ratio = new_policy_prob / old_policy_prob