class DiscreteBCQPolicy(*, model: Module, imitator: Module, optim: Optimizer, action_space: Discrete, discount_factor: float = 0.99, estimation_step: int = 1, target_update_freq: int = 8000, eval_eps: float = 0.001, unlikely_action_threshold: float = 0.3, imitation_logits_penalty: float = 0.01, reward_normalization: bool = False, is_double: bool = True, clip_loss_grad: bool = False, observation_space: Space | None = None, lr_scheduler: LRScheduler | MultipleLRSchedulers | None = None)[source]#

Implementation of discrete BCQ algorithm. arXiv:1910.01708.

  • model – a model following the rules (s_B -> action_values_BA)

  • imitator – a model following the rules in BasePolicy. (s -> imitation_logits)

  • optim – a torch.optim for optimizing the model.

  • discount_factor – in [0, 1].

  • estimation_step – the number of steps to look ahead

  • target_update_freq – the target network update frequency.

  • eval_eps – the epsilon-greedy noise added in evaluation.

  • unlikely_action_threshold – the threshold (tau) for unlikely actions, as shown in Equ. (17) in the paper.

  • imitation_logits_penalty – regularization weight for imitation logits.

  • estimation_step – the number of steps to look ahead.

  • target_update_freq – the target network update frequency (0 if you do not use the target network).

  • reward_normalization – normalize the returns to Normal(0, 1). TODO: rename to return_normalization?

  • is_double – use double dqn.

  • clip_loss_grad – clip the gradient of the loss in accordance with nature14236; this amounts to using the Huber loss instead of the MSE loss.

  • observation_space – Env’s observation space.

  • lr_scheduler – if not None, will be called in policy.update().

See also

Please refer to BasePolicy for more detailed explanation.

forward(batch: ObsBatchProtocol, state: dict | Batch | ndarray | None = None, **kwargs: Any) ImitationBatchProtocol[source]#

Compute action over the given batch data.

If you need to mask the action, please add a “mask” into batch.obs, for example, if we have an environment that has “0/1/2” three actions:

batch == Batch(
        obs="original obs, with batch_size=1 for demonstration",
        mask=np.array([[False, True, False]]),
        # action 1 is available
        # action 0 and 2 are unavailable

A Batch which has 3 keys:

  • act the action.

  • logits the network’s raw output.

  • state the hidden state.

See also

Please refer to forward() for more detailed explanation.

learn(batch: RolloutBatchProtocol, *args: Any, **kwargs: Any) TDiscreteBCQTrainingStats[source]#

Update policy with a given batch of data.


A dataclass object, including the data needed to be logged (e.g., loss).


In order to distinguish the collecting state, updating state and testing state, you can check the policy state by self.training and self.updating. Please refer to States for policy for more detailed explanation.


If you use torch.distributions.Normal and torch.distributions.Categorical to calculate the log_prob, please be careful about the shape: Categorical distribution gives “[batch_size]” shape while Normal distribution gives “[batch_size, 1]” shape. The auto-broadcasting of numerical operation with torch tensors will amplify this error.

train(mode: bool = True) Self[source]#

Set the module in training mode, except for the target network.

class DiscreteBCQTrainingStats(*, train_time: float = 0.0, smoothed_loss: dict = <factory>, loss: float, q_loss: float, i_loss: float, reg_loss: float)[source]#
i_loss: float#
q_loss: float#
reg_loss: float#