Compute a truncated SVD of an aMatrix — svd

Computes a rank-k truncated singular value decomposition of X, dispatching to the most suitable algorithm and backend based on matrix size, requested rank, and available hardware.

Usage

svd_factor(
  X,
  k = min(dim(X)),
  method = c("auto", "exact", "rsvd", "subspace"),
  n_oversamples = .amatrix_svd_factor_default_oversamples(k),
  n_iter = 2L
)

Arguments

X: An aMatrix (typically adgeMatrix or adgCMatrix).
k: Positive integer; the number of singular values and vectors to compute. Defaults to min(dim(X)).
method: One of "auto" (default), "exact", "rsvd", or "subspace". "auto" selects the algorithm based on matrix dimensions and the available backend.
n_oversamples: Non-negative integer; extra random vectors used to stabilize randomized algorithms. Ignored for method = "exact".
n_iter: Non-negative integer; number of power iterations for the subspace method. Ignored for method = "exact".

Value

An amSVD object containing slots u, d, v, and metadata.

Examples

m <- matrix(rnorm(30), nrow = 6)
A <- adgeMatrix(m)
fac <- svd_factor(A, k = 3L)
fac
#> amSVD [6x5 -> rank 3 | strict | exact/exact_svd@cpu | source: 20260702132121.033621-1cdf1817:am:117]
#>   d[1:min(3,k)]: 4.4865, 3.2935, 2.2253