Module matrix 

Sparse operands and the traits the solvers consume.

Index widths

SvdMat defaults to u32 column indices with u64 row pointers. Against usize-everywhere that is 12 bytes per non-zero instead of 16 for f64 data (8 instead of 16 for f32), and the separate pointer width keeps matrices with more than u32::MAX non-zeros representable. Callers who need wider indices can name them: SvdMat<f64, u64, u64>.

Re-exports

pub use kernels::DEFAULT_SCRATCH_BUDGET;

pub use masked::MaskedCsMat;

Modules

kernels

Parallel sparse × dense kernels.

masked

A submatrix view over a sparse matrix.

Traits

SparseMat

The operand interface the Krylov solvers (crate::lanczos, crate::irlba) need: shape and a matrix-vector product.

SparseMatDense

Blocked products, needed by the randomized solvers.

Functions

apply_centering

Subtract the rank-1 centering term from an uncentered product. Public so an implementor writing a fused mul_dense_centered can reuse it.

total_squared_norm

‖A‖²_F, or the centered version when means is supplied. What the solvers store in SvdRec::total_squared_norm.

Type Aliases

SvdMat

An owned sparse matrix. Defaults to u32 indices and u64 row pointers.

SvdMatView

A borrowed sparse matrix. Defaults to u32 indices and u64 row pointers.