Struct LowRankApproximation 

pub struct LowRankApproximation { /* private fields */ }

High-level API for low-rank matrix approximation.

Implementations

impl LowRankApproximation

pub fn new(config: LowRankConfig) -> Self

Create a new approximation engine with the given configuration.

pub fn approximate_matrix( &self, data: &[f64], rows: usize, cols: usize, ) -> Result<SvdResult, LowRankError>

Approximate a 2-D matrix stored in row-major order.

pub fn approximate_matmul( &self, a: &[f64], a_rows: usize, a_cols: usize, b: &[f64], b_rows: usize, b_cols: usize, ) -> Result<Vec<f64>, LowRankError>

Approximate the matrix product A @ B using low-rank factors of A.

Rather than computing the full product C = A B (which is O(a_rows · a_cols · b_cols)), we approximate A ≈ U Σ Vᵀ (rank-k) and then compute C ≈ (U Σ) (Vᵀ B). This can be cheaper when rank << min(a_rows, a_cols).

pub fn is_candidate(&self, rows: usize, cols: usize) -> bool

Return true if this matrix is large enough to be a candidate for low-rank approximation (based on min_dimension in the config).

pub fn optimal_rank(singular_values: &[f64], energy_threshold: f64) -> usize

Compute the smallest rank k such that the top-k singular values capture at least energy_threshold fraction of the total singular energy.

energy_threshold should be in [0, 1]. If the slice is empty the function returns 0.