Struct EmbeddingAggregator 

pub struct EmbeddingAggregator {
    pub config: EmbeddingAggregatorConfig,
    pub method: AggregationMethod,
    pub history: VecDeque<AggregationResult>,
    pub max_history: usize,
}

Aggregates multiple embedding vectors into a single representation.

Maintains a rolling history of the most recent aggregation results for monitoring and introspection.

Fields

config: EmbeddingAggregatorConfig

Configuration flags.

method: AggregationMethod

The pooling method.

history: VecDeque<AggregationResult>

Rolling history of past results.

max_history: usize

Maximum number of entries kept in history.

Implementations

impl EmbeddingAggregator

pub fn new(method: AggregationMethod, config: EmbeddingAggregatorConfig) -> Self

Create a new EmbeddingAggregator with the given method and config.

The history capacity defaults to 1 000 entries.

pub fn with_history_capacity( method: AggregationMethod, config: EmbeddingAggregatorConfig, max_history: usize, ) -> Self

Create a new aggregator with an explicit history capacity.

pub fn aggregate( &mut self, inputs: &[AggregationInput], ) -> Result<AggregationResult, AggregatorError>

Aggregate a slice of AggregationInput values using self.method.

All inputs must share the same dimensionality. Returns an AggregationResult and records it in the history ring.

pub fn aggregate_raw( &mut self, embeddings: &[Vec<f64>], ) -> Result<AggregationResult, AggregatorError>

Convenience method that treats all embeddings with weight 1.0 and applies self.method.

pub fn aggregate_with_input_weights( &mut self, inputs: &[AggregationInput], ) -> Result<AggregationResult, AggregatorError>

Aggregate using the weight field of each AggregationInput directly, ignoring the method’s own weight vector. Useful when weights are determined dynamically (e.g. relevance scores).

pub fn merge_results( results: &[AggregationResult], method: AggregationMethod, ) -> Result<AggregationResult, AggregatorError>

Aggregate the output vectors from several AggregationResults using the provided method.

pub fn l2_normalize(v: &[f64]) -> Vec<f64>

L2-normalise v in place. Returns v unchanged if its norm is ≈ 0.

pub fn l2_norm(v: &[f64]) -> f64

Compute the L2 (Euclidean) norm of v.

pub fn cosine_similarity(a: &[f64], b: &[f64]) -> f64

Cosine similarity between two vectors.

Returns 0.0 if either vector is the zero vector.

pub fn attention_scores(query: &[f64], embeddings: &[Vec<f64>]) -> Vec<f64>

Compute softmax attention scores: softmax(query · embedding[i] / sqrt(d)).

pub fn recent_history(&self, n: usize) -> Vec<&AggregationResult>

Return the n most-recent AggregationResults (newest last).

pub fn stats(&self) -> EaAggregatorStats

Compute statistics over the full history.