MCPcopy Create free account
hub / github.com/FreeformRobotics/OTS / DistributedSampler

Class DistributedSampler

lib/utils/data/distributed.py:7–58  ·  view source on GitHub ↗

Sampler that restricts data loading to a subset of the dataset. It is especially useful in conjunction with :class:`torch.nn.parallel.DistributedDataParallel`. In such case, each process can pass a DistributedSampler instance as a DataLoader sampler, and load a subset of the origina

Source from the content-addressed store, hash-verified

5
6
7class DistributedSampler(Sampler):
8 """Sampler that restricts data loading to a subset of the dataset.
9
10 It is especially useful in conjunction with
11 :class:`torch.nn.parallel.DistributedDataParallel`. In such case, each
12 process can pass a DistributedSampler instance as a DataLoader sampler,
13 and load a subset of the original dataset that is exclusive to it.
14
15 .. note::
16 Dataset is assumed to be of constant size.
17
18 Arguments:
19 dataset: Dataset used for sampling.
20 num_replicas (optional): Number of processes participating in
21 distributed training.
22 rank (optional): Rank of the current process within num_replicas.
23 """
24
25 def __init__(self, dataset, num_replicas=None, rank=None):
26 if num_replicas is None:
27 num_replicas = get_world_size()
28 if rank is None:
29 rank = get_rank()
30 self.dataset = dataset
31 self.num_replicas = num_replicas
32 self.rank = rank
33 self.epoch = 0
34 self.num_samples = int(math.ceil(len(self.dataset) * 1.0 / self.num_replicas))
35 self.total_size = self.num_samples * self.num_replicas
36
37 def __iter__(self):
38 # deterministically shuffle based on epoch
39 g = torch.Generator()
40 g.manual_seed(self.epoch)
41 indices = list(torch.randperm(len(self.dataset), generator=g))
42
43 # add extra samples to make it evenly divisible
44 indices += indices[:(self.total_size - len(indices))]
45 assert len(indices) == self.total_size
46
47 # subsample
48 offset = self.num_samples * self.rank
49 indices = indices[offset:offset + self.num_samples]
50 assert len(indices) == self.num_samples
51
52 return iter(indices)
53
54 def __len__(self):
55 return self.num_samples
56
57 def set_epoch(self, epoch):
58 self.epoch = epoch

Callers

nothing calls this directly

Calls

no outgoing calls

Tested by

no test coverage detected