lit_sb3.data.replay
[docs]
module
lit_sb3.data.replay
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53 | """PyTorch IterableDataset wrappers that bridge RL data collection.
with Lightning's DataLoader-based training loop.
"""
from __future__ import annotations
from collections.abc import Generator, Iterator
from itertools import count
from typing import TYPE_CHECKING
import torch
from torch.utils.data import IterableDataset
if TYPE_CHECKING:
from lit_sb3.buffers import ReplayBuffer, RolloutBuffer
from lit_sb3.core.types import RolloutBufferSamples
class ReplayBufferDataset(IterableDataset):
"""Finite IterableDataset for off-policy algorithms.
Yields a fixed number of dummy integer tokens. The actual data is NOT
provided by this dataset - instead, ``training_step`` handles replay buffer
sampling. This dataset simply drives the Lightning training loop for a
specified number of steps.
Args:
total_steps: Total number of training steps (= timer for entire learn()).
"""
def __init__(self, total_steps: int) -> None:
super().__init__()
self.total_steps = total_steps
def __iter__(self) -> Iterator[int]:
yield from range(self.total_steps)
def __len__(self) -> int:
return self.total_steps
class InfiniteReplayBufferDataset(IterableDataset):
"""Unbounded IterableDataset used when trainer-side RL budget control is authoritative.
Yields dummy integer tokens forever. The actual data is NOT provided by this
dataset - instead, ``training_step`` handles replay buffer sampling. This
dataset exists purely to keep Lightning advancing batches until the trainer
or callback-driven RL budget requests a stop.
"""
def __iter__(self) -> Iterator[int]:
yield from count()
|