Struct TensorDataLoader 

pub struct TensorDataLoader { /* private fields */ }

Batch data loader with deterministic shuffling and epoch tracking.

Samples are stored as (feature_vector, label) pairs. Each epoch iterates through all samples in (optionally shuffled) order, yielding fixed-size batches.

Implementations

impl TensorDataLoader

pub fn new(config: DataLoaderConfig, seed: u64) -> Self

Create a new data loader with the given config and PRNG seed.

pub fn add_sample(&mut self, data: Vec<f64>, label: &str)

Add a single training sample.

pub fn add_samples(&mut self, samples: Vec<(Vec<f64>, String)>)

Bulk-add training samples.

pub fn next_batch(&mut self) -> Option<DataBatch>

Return the next batch, or None if there are no samples.

When the current epoch is exhausted the loader automatically increments the epoch counter, reshuffles (if enabled), and begins yielding batches from the new epoch.

pub fn reset(&mut self)

Reset to the beginning of the current epoch without incrementing.

pub fn shuffle_indices(&mut self)

Perform a Fisher-Yates shuffle on self.indices using an FNV-1a-based PRNG seeded from seed + current_epoch.

pub fn total_samples(&self) -> usize

Total number of loaded samples.

pub fn total_batches(&self) -> usize

Total number of batches per epoch.

Returns ceil(samples / batch_size) normally, or floor(samples / batch_size) when drop_last is enabled.

pub fn current_epoch(&self) -> usize

Current epoch (zero-based).

pub fn progress(&self) -> f64

Fraction of the current epoch that has been consumed (0.0..=1.0).

pub fn stats(&self) -> DataLoaderStats

Snapshot of the loader’s current statistics.

Trait Implementations

impl Debug for TensorDataLoader

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.