Struct TripletSampler 

pub struct TripletSampler<S: SplitStore + EpochStateStore + SamplerStateStore + 'static> { /* private fields */ }

Sampler that draws anchors from a single shared epoch cursor and then selects chunks from those records. Ingestion happens on demand when sampling.

Implementations

impl<S: SplitStore + EpochStateStore + SamplerStateStore + 'static> TripletSampler<S>

pub fn new(config: SamplerConfig, split_store: Arc<S>) -> Self

Create a sampler from config and a split-state backend.

pub fn new_with_chunker( config: SamplerConfig, split_store: Arc<S>, chunker: Arc<dyn ChunkingAlgorithm>, ) -> Self

Create a sampler from config with a custom chunking implementation.

pub fn next_pair_batch_for_split( &self, split: SplitLabel, ) -> Result<SampleBatch, SamplerError>

Return an unweighted pair batch for split.

pub fn next_text_batch_for_split( &self, split: SplitLabel, ) -> Result<TextBatch, SamplerError>

Return an unweighted text batch for split.

pub fn next_triplet_batch_for_split( &self, split: SplitLabel, ) -> Result<TripletBatch, SamplerError>

Return an unweighted triplet batch for split.

pub fn next_pair_batch_with_weights_for_split( &self, split: SplitLabel, weights: &HashMap<SourceId, f32>, ) -> Result<SampleBatch, SamplerError>

Return a weighted pair batch for split using per-source weights.

pub fn next_text_batch_with_weights_for_split( &self, split: SplitLabel, weights: &HashMap<SourceId, f32>, ) -> Result<TextBatch, SamplerError>

Return a weighted text batch for split using per-source weights.

pub fn next_triplet_batch_with_weights_for_split( &self, split: SplitLabel, weights: &HashMap<SourceId, f32>, ) -> Result<TripletBatch, SamplerError>

Return a weighted triplet batch for split using per-source weights.

pub fn prefetch_triplet_batches( self: Arc<Self>, split: SplitLabel, capacity: usize, ) -> BatchPrefetcher<TripletBatch>

Spawn a background prefetcher for triplet batches.

pub fn prefetch_triplet_batches_with_weights( self: Arc<Self>, split: SplitLabel, capacity: usize, weights: HashMap<SourceId, f32>, ) -> BatchPrefetcher<TripletBatch>

Spawn a background prefetcher for weighted triplet batches.

pub fn prefetch_pair_batches( self: Arc<Self>, split: SplitLabel, capacity: usize, ) -> BatchPrefetcher<SampleBatch>

Spawn a background prefetcher for pair batches.

pub fn prefetch_pair_batches_with_weights( self: Arc<Self>, split: SplitLabel, capacity: usize, weights: HashMap<SourceId, f32>, ) -> BatchPrefetcher<SampleBatch>

Spawn a background prefetcher for weighted pair batches.

pub fn prefetch_text_batches( self: Arc<Self>, split: SplitLabel, capacity: usize, ) -> BatchPrefetcher<TextBatch>

Spawn a background prefetcher for text batches.

pub fn prefetch_text_batches_with_weights( self: Arc<Self>, split: SplitLabel, capacity: usize, weights: HashMap<SourceId, f32>, ) -> BatchPrefetcher<TextBatch>

Spawn a background prefetcher for weighted text batches.

pub fn text_recipes(&self) -> Vec<TextRecipe>

Return the currently active text recipes.

pub fn register_source( &self, source: Box<dyn DataSource + 'static>, ) -> Result<(), SamplerError>

Register a data source for ingestion and sampling.

Returns an error if the source’s id() matches the reserved __*__ pattern used for internal synthetic/metadata source identifiers.

pub fn set_epoch(&self, epoch: u64) -> Result<(), SamplerError>

Force sampler epoch to epoch (advanced deterministic replay control).

pub fn save_sampler_state( &self, save_to: Option<&Path>, ) -> Result<(), SamplerError>

Persist sampler and split runtime state for restart-resume.

When save_to is Some(path), current persisted runtime state is also mirrored to path when supported by the split-store backend.

Trait Implementations

impl<S: SplitStore + EpochStateStore + SamplerStateStore + 'static> Sampler for TripletSampler<S>

fn next_pair_batch( &self, split: SplitLabel, ) -> Result<SampleBatch, SamplerError>

Returns a batch of pairs. Consumes the shared epoch cursor for anchor selection.

fn next_pair_batch_with_weights( &self, split: SplitLabel, weights: &HashMap<SourceId, f32>, ) -> Result<SampleBatch, SamplerError>

Returns a batch of pairs with per-call source weights.

fn next_text_batch(&self, split: SplitLabel) -> Result<TextBatch, SamplerError>

Returns a batch of text samples. Consumes the shared epoch cursor for anchor selection.

fn next_text_batch_with_weights( &self, split: SplitLabel, weights: &HashMap<SourceId, f32>, ) -> Result<TextBatch, SamplerError>

Returns a batch of text samples with per-call source weights.

fn next_triplet_batch( &self, split: SplitLabel, ) -> Result<TripletBatch, SamplerError>

Returns a batch of triplets. Consumes the shared epoch cursor for anchor selection.

fn next_triplet_batch_with_weights( &self, split: SplitLabel, weights: &HashMap<SourceId, f32>, ) -> Result<TripletBatch, SamplerError>

Returns a batch of triplets with per-call source weights.