Repository navigation
Formalize documentation, CI and PyPI publishing (#48, #49, #50) - #51
Merged
Merged
Conversation
…(issue #48) - test.yml runs pytest on Python 3.10-3.12, Coveralls coverage and the PyPI version gate; publish-pypi.yml verifies the v* tag against __version__, builds sdist+wheel, runs twine check and uploads. - Add scripts/check_package_versions.py, [build-system] and a `test` extra. - Add CompStats/tests/test_packaging.py; update badge URLs and CLAUDE.md (new Releasing section); drop the stale nosetests.xml ignore line. Verification notes: `python -m build` is not installed in .venv (the issue assumed it was) and nothing may be installed, so the build was checked by calling setuptools.build_meta.build_sdist/build_wheel on a scratch copy: compstats-0.2.0.tar.gz and compstats-0.2.0-py3-none-any.whl built. check_package_versions.py printed "OK: ok" (pypi 0.1.15, local 0.2.0). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…sue #49) - Rewrite docs/source/conf.py (furo, no sys.path, intersphinx for python, numpy, pandas, scipy, sklearn, statsmodels, seaborn, matplotlib); add docs/requirements.txt and utils_api.rst; drop .readthedocs.yaml. - Restructure index.rst (introduction, statistical background, architecture, invariants, development, citing with both papers); update metrics_api.rst and performance_api.rst. - Fix all 258 strict-build warnings by editing docstring text only: parameter names and .measure factories become inline literals (including the metrics_docs template), unqualified/abbreviated type names are qualified, malformed reST in performance.py is repaired; improve the weak docstrings listed in the issue. No executable code changed (verified by comparing the ASTs with docstrings stripped; only the metrics_docs template string differs). - Add the `docs` job (strict build) to test.yml and CompStats/tests/test_docs.py; update CLAUDE.md. `python -m sphinx -n -W --keep-going -b html docs/source docs/build/html` ends with "build succeeded." and no warning. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… to GitHub Pages (issue #50) - Quarto: new navbar, home page as index.qmd, new Install, Methodology and Citing pages; the Colab install cells are shown but not executed (#| eval: false); sklearn_users is seeded and ends with the comparison against the best system; quarto/_freeze refreshed so CI renders without Python. - README.rst rewritten (short, formal, no images); pyproject URLs point at the GitHub Pages site; legacy docs/*.ipynb removed. - website.yml replaces publish.yml: renders the site from the freeze, builds the Sphinx reference into /dev/ and deploys with actions/deploy-pages from main. - CLAUDE.md gets a Documentation section; docs/source/index.rst documents the user-site preview; test_docs.py gains the freeze, navbar, README, workflow and project-URL tests. Verified: the executing render, the render with QUARTO_PYTHON=/nonexistent (no kernel started, freeze unchanged), the strict Sphinx build into quarto/_site/dev, docutils on README.rst, and the full pytest suite. Closes #45 Closes #46 Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
#48 #49 #50) Corrects a review finding found after the fact; these edits are not part of the original issues and must not be reverted as redundant. - Remove the "developed by INGEOTEC" attribution from the home page and the INGEOTEC mention from the site footer (authors named instead). - Replace "validated on eight competitions" with the article's wording (designed to be universally applicable, illustrated with eight cases). - State why classical tests do not apply (Dietterich 1998, Demsar 2006) and the purpose of the methodology on the home page and in the README. - Methodology page: hypotheses and partial-overlap reading, 0.05 decision rule and rationale of the paired bootstrap (Koehn 2004), reasons for multiple comparison correction, competitiveness restructured into three aspects with PPI separated as potential for improvement; add three references. - docs/source/index.rst: same competitiveness correction. - test_packaging.py no longer contains the legacy strings it checks for; add test_site_follows_the_article. Verified: strict Sphinx build, README with docutils, render from the freeze without Python (freeze unchanged), full suite 100 passed (run with MPLBACKEND=Agg, as CI does; the macOS GUI backend blocks in plt.show()). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…-2: #50) Corrects a review finding found after the fact: citations were plain text and the reader had no link to the full reference. Not part of the original issue; do not revert as redundant. - Add quarto/references.bib and `bibliography: references.bib`; citations on the home and methodology pages are now `[@key]` links to complete entries (authors, title, venue, volume, pages, DOI or URL), and the hand-written reference list is replaced by the generated one. - Add volume and pages of the 2024 article to the BibTeX in citing.qmd and README.rst. - test_docs.py: methodology test checks citation through the bibliography; new test_site_bibliography. Verified: render from the freeze without Python (freeze unchanged), citation targets present, README valid, full suite 101 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…e (review fix-3: #50) Corrects a review finding found after the fact: the arXiv link of the 2024 article