DiscreteIQLLoss

class torchrl.objectives.DiscreteIQLLoss(*args, **kwargs)

離散 IQL 損失的 TorchRL 實現。

在“Implicit Q-Learning 的離線強化學習”中提出 https://arxiv.org/abs/2110.06169

引數:

• actor_network (ProbabilisticActor) – 隨機策略

• qvalue_network (TensorDictModule) – Q(s, a) 引數化模型。

• value_network (TensorDictModule, optional) – V(s) 引數化模型。

關鍵字引數:

• action_space (str or TensorSpec) – 動作空間。必須是以下之一 "one-hot"、"mult_one_hot"、"binary" 或 "categorical"，或者相應 spec 的例項 (torchrl.data.OneHot、torchrl.data.MultiOneHot、torchrl.data.Binary 或 torchrl.data.Categorical)。

• num_qvalue_nets (integer, optional) – 使用的 Q 值網路的數量。預設為 2。

• loss_function (str, optional) – 要用於值函式損失的損失函式。預設為 “smooth_l1”。

• temperature (float, optional) – 逆溫度 (beta)。對於較小的超引數值，目標函式類似於行為克隆，而對於較大的值，它則試圖恢復 Q 函式的最大值。

• expectile (float, optional) – expectile \(\tau\)。較大的 \(\tau\) 值對於需要動態規劃（“stichting”）的 antmaze 任務至關重要。

• priority_key (str, optional) – [已棄用，請改用 .set_keys(priority_key=priority_key)] 用於寫入優先順序（用於優先重放緩衝區）的 tensordict 鍵。預設為 “td_error”。

• separate_losses (bool, 可選) – 如果為 True，則策略和評估器之間的共享引數將僅針對策略損失進行訓練。預設為 False，即梯度將傳播到策略和評估器損失的共享引數。

• reduction (str, optional) – 指定應用於輸出的約簡："none" | "mean" | "sum"。"none"：不應用約簡，"mean"：輸出的總和將除以輸出中的元素數量，"sum"：將對輸出進行求和。預設為 "mean"。

示例

>>> import torch
>>> from torch import nn
>>> from torchrl.data.tensor_specs import OneHot
>>> from torchrl.modules.distributions.discrete import OneHotCategorical
>>> from torchrl.modules.tensordict_module.actors import ProbabilisticActor
>>> from torchrl.modules.tensordict_module.common import SafeModule
>>> from torchrl.objectives.iql import DiscreteIQLLoss
>>> from tensordict import TensorDict
>>> n_act, n_obs = 4, 3
>>> spec = OneHot(n_act)
>>> module = SafeModule(nn.Linear(n_obs, n_act), in_keys=["observation"], out_keys=["logits"])
>>> actor = ProbabilisticActor(
...     module=module,
...     in_keys=["logits"],
...     out_keys=["action"],
...     spec=spec,
...     distribution_class=OneHotCategorical)
>>> qvalue = SafeModule(
...     nn.Linear(n_obs, n_act),
...     in_keys=["observation"],
...     out_keys=["state_action_value"],
... )
>>> value = SafeModule(
...     nn.Linear(n_obs, 1),
...     in_keys=["observation"],
...     out_keys=["state_value"],
... )
>>> loss = DiscreteIQLLoss(actor, qvalue, value)
>>> batch = [2, ]
>>> action = spec.rand(batch).long()
>>> data = TensorDict({
...         "observation": torch.randn(*batch, n_obs),
...         "action": action,
...         ("next", "done"): torch.zeros(*batch, 1, dtype=torch.bool),
...         ("next", "terminated"): torch.zeros(*batch, 1, dtype=torch.bool),
...         ("next", "reward"): torch.randn(*batch, 1),
...         ("next", "observation"): torch.randn(*batch, n_obs),
...     }, batch)
>>> loss(data)
TensorDict(
    fields={
        entropy: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.float32, is_shared=False),
        loss_actor: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.float32, is_shared=False),
        loss_qvalue: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.float32, is_shared=False),
        loss_value: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.float32, is_shared=False)},
    batch_size=torch.Size([]),
    device=None,
    is_shared=False)

此類也相容非 tensordict 基於的模組，並且可以在不依賴任何 tensordict 相關原語的情況下使用。在這種情況下，預期的關鍵字引數是：["action", "next_reward", "next_done", "next_terminated"] + actor、value 和 qvalue 網路的 in_keys。返回值是一個按以下順序排列的張量元組：["loss_actor", "loss_qvalue", "loss_value", "entropy"]。

示例

>>> import torch
>>> import torch
>>> from torch import nn
>>> from torchrl.data.tensor_specs import OneHot
>>> from torchrl.modules.distributions.discrete import OneHotCategorical
>>> from torchrl.modules.tensordict_module.actors import ProbabilisticActor
>>> from torchrl.modules.tensordict_module.common import SafeModule
>>> from torchrl.objectives.iql import DiscreteIQLLoss
>>> _ = torch.manual_seed(42)
>>> n_act, n_obs = 4, 3
>>> spec = OneHot(n_act)
>>> module = SafeModule(nn.Linear(n_obs, n_act), in_keys=["observation"], out_keys=["logits"])
>>> actor = ProbabilisticActor(
...     module=module,
...     in_keys=["logits"],
...     out_keys=["action"],
...     spec=spec,
...     distribution_class=OneHotCategorical)
>>> qvalue = SafeModule(
...     nn.Linear(n_obs, n_act),
...     in_keys=["observation"],
...     out_keys=["state_action_value"],
... )
>>> value = SafeModule(
...     nn.Linear(n_obs, 1),
...     in_keys=["observation"],
...     out_keys=["state_value"],
... )
>>> loss = DiscreteIQLLoss(actor, qvalue, value)
>>> batch = [2, ]
>>> action = spec.rand(batch).long()
>>> loss_actor, loss_qvalue, loss_value, entropy = loss(
...     observation=torch.randn(*batch, n_obs),
...     action=action,
...     next_done=torch.zeros(*batch, 1, dtype=torch.bool),
...     next_terminated=torch.zeros(*batch, 1, dtype=torch.bool),
...     next_observation=torch.zeros(*batch, n_obs),
...     next_reward=torch.randn(*batch, 1))
>>> loss_actor.backward()

也可以使用 DiscreteIQLLoss.select_out_keys() 方法過濾輸出鍵。

示例

>>> _ = loss.select_out_keys('loss_actor', 'loss_qvalue', 'loss_value')
>>> loss_actor, loss_qvalue, loss_value = loss(
...     observation=torch.randn(*batch, n_obs),
...     action=action,
...     next_done=torch.zeros(*batch, 1, dtype=torch.bool),
...     next_terminated=torch.zeros(*batch, 1, dtype=torch.bool),
...     next_observation=torch.zeros(*batch, n_obs),
...     next_reward=torch.randn(*batch, 1))
>>> loss_actor.backward()

default_keys

別名：_AcceptedKeys

forward(tensordict: TensorDictBase = None) → TensorDictBase

它旨在讀取一個輸入的 TensorDict 並返回另一個包含名為“loss*”的損失鍵的 tensordict。

將損失分解為其組成部分可以被訓練器用於在訓練過程中記錄各種損失值。輸出 tensordict 中存在的其他標量也將被記錄。

引數:

tensordict – 一個輸入的 tensordict，包含計算損失所需的值。

返回:

一個沒有批處理維度的新 tensordict，其中包含各種損失標量，這些標量將被命名為“loss*”。重要的是，損失必須以這個名稱返回，因為它們將在反向傳播之前被訓練器讀取。