Struct DataLoader 

pub struct DataLoader<'a, B: Backend> { /* private fields */ }

A DataLoader wraps a Dataset and produces batches of tensors.

Each batch is a HashMap<String, Tensor<B>> with keys "input" and "target", matching the format expected by Executor::run() and Trainer::train().

Implementations

impl<'a, B: Backend> DataLoader<'a, B>

pub fn new( dataset: &'a dyn Dataset, device: B::Device, config: DataLoaderConfig, ) -> Self

Create a new DataLoader over a dataset.

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

Add a transform to apply to each sample.

pub fn num_batches(&self) -> usize

The number of batches per epoch.

pub fn len(&self) -> usize

Total number of samples.

pub fn is_empty(&self) -> bool

Whether the dataset is empty.

pub fn reshuffle(&mut self)

Reshuffle indices (call at the start of each epoch).

pub fn epoch_batches( &mut self, input_name: &str, target_name: &str, ) -> Result<Vec<HashMap<String, Tensor<B>>>, Error>

Produce all batches for one epoch as a Vec of HashMap tensors.

Each HashMap has keys matching the input/target names you’ll pass to the executor. By default: "input" and "target".

pub fn iter_batches( &mut self, input_name: &str, target_name: &str, ) -> BatchIterator<'_, 'a, B>

Iterate over batches one at a time (lower memory than epoch_batches).