Struct AsyncDataLoader 

pub struct AsyncDataLoader<B: Backend> { /* private fields */ }

A data loader that prefetches batches on background threads.

On each call to iter_epoch, the loader:

1. Optionally reshuffles indices.
2. Spawns worker threads that load, transform, and collate batches.
3. Returns an iterator that pulls ready batches from a bounded channel.

The channel capacity is prefetch_factor * max(num_workers, 1), so at most that many batches are materialised in memory at any time.

The dataset is held via Arc<dyn Dataset> so it can be safely shared with background worker threads.

Implementations

impl<B: Backend> AsyncDataLoader<B>

where B::Device: Clone + Send + Sync + 'static,

pub fn new( dataset: Arc<dyn Dataset>, device: B::Device, config: AsyncDataLoaderConfig, ) -> Self

Create a new async data loader.

pub fn with_transform(self, t: Arc<dyn Transform>) -> Self

Add a transform.

pub fn num_batches(&self) -> usize

Number of batches per epoch.

pub fn reshuffle(&mut self)

Reshuffle indices.

pub fn iter_epoch( &mut self, input_name: &str, target_name: &str, ) -> PrefetchIterator<B>

Iterate over one epoch of prefetched batches.

Spawns background workers that load batches into a bounded channel. The returned iterator yields Result<Batch<B>> — one per batch.

The background workers are joined when the iterator is dropped.