Struct EmbeddingPipelineManager 

pub struct EmbeddingPipelineManager { /* private fields */ }

Multi-stage embedding pipeline manager.

Supports both text-input (process_text) and pre-computed embedding input (process_embeddings). Thread-safe — the internal statistics are protected by a Mutex.

Implementations

impl EmbeddingPipelineManager

pub fn new(config: EpmPipelineConfig) -> Result<Self, EpmPipelineError>

Create a new manager from the given config.

pub fn process_text( &self, ids: Vec<String>, texts: Vec<String>, corpus: Option<&[String]>, ) -> Result<EmbeddingBatch, EpmPipelineError>

Process a batch of raw text strings through all pipeline stages.

corpus provides additional documents for IDF estimation; when None the current batch is used as the corpus.

pub fn process_embeddings( &self, ids: Vec<String>, embeddings: Vec<Vec<f64>>, ) -> Result<EmbeddingBatch, EpmPipelineError>

Process pre-computed embeddings through the non-text pipeline stages.

pub fn add_stage( &mut self, stage: EpmPipelineStage, ) -> Result<(), EpmPipelineError>

Append a stage to the pipeline, then re-validate.

pub fn remove_stage(&mut self, index: usize) -> Result<(), EpmPipelineError>

Remove the stage at index.

pub fn validate_config(&self) -> Result<(), EpmPipelineError>

Validate the current stage configuration for ordering correctness.

Rules enforced:

• TfIdfWeighting must be preceded by at least one of Tokenize, StopWordFilter, or NGram.
• DimensionReduce with target_dim == 0 is rejected.
• NGram with n == 0 is rejected.

pub fn benchmark(&self, texts: &[String], n_runs: usize) -> Vec<StageTiming>

Run each stage independently on texts for n_runs repetitions and return per-stage timing information.

pub fn stats(&self) -> EpmPipelineStats

Return a snapshot of cumulative statistics.

pub fn config(&self) -> &EpmPipelineConfig

Immutable reference to the current config.