Study introduces a new framework for policy learning without positivity assumption.
problem Learning optimal treatment assignment policies from observational data with constraints.
method Incremental propensity score policies and semiparametric efficiency theory.
result Validated framework's performance through numerical experiments.