pub struct SvdsOptions<T: Real = f64> {
pub which: SvdWhich,
pub ncv: Option<usize>,
pub tol: f64,
pub max_restarts: Option<usize>,
pub v0: Option<Vec<T>>,
pub seed: u64,
pub reorth_eta: f64,
pub vectors: bool,
pub allow_partial: bool,
}Expand description
Options of svds(). Defaults are scipy’s svds where it has one
(cited from memory), with ARPACK’s dimension rule.
Fields§
§which: SvdWhichWhich values (default SvdWhich::Largest).
ncv: Option<usize>Krylov dimension ncv (number of right Lanczos vectors kept before
a restart); None = min(min(m, n), max(2k + 1, 20)). Must satisfy
k < ncv ≤ min(m, n).
tol: f64τ of the convergence test (step B5); 0 (default) means
T::EPSILON. Absolute in units of σ̃_1, the largest Ritz value:
a relative test ≤ τ σ̃_i is unattainable for small σ_i and is
not offered.
max_restarts: Option<usize>Restart limit; None = 10 · min(m, n).
v0: Option<Vec<T>>Starting right vector p_1 ∈ R^n; None = Gaussian from
SvdsOptions::seed (Algorithm G).
seed: u64Seed of the starting vector and of the replacement vectors after a breakdown (default 0).
reorth_eta: f64DGKS threshold of both reorthogonalizations (default 1/√2).
vectors: boolCompute the singular vectors (default true); with false, u
and v of the result are 0-column blocks.
allow_partial: boolOn non-convergence return the converged triplets (Ok, fewer than
k) instead of StatlaError::NoConvergence (default false).