Struct EmbeddingAggregator 

pub struct EmbeddingAggregator { /* private fields */ }

Stateful embedding aggregator that tracks total aggregation operations.

Implementations

impl EmbeddingAggregator

pub fn new(config: AggregatorConfig) -> Self

Create a new aggregator with the given configuration.

pub fn aggregate(&mut self, tokens: &[Vec<f32>]) -> Option<AggregatedEmbedding>

Aggregate a sequence of token embeddings using the default strategy.

Each inner Vec<f32> is a single token’s embedding. All token embeddings must have the same dimensionality.

pub fn aggregate_with( &mut self, tokens: &[Vec<f32>], strategy: PoolingStrategy, attention_weights: Option<&[f32]>, ) -> Option<AggregatedEmbedding>

Aggregate using a specific strategy.

attention_weights is required when strategy == AttentionWeighted and is ignored for other strategies.

pub fn aggregate_batch(&mut self, batch: &[Vec<Vec<f32>>]) -> BatchResult

Aggregate a batch of token sequences using the default strategy.

pub fn aggregate_batch_with( &mut self, batch: &[Vec<Vec<f32>>], strategy: PoolingStrategy, ) -> BatchResult

Aggregate a batch with a specific strategy.

pub fn hierarchical_aggregate( &mut self, sentences: &[Vec<Vec<f32>>], paragraph_boundaries: &[usize], ) -> Option<HierarchicalResult>

Perform hierarchical aggregation: sentence -> paragraph -> document.

• sentences – each entry is a sequence of token embeddings forming one sentence.
• paragraph_boundaries – indices into sentences where a new paragraph starts (e.g. [0, 3, 7] means sentences 0..3 form paragraph 0, 3..7 form paragraph 1, etc.).

Uses mean pooling at every level.

pub fn compare_strategies( &mut self, tokens: &[Vec<f32>], strategy_a: PoolingStrategy, strategy_b: PoolingStrategy, ) -> (Option<AggregatedEmbedding>, Option<AggregatedEmbedding>)

Compare two pooling strategies on the same tokens and return both results.

pub fn total_aggregations(&self) -> u64

Return the total number of individual aggregation operations performed.

pub fn config(&self) -> &AggregatorConfig

Return the current configuration.

pub fn strategy_summary( results: &[AggregatedEmbedding], ) -> HashMap<PoolingStrategy, usize>

Build a summary of aggregation results per strategy from provided labels.