espnet2.iterators.category_iter_factory.CategoryIterFactory

source

class espnet2.iterators.category_iter_factory.CategoryIterFactory(dataset, batches: AbsSampler | Sequence[Sequence[Any]], num_iters_per_epoch: int = None, seed: int = 0, sampler_args: dict = None, shuffle: bool = False, num_workers: int = 0, collate_fn=None, pin_memory: bool = False)

Bases: AbsIterFactory

Build iterator for each epoch.

This class creates a PyTorch DataLoader with the following features:

• The random seed is determined by the epoch number, ensuring reproducibility

when resuming training.

• The number of samples for each epoch can be restricted, controlling the interval between training and evaluation.

dataset

The dataset to be used by the DataLoader.

• Type: Any

sampler

An instance of a sampler that defines how to sample data from the dataset.

• Type:AbsSampler

num_iters_per_epoch

The number of iterations to run per epoch.

• Type: int

sampler

_args

Arguments to configure the sampler.

• Type: dict

shuffle

Whether to shuffle the data at the beginning of each epoch.

• Type: bool

seed

The seed for random number generation.

• Type: int

num_workers

The number of subprocesses to use for data loading.

• Type: int

collate_fn

Function to merge a list of samples to form a mini-batch.

• Type: callable

pin_memory

If True, the data loader will copy Tensors into CUDA pinned memory before returning them.

• Type: bool

• Parameters:

  • dataset (Any) – The dataset from which to load the data.
  • batches (Union [AbsSampler , Sequence *[*Sequence *[*Any ] ] ]) – Either a sampler or a sequence of batches to use for sampling.
  • num_iters_per_epoch (int , optional) – The number of iterations to run per epoch. Defaults to None.
  • seed (int , optional) – The random seed. Defaults to 0.
  • sampler_args (dict , optional) – Additional arguments for the sampler. Defaults to None.
  • shuffle (bool , optional) – Whether to shuffle the dataset. Defaults to False.
  • num_workers (int , optional) – Number of worker processes for data loading. Defaults to 0.
  • collate_fn (callable , optional) – Function to merge a list of samples to form a mini-batch. Defaults to None.
  • pin_memory (bool , optional) – If True, data loader will copy Tensors into CUDA pinned memory. Defaults to False.

• Returns: A PyTorch DataLoader instance configured for the dataset and sampler.

• Return type: DataLoader

####### Examples

Create an instance of CategoryIterFactory

iter_factory = CategoryIterFactory(

dataset=my_dataset, batches=my_batches, num_iters_per_epoch=100, seed=42, shuffle=True

)

Build the DataLoader for epoch 1

data_loader = iter_factory.build_iter(epoch=1)

NOTE

This class is designed to work with PyTorch DataLoader and provides additional control over the sampling process.

• Raises:
  • RuntimeError – If the batch size is less than the world size during
  • distributed training. –

build_iter(epoch: int, shuffle: bool | None = None) → DataLoader

Build iterator for each epoch.

This class simply creates a PyTorch DataLoader with the following features:

• The random seed is determined according to the number of epochs,

ensuring reproducibility when resuming training.

• It allows restriction on the number of samples for one epoch, controlling the interval between training and evaluation.

dataset

The dataset to be loaded.

• Type: Any

sampler

The sampler to sample batches from the dataset.

• Type:AbsSampler

num_iters_per_epoch

The number of iterations per epoch.

• Type: int, optional

sampler

_args

Arguments for the sampler.

• Type: dict, optional

shuffle

Whether to shuffle the data at every epoch.

• Type: bool

seed

Random seed for reproducibility.

• Type: int

num_workers

Number of workers for data loading.

• Type: int

collate_fn

Function to collate data samples.

• Type: callable, optional

pin_memory

Whether to pin memory for faster data transfer.

• Type: bool

• Parameters:

  • dataset (Any) – The dataset to be used for the iterator.
  • batches (Union [AbsSampler , Sequence *[*Sequence *[*Any ] ] ]) – Batches of data or an instance of AbsSampler.
  • num_iters_per_epoch (int , optional) – Number of iterations per epoch.
  • seed (int , optional) – Seed for random number generation. Default is 0.
  • sampler_args (dict , optional) – Additional arguments for the sampler.
  • shuffle (bool , optional) – Whether to shuffle the dataset. Default is False.
  • num_workers (int , optional) – Number of worker threads for loading data.
  • collate_fn (callable , optional) – Function to merge a list of samples into a batch.
  • pin_memory (bool , optional) – If True, the data loader will copy tensors into CUDA pinned memory before returning them.

• Returns: A PyTorch DataLoader for the specified dataset and batches.

• Return type: DataLoader

####### Examples

factory = CategoryIterFactory(dataset=my_dataset, batches=my_batches, : num_iters_per_epoch=100, seed=42)

dataloader = factory.build_iter(epoch=1, shuffle=True)

• Raises:RuntimeError – If the batch size is less than the world size in distributed training.