What’s New in Revision 0.12.3

These are the changes in SpectroChemPy-0.12.3. See Release notes for a full changelog, including other versions of SpectroChemPy.

Bug Fixes

  • Public deprecation warnings no longer promise API removals in 0.13.0 for deprecated plotting aliases, the historical processing.alignement import path, deprecated plugin root aliases, concatenate(..., force_stack=True), or deprecated writer protocol= / description= kwargs. These paths remain available with warnings, and their eventual removal is now governed by the accepted SpectroChemPy deprecation lifecycle policy instead of an expired fixed-version target.

  • Analysis and fitting outputs now follow deterministic derived-metadata rules instead of inheriting path-dependent combinations of source identity and wrapper defaults. Single-source outputs preserve the captured scientific provenance from fit() time while generating canonical derived name/title/description/history text and clearing filename; direct NDDataset arguments remain authoritative for that call only, array-like inputs do not leak stale provenance, and input datasets are not mutated. Supervised PLSRegression.predict() now applies the accepted multi-source provenance order (Xtrain, Ytrain, then the prediction input), Optimize.result.residuals now follows the same canonical derived-output policy as fitted data, latent analysis factors now follow the accepted unitless/default-units policy, source-domain reconstructions and residuals preserve exact source units, supervised predictions preserve exact Ytrain units, and SVD diagnostic NDDataset outputs now use the common diagnostic metadata contract while raw U/s/VT factors remain unchanged.

  • Analysis outputs now align their public geometry support with the accepted family-specific rules: latent factors and profiles stay unmasked unless an exact restored full-row/full-column authority exists, source-domain reconstructions and fitted/residual outputs recover the public X coordset and mask only when their geometry matches exactly, PLSRegression predictions rebuild their observation axis from Xpredict and target axis from Ytrain without whole-source geometry copies, and diagnostic outputs remain generated and unmasked.

  • Sphinx Gallery example pages now use a consistent RST heading hierarchy. Top-level page titles are emitted at the correct level in the generated gallery, and nested section markers were normalized so the HTML navigation is structured consistently. This change is documentation-only and does not alter the scientific behavior of the examples.