Skip to content

Changelog

All notable changes to qlpca are recorded here. The project follows semantic versioning; while the version is 0.x the interface may still change between minor versions.

Version 0.1.0 was never published: it existed only as the version number of the thesis-replication state of the repository. 0.2.0 is the first release.

Unreleased

  • README, documentation site and CITATION.cff now give the Zenodo DOI (concept DOI 10.5281/zenodo.24260718; version 0.2.0: 10.5281/zenodo.24260719) and installation by pip install qlpca.

0.2.0 — 2026-11-15

First public release: the Python package and the replication of the PhD thesis (Lavado, 2012) in one repository.

Package

  • QLPCA estimator with two modes: a modernised default (docs/DESIGN.md, D1-D10) and legacy=True, which reproduces the 2012 MATLAB code and thesis Tables 4.3-4.8.
  • scikit-learn transformer (BaseEstimator, TransformerMixin): get_params / set_params, clone, pipelines, set_output, get_feature_names_out; tested with scikit-learn's estimator checks (exclusions in docs/PACKAGING.md).
  • Input validation with clear messages: NaN/inf, constant variables, fewer than two observations; SmallSampleWarning when n < 5 (degree + knots). Missing data are not supported.
  • pandas DataFrames accepted; column names kept in feature_names_in_ and used by the plots.
  • ExtrapolationWarning (exported in qlpca): transform and transform_variables warn when observations fall outside the training range, with their number and the variables (at most five named); the warning points at the caller's line; the values returned (linear extrapolation) are unchanged (docs/DESIGN.md, D11). README section "Use with scikit-learn": what works, the descriptive nature of the method, extrapolation and heavy tails.
  • Splines (qlpca.splines), diagnostics (qlpca.diagnostics), plots (qlpca.plots).
  • Data: the chapter 4 data set ships with the package (qlpca.datasets.load_st_gender, not covered by the code licence); make_cylinder simulates the cylinder problem.
  • New dependency: scikit-learn >= 1.6.

Replication

  • Notebooks 01-07 (07 is an introduction only); results compared with the thesis in the README.

Documentation

  • Documentation site (https://nlavadodatasci.github.io/quasi-linear-PCA/): MkDocs Material, API reference from the numpydoc docstrings (mkdocstrings), user guide, choice of knots (D12), thesis notebooks (mkdocs-jupyter, not re-executed), comparison with princals, Kelmarsh application, DESIGN.md and PACKAGING.md. Built with mkdocs build --strict in CI and deployed to GitHub Pages from main; extra docs in pyproject.toml.
  • Docstrings of all public functions in numpydoc format (no change of signatures or defaults).

Authors and metadata

  • Software authors: Nuno Lavado and Teresa Calapez (CITATION.cff, pyproject.toml, README "Authors"); .zenodo.json with the creators, licence, keywords and related identifiers for the Zenodo archive.

Infrastructure

  • Version read from the package metadata (single source in pyproject.toml).
  • CI: tests on Python 3.10-3.13, notebook execution, wheel install with the README quick start.
  • Publishing workflow to TestPyPI and PyPI through Trusted Publishing.