was typed in the home page prose instead of being part of its reference. Not part of the original issue; do not revert as redundant. - Add quarto/compstats.csl (author-date, prints the BibTeX `note` field, which Quarto's default Chicago style does not) and `csl: compstats.csl`. - references.bib: `note` with the preprint URL for Nava-Munoz2024; double braces around the Efron1994 publisher. - index.qmd: drop the hand-typed preprint link. - test_site_bibliography checks the style, the note and the absence of the link in the prose. Verified: render from the freeze without Python (freeze unchanged), preprint link inside the reference entry on both pages, full suite 101 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Corrects a review finding found after the fact: every code cell of the three notebook pages was collapsed behind a "Code" link. Not part of the original issue; do not revert as redundant. - Remove `code-fold: true` from quarto/_quarto.yml (Quarto's default shows the code unfolded). - Add test_site_code_is_visible. Verified: render from the freeze without Python (freeze unchanged), no code-fold in the three notebook pages and the code still present, full suite 102 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…fix-5: #50) Corrects a review finding found after the fact; these edits are not part of the original issue and must not be reverted as redundant. - Colab install cell of the three notebooks: `try: import CompStats / except ImportError:` guard that installs with `uv pip install -q --system`, falling back to pip; still `#| eval: false`. - sklearn_users.ipynb: explain `statistic` and `se` of the first output, explain the figure of "Rest of the classifiers", state that the package is not limited to classification and add a Regression section (diabetes, mean_absolute_error, plot, Holm-adjusted p-values). - quarto/_freeze regenerated; test_docs.py and CLAUDE.md follow the new install cell; new test_sklearn_users_covers_regression. Verified: executing render, then render without Python with the freeze unchanged, inserted sentences checked against the regenerated outputs, full suite 103 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…w fix-6: #50) Corrects a review finding found after the fact: the prose joined clauses with semicolons, colons and dashes far more often than the published article. Not part of the original issue; do not revert as redundant. - index.qmd, install.qmd, methodology.qmd, the markdown cells of the three notebooks and the prose of README.rst: one idea per sentence, sentence-initial connectors or subordination instead of semicolons, colons only before enumerations, no dashes as punctuation. Meaning, numbers, code, formulas, headings and citations are unchanged. - Notebooks edited only in markdown cells through nbformat (code cells verified identical); quarto/_freeze regenerated. - test_site_prose_style enforces the punctuation limits; CLAUDE.md gets the rule. The test helper now skips Jupyter's .ipynb_checkpoints directories. Verified: executing render, then render without Python with the freeze unchanged, outputs quoted by the text unchanged, full suite 104 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This was referenced Nov 20, 2026
mgraffg
added a commit
that referenced
this pull request
Nov 20, 2026
* Crear la plantilla del sitio en Quarto (estructura tipo LegalIA) y la página de Introducción Migra quarto/ de un dashboard de una sola página a un proyecto Quarto de tipo website: _quarto.yml con navbar/footer/tema propio, index.ipynb como libreta de inicio con la Introducción basada en el artículo (arXiv:2403.04693) y las tarjetas de instalación, y pages/ como destino para las libretas que agregarán #38, #39 y #40. Closes #43 Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Agregar identificadores únicos a las celdas de markdown y actualizar la información del lenguaje en el notebook de Quarto * Crear la página 'scikit-learn Users' (libreta) con la API actual (closes #38) Migra la sección "scikit-learn Users" del borrador anterior (quarto/CompStats.qmd) a quarto/pages/sklearn_users.ipynb, siguiendo el formato de libreta/website adoptado en #43. Los ejemplos ya usaban la API actual (func/BiB vía CompStats.metrics.f1_score); el flujo se verificó ejecutando con `quarto render`. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Crear la página 'Competition Organizers' (libreta) usando el dataset de INGEOTEC/Delitos (closes #39) Sólo las etiquetas del corpus son redistribuibles (no el texto de los tuits, por los términos de servicio de Twitter/X), así que la libreta descarga las etiquetas de la partición de prueba y simula cuatro "submissions" con distinto nivel de acierto para ilustrar el flujo de un organizador: construir el leaderboard con f1_score, revisar los intervalos de confianza, y calcular difference()/p-values respecto al mejor. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Expandir la libreta 'Competition Organizers' a cinco competidores y agregar corrección por comparaciones múltiples Reemplaza el baseline de clase mayoritaria por un quinto equipo simulado, de modo que Team Alpha y Team Beta comparten el mismo nivel de habilidad (pero con predicciones distintas, no idénticas) y quedan estadísticamente empatados en primer lugar, mientras que Team Gamma, Team Delta y Team Epsilon son competitivos pero con rendimiento menor. Se fija la semilla del remuestreo bootstrap (np.random.seed) para que el resultado sea reproducible: Team Gamma queda significativamente por debajo del mejor con el p-value crudo, Holm y FDR, pero empatado bajo la corrección de Bonferroni, lo que ilustra cómo cambian tanto los p-values como las gráficas (Perf.plot/Difference.plot) según el método de corrección usado. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Crear la página 'Multiple Testing' (libreta) con ejemplo de tres métricas de rendimiento Muestra un Perf multi-medida (precision, recall, F1) sobre tres sistemas sintéticos con distinto trade-off precision/recall, donde cada uno gana en una métrica distinta, y usa Difference.p_value(correction=...) para corregir por comparaciones múltiples entre medidas (closes #40) Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Reestructurar docs/source/index.rst como documentación de desarrollo La página principal de Sphinx deja de ser el tutorial completo (Quickstart Guide con la API funcional antigua de performance()) y pasa a incluir una introducción breve, una nota que remite al sitio de Quarto para el tutorial de uso, y la arquitectura interna (antes en architecture.rst, ahora fusionada en index.rst) como documentación orientada a quien contribuye al código. También se agrega el docstring faltante en StatisticSamples.__getitem__. Closes #44 Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Enlazar el sitio principal (Quarto) con la documentación de la API en ReadTheDocs Se agrega un enlace "API Docs" en el navbar de quarto/_quarto.yml hacia https://compstats.readthedocs.io/en/latest/. El enlace de vuelta desde Sphinx hacia https://ingeotec.github.io/CompStats/ ya existía en docs/source/index.rst (agregado al cerrar #44). Closes #42 Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Agregar enlaces 'Open in Colab' a las libretas del sitio (closes #41) Cada libreta bajo quarto/ (index.ipynb y pages/*.ipynb) recibe un badge que enlaza al notebook crudo en la rama main para abrirlo directamente en Colab. Las tres libretas con código (sklearn_users, competition_organizers, multiple_testing) también reciben una celda `pip install -q CompStats` al inicio, ya que Colab no trae el paquete preinstalado; el resto de sus dependencias (seaborn, statsmodels, pandas) llegan de forma transitiva por ser dependencias declaradas de CompStats en pyproject.toml. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Configure issue-pilot for this repository (#47) * Formalize documentation, CI and PyPI publishing (#48, #49, #50) (#51) * Replace legacy CI with pytest and a tag-driven PyPI publish workflow (issue #48) - test.yml runs pytest on Python 3.10-3.12, Coveralls coverage and the PyPI version gate; publish-pypi.yml verifies the v* tag against __version__, builds sdist+wheel, runs twine check and uploads. - Add scripts/check_package_versions.py, [build-system] and a `test` extra. - Add CompStats/tests/test_packaging.py; update badge URLs and CLAUDE.md (new Releasing section); drop the stale nosetests.xml ignore line. Verification notes: `python -m build` is not installed in .venv (the issue assumed it was) and nothing may be installed, so the build was checked by calling setuptools.build_meta.build_sdist/build_wheel on a scratch copy: compstats-0.2.0.tar.gz and compstats-0.2.0-py3-none-any.whl built. check_package_versions.py printed "OK: ok" (pypi 0.1.15, local 0.2.0). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Make the Sphinx developer reference formal and strict-build clean (issue #49) - Rewrite docs/source/conf.py (furo, no sys.path, intersphinx for python, numpy, pandas, scipy, sklearn, statsmodels, seaborn, matplotlib); add docs/requirements.txt and utils_api.rst; drop .readthedocs.yaml. - Restructure index.rst (introduction, statistical background, architecture, invariants, development, citing with both papers); update metrics_api.rst and performance_api.rst. - Fix all 258 strict-build warnings by editing docstring text only: parameter names and .measure factories become inline literals (including the metrics_docs template), unqualified/abbreviated type names are qualified, malformed reST in performance.py is repaired; improve the weak docstrings listed in the issue. No executable code changed (verified by comparing the ASTs with docstrings stripped; only the metrics_docs template string differs). - Add the `docs` job (strict build) to test.yml and CompStats/tests/test_docs.py; update CLAUDE.md. `python -m sphinx -n -W --keep-going -b html docs/source docs/build/html` ends with "build succeeded." and no warning. