Attention

June 2024 Status Update: Removing DataPipes and DataLoader V2

We are re-focusing the torchdata repo to be an iterative enhancement of torch.utils.data.DataLoader. We do not plan on continuing development or maintaining the [DataPipes] and [DataLoaderV2] solutions, and they will be removed from the torchdata repo. We’ll also be revisiting the DataPipes references in pytorch/pytorch. In release torchdata==0.8.0 (July 2024) they will be marked as deprecated, and in 0.9.0 (Oct 2024) they will be deleted. Existing users are advised to pin to torchdata==0.8.0 or an older version until they are able to migrate away. Subsequent releases will not include DataPipes or DataLoaderV2. Please reach out if you suggestions or comments (please use this issue for feedback)

BatchMapper

class torchdata.datapipes.iter.BatchMapper(datapipe: IterDataPipe, fn: Callable, batch_size: int, input_col=None)

Combines elements from the source DataPipe to batches and applies a function over each batch, then flattens the outputs to a single, unnested IterDataPipe (functional name: map_batches).

Parameters:

• datapipe – Source IterDataPipe

• fn – The function to be applied to each batch of data

• batch_size – The size of batch to be aggregated from datapipe

• input_col –

  Index or indices of data which fn is applied, such as:

  • None as default to apply fn to the data directly.

  • Integer(s) is used for list/tuple.

  • Key(s) is used for dict.

Example

>>> from torchdata.datapipes.iter import IterableWrapper
>>> def fn(batch):
>>>     return [d + 1 for d in batch]
>>> source_dp = IterableWrapper(list(range(5)))
>>> mapped_dp = source_dp.map_batches(fn, batch_size=3)
>>> list(mapped_dp)
[1, 2, 3, 4, 5]

Notes

Compared with map, the reason that map_batches doesn’t take output_col argument is the size of fn output is not guaranteed to be the same as input batch. With different size, this operation cannot assign data back to original data structure.

And, this operation is introduced based on the use case from TorchText. A pybinded C++ vectorized function can be applied for efficiency.