DiscreteIQLLoss

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

TorchRL 對離散 IQL 損失函式的實現。

選自論文“Offline Reinforcement Learning with Implicit Q-Learning” https://arxiv.org/abs/2110.06169

引數:

• actor_network (ProbabilisticActor) – 隨機策略網路

• 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 "one-hot", "mult_one_hot", "binary" 或 "categorical" 中的一個，或是相應 spec (torchrl.data.OneHot, torchrl.data.MultiOneHot, torchrl.data.Binary 或 torchrl.data.Categorical) 的例項。

• num_qvalue_nets (integer, optional) – 使用的 Q-Value 網路數量。預設為 2。

• loss_function (str, optional) – 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，目標函式的行為類似於行為克隆；而對於較大的值，它試圖恢復 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) 代替] 用於寫入優先順序的 tensordict 鍵（用於優先經驗回放緩衝區）。預設為 “td_error”。

• separate_losses (bool, optional) – 如果為 True，策略和 critic 之間的共享引數將僅透過策略損失進行訓練。預設為 False，即梯度會傳播到策略和 critic 損失的共享引數。

• reduction (str, optional) – 指定應用於輸出的歸約方式："none" | "mean" | "sum"。"none"：不應用歸約；"mean"：輸出的總和將除以輸出中的元素數量；"sum"：輸出將被求和。預設值："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)

此類也相容非基於 tensordict 的模組，並且無需使用任何 tensordict 相關原語即可使用。在這種情況下，預期的關鍵字引數為：["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()

輸出鍵也可以使用 DiscreteIQLLoss.select_out_keys() 方法進行過濾。

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

_AcceptedKeys 的別名

forward(tensordict: TensorDictBase = None) → TensorDictBase

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

將損失分解到其組成部分，然後訓練器可以使用這些元件來記錄訓練過程中的各種損失值。輸出 tensordict 中存在的其他標量值也將被記錄。

引數:

tensordict – 包含計算損失所需值的輸入 tensordict。

返回值:

一個不包含批處理維度的新 tensordict，其中包含各種損失標量，這些標量將命名為“loss*”。將損失值以這個名稱返回至關重要，因為訓練器會在反向傳播之前讀取它們。