Policy evaluation trees

Abstract

Technology for improving evaluation of policies comprising multiple rules is disclosed. By generating a policy evaluation tree controlling, for any given policy state, which rules should be evaluated next, policy optimization can be performed off-line prior to policy evaluation. For a policy, a policy evaluation tree can be generated such that each node in the tree corresponds to a policy state and each child node corresponds to a policy state that may result from an action that may be taken from its parent policy state. Policy evaluation trees may be generated by iteratively generating, from an initial policy state, possible next states as child states until a result of the policy is determined. Some next possible policy states may be pruned from the tree based on conditions such as having a high cost of evaluation compared to the likelihood a rule will yield an interesting result.