation