TensorDictMaxValueWriter

class torchrl.data.replay_buffers.TensorDictMaxValueWriter(rank_key=None, reduction: str = 'sum', **kwargs)

一个 Writer 类，用于可组合回放缓冲区，根据某个排序键保留最优元素。

参数：

• rank_key (str or tuple of str) – 用于对元素进行排序的键。默认为 ("next", "reward")。

• reduction (str) – 如果排序键包含多个元素，则使用的归约方法。可以是 "max"、"min"、"mean"、"median" 或 "sum"。

示例

>>> import torch
>>> from tensordict import TensorDict
>>> from torchrl.data import LazyTensorStorage, TensorDictReplayBuffer, TensorDictMaxValueWriter
>>> from torchrl.data.replay_buffers.samplers import SamplerWithoutReplacement
>>> rb = TensorDictReplayBuffer(
...     storage=LazyTensorStorage(1),
...     sampler=SamplerWithoutReplacement(),
...     batch_size=1,
...     writer=TensorDictMaxValueWriter(rank_key="key"),
... )
>>> td = TensorDict({
...     "key": torch.tensor(range(10)),
...     "obs": torch.tensor(range(10))
... }, batch_size=10)
>>> rb.extend(td)
>>> print(rb.sample().get("obs").item())
9
>>> td = TensorDict({
...     "key": torch.tensor(range(10, 20)),
...     "obs": torch.tensor(range(10, 20))
... }, batch_size=10)
>>> rb.extend(td)
>>> print(rb.sample().get("obs").item())
19
>>> td = TensorDict({
...     "key": torch.tensor(range(10)),
...     "obs": torch.tensor(range(10))
... }, batch_size=10)
>>> rb.extend(td)
>>> print(rb.sample().get("obs").item())
19

注意

这个类与多维存储不兼容。这并不意味着禁止存储轨迹，而是存储的轨迹必须按每个轨迹的基础存储。以下是一些该类的有效和无效用法示例。首先，一个扁平缓冲区，我们在其中存储单个转换

>>> from torchrl.data import TensorStorage
>>> # Simplest use case: data comes in 1d and is stored as such
>>> data = TensorDict({
...     "obs": torch.zeros(10, 3),
...     "reward": torch.zeros(10, 1),
... }, batch_size=[10])
>>> rb = TensorDictReplayBuffer(
...     storage=LazyTensorStorage(max_size=100),
...     writer=TensorDictMaxValueWriter(rank_key="reward")
... )
>>> # We initialize the buffer: a total of 100 *transitions* can be stored
>>> rb.extend(data)
>>> # Samples 5 *transitions* at random
>>> sample = rb.sample(5)
>>> assert sample.shape == (5,)

其次，一个存储轨迹的缓冲区。最大信号在每个批次中聚合（例如，每个 rolluot 的奖励被求和）

>>> # One can also store batches of data, each batch being a sub-trajectory
>>> env = ParallelEnv(2, lambda: GymEnv("Pendulum-v1"))
>>> # Get a batch of [2, 10] -- format is [Batch, Time]
>>> rollout = env.rollout(max_steps=10)
>>> rb = TensorDictReplayBuffer(
...     storage=LazyTensorStorage(max_size=100),
...     writer=TensorDictMaxValueWriter(rank_key="reward")
... )
>>> # We initialize the buffer: a total of 100 *trajectories* (!) can be stored
>>> rb.extend(rollout)
>>> # Sample 5 trajectories at random
>>> sample = rb.sample(5)
>>> assert sample.shape == (5, 10)

如果数据是批量形式的，但需要扁平缓冲区，我们可以在扩展缓冲区之前简单地扁平化数据

>>> rb = TensorDictReplayBuffer(
...     storage=LazyTensorStorage(max_size=100),
...     writer=TensorDictMaxValueWriter(rank_key="reward")
... )
>>> # We initialize the buffer: a total of 100 *transitions* can be stored
>>> rb.extend(rollout.reshape(-1))
>>> # Sample 5 trajectories at random
>>> sample = rb.sample(5)
>>> assert sample.shape == (5,)

无法创建一个沿时间维度扩展的缓冲区，这通常是使用批量轨迹缓冲区时推荐的方式。由于轨迹是重叠的，聚合奖励值并进行比较很困难，甚至不可能。这个构造函数无效（注意 ndim 参数）

>>> rb = TensorDictReplayBuffer(
...     storage=LazyTensorStorage(max_size=100, ndim=2),  # Breaks!
...     writer=TensorDictMaxValueWriter(rank_key="reward")
... )

add(data: Any) → int | torch.Tensor

在适当的索引处插入单个数据元素，并返回该索引。

传递给此模块的数据中的 rank_key 应被构造为 []。如果它具有更多维度，将使用 reduction 方法将其归约到单个值。

extend(data: TensorDictBase) → None

在适当的索引处插入一系列数据点。

传递给此模块的数据中的 rank_key 应被构造为 [B]。如果它具有更多维度，将使用 reduction 方法将其归约到单个值。

get_insert_index(data: Any) → int

返回应该插入数据的索引，如果数据不应被插入，则返回 None。