torchtnt.utils.data.InOrderIterator

class torchtnt.utils.data.InOrderIterator(individual_dataloaders: Mapping[str, Union[DataLoader, Iterable]], iteration_strategy: InOrder)

InOrderIterator returns all batches from a single dataset till it is exhausted and then moves to the next one.

By default, the order is same as the keys of the input dataloader dict. This can be overridden to provide custom order. Repetition is supported.

Returns batches of the format: {dataloader_name: batch_from_dataloader}

Parameters:

• individual_dataloaders (Mapping[str, Union[DataLoader, Iterable]]) – A mapping of DataLoaders or Iterables with dataloader name as key
• value. (and dataloader/iterable object as) –
• iteration_strategy (RandomizedBatchSampler) – A RandomizedBatchSampler dataclass indicating how the dataloaders are iterated over.

Examples

>>> loaders = {'a': torch.utils.data.DataLoader(range(4), batch_size=4),
    'b': torch.utils.data.DataLoader(range(15), batch_size=5)}
>>> in_order_strategy = InOrder()
>>> combined_iterator = RandomizedBatchSamplerIterator(loaders, in_order_strategy)
>>> for item in combined_iterator:
        print(item)
{'a': tensor([0, 1, 2, 3])}
{'b': tensor([0, 1, 2, 3, 4])}
{'b': tensor([5, 6, 7, 8, 9])}
{'b': tensor([10, 11, 12, 13, 14])}

__init__(individual_dataloaders: Mapping[str, Union[DataLoader, Iterable]], iteration_strategy: InOrder) → None

Methods

__init__(individual_dataloaders, ...)