lazymatrix::traits

LazyMatrix for Rust · 0.8.1

Source

pub mod traits

Backend-agnostic trait surface for LazyMatrix.

The traits separate numeric operations from matrix capabilities:

• Scalar — the numeric element type, a blanket-implemented bundle of num-traits bounds.
• MatrixErrorType — the shared operational error type for a backend.
• MatrixShape, MatVec / MatTransposeVec, and their reusable-output MatVecInto / MatTransposeVecInto counterparts — the matrix-free linear-operator interface, implemented both by concrete backend matrices and by LazyMatrix itself. Products return Result. MatVecScaledInto and MatTransposeVecScaledInto provide fused alpha/beta products for all supported backends and crate::WithIntercept.
• MatMat / MatTransposeMat and MatMatInto / MatTransposeMatInto batch multiple right-hand sides. MatrixRead borrows dense inputs, and MatrixOwned selects allocating output storage.
• Solver-facing vector algebra (DotProduct, L2Norm, ScaledAddAssign, and ScaleAssign).
• WeightedGramInto computes dense coefficient-space products into MatrixWrite storage. WeightedGramKernel lets backends apply normalization during accumulation to preserve small centered variations.
• WeightedColumnSumsInto and WeightedColumnSumsKernel compute stable weighted sums for intercept Gram cross terms.
• VectorOwned constructs backend-compatible coefficient scratch for lazy operator components.
• The five normalization-specific vector traits (ElemDivAssign, DotSlice, SubScalarAssign, SumEntries, ScaledSubSlice) — the elementwise primitives that fold the lazy normalization into a backend vector. They are phrased as a backend vector against a coefficient slice &[F], which is exactly the shape the centering/scaling math needs.
• ColumnStats — column statistics computed directly over a (possibly sparse) backend matrix, used by LazyMatrix::new.
• VectorView / VectorViewMut — storage-independent borrowed vector access, including strided backend views.
• RawColumn / RawColumns and LogicalColumn / Columns — the backend and normalized sides of storage-independent column access.
• SparseColumns — the stronger borrowed access capability for contiguous sparse columns.
• SparseRows — borrowed access to contiguous sparse rows.
• MaterializeDense — explicit normalized dense materialization from any supported input. MatrixOwned allocates the selected dense backend, and DenseNormalize permits infallible in-place dense normalization.
• SubmatrixKernel — rectangular products and materialization over raw storage, used by borrowed crate::Submatrix views.
• ReadBlock — experimental fallible rectangular reads into caller-owned storage, returning DenseBlock views without hiding I/O in borrowing.