pytorch_pfn_extras.distributed.DistributedValidationSampler#

class pytorch_pfn_extras.distributed.DistributedValidationSampler(dataset, num_replicas=None, rank=None, shuffle=True, seed=0)#

Bases: Sampler

Distributed sampler without duplication

This sampler splits the input dataset to each worker process in distributed setup without allowing repetition. It is for evaluation purpose such as DistributedEvaluator. This does not guarantee each worker to get the same number of samples, so for training do not use this sampler (use PyTorch DistributedSampler instead).

Methods

__init__(dataset[, num_replicas, rank, ...])

__init__(dataset, num_replicas=None, rank=None, shuffle=True, seed=0)#
Parameters:

  • dataset (Sized) –

  • num_replicas (Optional[int]) –

  • rank (Optional[int]) –

  • shuffle (bool) –

  • seed (int) –

Return type:

None