reagent/training/cfeval/bandit_reward_network_trainer.py [20:45]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    def __init__(
        self,
        reward_net: ModelBase,
        optimizer: Optimizer__Union = field(  # noqa: B008
            default_factory=Optimizer__Union.default
        ),
        loss_type: LossFunction = LossFunction.MSE,
        reward_ignore_threshold: Optional[float] = None,
        weighted_by_inverse_propensity: bool = False,
    ) -> None:
        super().__init__()
        self.reward_net = reward_net
        self.optimizer = optimizer
        self.loss_type = loss_type
        self.reward_ignore_threshold = reward_ignore_threshold
        self.weighted_by_inverse_propensity = weighted_by_inverse_propensity
        self.loss_fn = _get_loss_function(
            loss_type, reward_ignore_threshold, weighted_by_inverse_propensity
        )

    def configure_optimizers(self):
        optimizers = []
        optimizers.append(
            self.optimizer.make_optimizer_scheduler(self.reward_net.parameters())
        )
        return optimizers
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



reagent/training/reward_network_trainer.py [69:94]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    def __init__(
        self,
        reward_net: ModelBase,
        optimizer: Optimizer__Union = field(  # noqa: B008
            default_factory=Optimizer__Union.default
        ),
        loss_type: LossFunction = LossFunction.MSE,
        reward_ignore_threshold: Optional[float] = None,
        weighted_by_inverse_propensity: bool = False,
    ) -> None:
        super().__init__()
        self.reward_net = reward_net
        self.optimizer = optimizer
        self.loss_type = loss_type
        self.reward_ignore_threshold = reward_ignore_threshold
        self.weighted_by_inverse_propensity = weighted_by_inverse_propensity
        self.loss_fn = _get_loss_function(
            loss_type, reward_ignore_threshold, weighted_by_inverse_propensity
        )

    def configure_optimizers(self):
        optimizers = []
        optimizers.append(
            self.optimizer.make_optimizer_scheduler(self.reward_net.parameters())
        )
        return optimizers
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -