Owner import path¶
jaxstro.numerics.linear_algebra
Purpose¶
Small dense vector, solve, covariance, correlation, and conditioning helpers.
Public records and callables¶
norm2, project_onto, condition_number, weighted_lstsq, qr_solve,
svd_solve, covariance_matrix, correlation_from_covariance,
correlation_matrix, is_positive_definite, add_diagonal_jitter, and
positive_definite_jitter.
Shape and dtype expectations¶
Vectors and matrices are floating arrays with explicit trailing vector or matrix axes. Solves require compatible row, column, weight, and right-hand-side dimensions.
JAX transforms and AD classification¶
Dense algebra composes with JIT and smooth-pathwise AD where rank and selected singular subspaces remain stable. Diagnostics are nonsmooth at rank changes and repeated singular values.
Failure behavior¶
Concrete invalid weights, covariance shapes, non-finite values, and negative
variances raise. Exact rank deficiency makes condition_number return positive
infinity, never NaN.
Contract and evidence links¶
See Linear algebra helpers and Validation.
Canonical import example¶
from jaxstro.numerics.linear_algebra import condition_number