Module sparse 

Sparse matrix types: CSR and COO representations. Sparse matrix representations and arithmetic.

Provides Compressed Sparse Row (SparseCsr) and Coordinate (SparseCoo) formats, plus element-wise and matrix-matrix operations.

Determinism Contract

• Row-wise reductions use [kahan_sum_f64] or [binned_sum_f64].
• [sparse_matmul] accumulates column contributions in a [BTreeMap] for deterministic iteration order, then reduces each column with [binned_sum_f64].
• No HashMap, no parallel iteration.

Structs

SparseCoo

Coordinate (COO / triplet) sparse matrix representation.

SparseCsr

Compressed Sparse Row (CSR) matrix representation.

Functions

sparse_add

Element-wise addition of two CSR matrices (same dimensions).

sparse_matmul

Sparse matrix-matrix multiplication (SpGEMM): C = A * B. Uses row-wise accumulation with BTreeMap for deterministic column ordering. All floating-point reductions use binned summation.

sparse_mul

Element-wise multiplication (Hadamard product) of two CSR matrices. Only positions where BOTH matrices have non-zeros produce non-zeros.

sparse_scalar_mul

Scalar multiplication: every non-zero element is multiplied by s.

sparse_sub

Element-wise subtraction of two CSR matrices (same dimensions).

sparse_transpose

Transpose a CSR matrix. Returns a new CSR where rows and columns are swapped.