Module fallback 

Fallback implementations when SciRS2 is not available

These are used (via the #[cfg(not(feature = "scirs2"))] call sites in reconstruction/linear_inversion.rs and friends) only when the crate is built with --no-default-features (i.e. without scirs2-linalg). Every matrix operation below is a real, dimension-correct (if numerically naive) implementation rather than a fixed-size placeholder, so a no-scirs2 build computes with the actual input shape instead of silently substituting a hardcoded 2x2 result. The handful of decompositions that are genuinely impractical to hand-roll correctly (full eigenvalue decomposition, SVD) honestly return Err instead.

Modules

distributions

Distribution types and functions

graph

Graph analysis fallback functions

Structs

KSTestResult

Kolmogorov-Smirnov test result structure

OptimizeResult

Fallback optimization result

SWTestResult

Shapiro-Wilk test result structure

TTestResult

T-test result structure

Enums

Alternative

Alternative hypothesis type

Functions

cholesky

Real Cholesky decomposition (lower-triangular L such that L L^T = matrix). Works for any square NxN symmetric positive-definite matrix (not just 2x2); returns an honest error if the matrix is not positive-definite.

corrcoef

Real 2x2 Pearson correlation matrix for two equal-length samples x/y (a pairwise correlation matrix for two variables is always 2x2, so unlike inv/qr/etc. the size here was never the issue – the value was; this now computes the actual correlation instead of a fixed identity matrix).

det

Real matrix determinant via LU decomposition with partial pivoting. Works for any square NxN matrix (not just 2x2).

eig

Full eigenvalue decomposition is not implemented in this no-SciRS2 fallback (a numerically stable general eigensolver – e.g. the shifted-QR algorithm with deflation – is well beyond a “naive” reimplementation); honestly report the limitation rather than return a fixed, wrong-sized identity-like result.

inv

Real matrix inversion via Gauss-Jordan elimination with partial pivoting. Works for any square NxN matrix (not just 2x2); returns an honest error for non-square or singular/near-singular input.

ks_2samp

Fallback Kolmogorov-Smirnov 2-sample test

matrix_norm

Fallback matrix norm calculation: the Frobenius norm, computed from the actual matrix entries (not a fixed constant). Only the Frobenius norm is supported here – spectral/nuclear norms would require SVD, which this no-scirs2 fallback does not implement.

mean

Fallback statistical mean calculation

minimize

Fallback optimization function

pearsonr

Fallback Pearson correlation calculation

qr

Real QR decomposition via modified Gram-Schmidt orthogonalization. Works for any MxN matrix (not just 2x2).

shapiro_wilk

Fallback Shapiro-Wilk test

spearmanr

Fallback Spearman correlation calculation

std

Fallback standard deviation calculation

svd

Singular value decomposition is not implemented in this no-SciRS2 fallback (a numerically stable general SVD – e.g. Golub-Kahan bidiagonalization followed by an implicit-shift QR sweep – is well beyond a “naive” reimplementation); honestly report the limitation rather than return a fixed, wrong-sized identity-like result.

trace

Real matrix trace: sum of the diagonal entries (dimension-correct for any min(rows, cols)-length diagonal, not just 2x2).

ttest_1samp

Fallback t-test (one sample)

ttest_ind

Fallback t-test (independent samples)

var

Fallback variance calculation