Struct SemanticDimensionReducer 

pub struct SemanticDimensionReducer { /* private fields */ }

Semantic dimension reducer supporting random projection, PCA, and truncation.

Implementations

impl SemanticDimensionReducer

pub fn new(config: ReducerConfig) -> Self

Create a new dimension reducer with the given configuration.

pub fn fit(&mut self, embeddings: &[Vec<f64>]) -> Result<(), String>

Fit the reducer to the given embeddings.

• RandomProjection: generates a Gaussian random matrix from the seed
• PCA: computes top eigenvectors via power iteration
• Truncation: no-op (always ready)

pub fn transform(&mut self, embedding: &[f64]) -> Result<Vec<f64>, String>

Transform a single embedding to lower dimensionality.

pub fn fit_transform( &mut self, embeddings: &[Vec<f64>], ) -> Result<Vec<Vec<f64>>, String>

Fit the reducer and transform all embeddings in one step.

pub fn reconstruction_error(&self, original: &[f64], reduced: &[f64]) -> f64

Compute the reconstruction error (MSE) between the original and back-projected embedding.

For RandomProjection, uses pseudo-inverse (transpose for column-normalized matrices). For PCA, uses transpose of components. For Truncation, pads with zeros.

pub fn is_fitted(&self) -> bool

Whether the reducer has been fitted.

pub fn reset(&mut self)

Reset the reducer, clearing the projection matrix and fitted state.

pub fn stats(&self) -> ReducerStats

Get statistics about the reducer.