colour-science / colour / 34475070726

97%

develop: 98%

Build Type

Pull #1419

github

Committed by MichaelMauderer

Commit Message

Separate cubic-spline and PCHIP interpolation into three clean layers.

Replace the tangled scipy-or-native conditionals with an explicit
architecture:
  - _ArrayInterpolator (+ _ArrayCubicSplineInterpolator, _ArrayPchipInterpolator):
    the native, autodiff-preserving numerics, standalone and scipy-free,
    exposing the scipy interp1d/PchipInterpolator x/y/__call__ contract.
  - scipy.interpolate.interp1d / PchipInterpolator: the NumPy reference, used
    directly.
  - _BackendAwareInterpolator: the unifying dispatcher; public
    CubicSplineInterpolator and PchipInterpolator subclass it and only supply
    the two implementation factories. It routes NumPy queries to scipy and
    everything else (incl. non-NumPy queries against a NumPy-built interpolator)
    to the native impl.

Strengthen test__call__scipy_parity to compare the native impl against an
independently constructed scipy interpolator, so it exercises the native path
on every backend including NumPy.

Pull Request Pull Request #1419: Array api improvements to preserve computaional graph

Coverage Stats

487 of 837 new or added lines in 42 files covered. (58.18%)

1 existing line in 1 file now uncovered.

49560 of 51123 relevant lines covered (96.94%)

0.97 hits per line