An empirical risk minimization approach for offline inverse RL and Dynamic Discrete Choice models