Struct CorpusWorkspace 

pub struct CorpusWorkspace<T> { /* private fields */ }

A corpus prepared once for repeated scoring under a fixed Metric.

Build it with CorpusWorkspace::build, then score many queries with query_corpus_scores / query_corpus_scores_into or rerank a single query with rerank_with. The corpus and metric are fixed for the life of the workspace.

Implementations

impl<T: NabledReal> CorpusWorkspace<T>

pub fn build( corpus: &ArrayView2<'_, T>, metric: Metric, ) -> Result<Self, EmbeddingError>

Precompute reusable corpus state for metric.

For Metric::Cosine this computes and caches the corpus row norms (rejecting zero-norm rows up front, exactly as the stateless cosine path would).

Errors

Returns EmbeddingError::EmptyInput when corpus is empty and EmbeddingError::ZeroNorm when a cosine corpus contains a zero-norm row.

pub const fn metric(&self) -> Metric

The metric this workspace was built for.

pub fn len(&self) -> usize

Number of corpus rows.

pub fn is_empty(&self) -> bool

Whether the corpus has no rows. Always false for a successfully built workspace.

pub fn dim(&self) -> usize

Feature dimension of the corpus rows.

pub fn query_corpus_scores( &self, queries: &ArrayView2<'_, T>, ) -> Result<Array2<T>, EmbeddingError>

Score every queries row against the cached corpus, reusing precomputed corpus state.

Returns a (queries.nrows(), corpus_len) matrix identical to the stateless query_corpus_scores result.

Errors

Returns EmbeddingError::EmptyInput for empty queries, EmbeddingError::DimensionMismatch when the feature dimension differs, and EmbeddingError::ZeroNorm for cosine queries with zero-norm rows.

pub fn query_corpus_scores_into( &self, queries: &ArrayView2<'_, T>, out: &mut Array2<T>, ) -> Result<(), EmbeddingError>

Score every queries row against the cached corpus into a caller-provided out buffer.

out must be shaped (queries.nrows(), corpus_len).

Errors

See query_corpus_scores; also returns EmbeddingError::DimensionMismatch when out is mis-sized.

pub fn rerank_with( &self, query: &ArrayView1<'_, T>, k: usize, ) -> Result<Vec<Neighbor<T>>, EmbeddingError>

Rerank the cached corpus against a single query, returning the best k neighbors.

Equivalent to rerank against this workspace’s corpus and metric, but reusing the precomputed corpus state.

Errors

See query_corpus_scores.

pub fn knn_with( &self, queries: &ArrayView2<'_, T>, k: usize, ) -> Result<Vec<Vec<Neighbor<T>>>, EmbeddingError>

Exact brute-force kNN over the cached corpus for every queries row.

Per-query best-first neighbor lists, reusing the precomputed corpus state.

Errors

See query_corpus_scores.

Trait Implementations

impl<T: Clone> Clone for CorpusWorkspace<T>

fn clone(&self) -> CorpusWorkspace<T>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<T: Debug> Debug for CorpusWorkspace<T>

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

Formats the value using the given formatter.