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Formalize the Quarto user site and README, and publish both doc sites to GitHub Pages (issue #50) - Quarto: new navbar, home page as index.qmd, new Install, Methodology and Citing pages; the Colab install cells are shown but not executed (#| eval: false); sklearn_users is seeded and ends with the comparison against the best system; quarto/_freeze refreshed so CI renders without Python. - README.rst rewritten (short, formal, no images); pyproject URLs point at the GitHub Pages site; legacy docs/*.ipynb removed. - website.yml replaces publish.yml: renders the site from the freeze, builds the Sphinx reference into /dev/ and deploys with actions/deploy-pages from main. - CLAUDE.md gets a Documentation section; docs/source/index.rst documents the user-site preview; test_docs.py gains the freeze, navbar, README, workflow and project-URL tests. Verified: the executing render, the render with QUARTO_PYTHON=/nonexistent (no kernel started, freeze unchanged), the strict Sphinx build into quarto/_site/dev, docutils on README.rst, and the full pytest suite. Closes #45 Closes #46 Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Align the documentation prose with the published article (review fix-1: #48 #49 #50) Corrects a review finding found after the fact; these edits are not part of the original issues and must not be reverted as redundant. - Remove the "developed by INGEOTEC" attribution from the home page and the INGEOTEC mention from the site footer (authors named instead). - Replace "validated on eight competitions" with the article's wording (designed to be universally applicable, illustrated with eight cases). - State why classical tests do not apply (Dietterich 1998, Demsar 2006) and the purpose of the methodology on the home page and in the README. - Methodology page: hypotheses and partial-overlap reading, 0.05 decision rule and rationale of the paired bootstrap (Koehn 2004), reasons for multiple comparison correction, competitiveness restructured into three aspects with PPI separated as potential for improvement; add three references. - docs/source/index.rst: same competitiveness correction. - test_packaging.py no longer contains the legacy strings it checks for; add test_site_follows_the_article. Verified: strict Sphinx build, README with docutils, render from the freeze without Python (freeze unchanged), full suite 100 passed (run with MPLBACKEND=Agg, as CI does; the macOS GUI backend blocks in plt.show()). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Cite works on the user site through Quarto's bibliography (review fix-2: #50) Corrects a review finding found after the fact: citations were plain text and the reader had no link to the full reference. Not part of the original issue; do not revert as redundant. - Add quarto/references.bib and `bibliography: references.bib`; citations on the home and methodology pages are now `[@key]` links to complete entries (authors, title, venue, volume, pages, DOI or URL), and the hand-written reference list is replaced by the generated one. - Add volume and pages of the 2024 article to the BibTeX in citing.qmd and README.rst. - test_docs.py: methodology test checks citation through the bibliography; new test_site_bibliography. Verified: render from the freeze without Python (freeze unchanged), citation targets present, README valid, full suite 101 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Carry the preprint link in the bibliography with a site citation style (review fix-3: #50) Corrects a review finding found after the fact: the arXiv link of the 2024 article was typed in the home page prose instead of being part of its reference. Not part of the original issue; do not revert as redundant. - Add quarto/compstats.csl (author-date, prints the BibTeX `note` field, which Quarto's default Chicago style does not) and `csl: compstats.csl`. - references.bib: `note` with the preprint URL for Nava-Munoz2024; double braces around the Efron1994 publisher. - index.qmd: drop the hand-typed preprint link. - test_site_bibliography checks the style, the note and the absence of the link in the prose. Verified: render from the freeze without Python (freeze unchanged), preprint link inside the reference entry on both pages, full suite 101 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Show all code on the user site without a fold link (review fix-4: #50) Corrects a review finding found after the fact: every code cell of the three notebook pages was collapsed behind a "Code" link. Not part of the original issue; do not revert as redundant. - Remove `code-fold: true` from quarto/_quarto.yml (Quarto's default shows the code unfolded). - Add test_site_code_is_visible. Verified: render from the freeze without Python (freeze unchanged), no code-fold in the three notebook pages and the code still present, full suite 102 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Improve the scikit-learn tutorial and the Colab install cell (review fix-5: #50) Corrects a review finding found after the fact; these edits are not part of the original issue and must not be reverted as redundant. - Colab install cell of the three notebooks: `try: import CompStats / except ImportError:` guard that installs with `uv pip install -q --system`, falling back to pip; still `#| eval: false`. - sklearn_users.ipynb: explain `statistic` and `se` of the first output, explain the figure of "Rest of the classifiers", state that the package is not limited to classification and add a Regression section (diabetes, mean_absolute_error, plot, Holm-adjusted p-values). - quarto/_freeze regenerated; test_docs.py and CLAUDE.md follow the new install cell; new test_sklearn_users_covers_regression. Verified: executing render, then render without Python with the freeze unchanged, inserted sentences checked against the regenerated outputs, full suite 103 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Actualizar metadatos en el cuaderno de scikit-learn para evitar colapsar las celdas * Rewrite the site prose in the punctuation style of the article (review fix-6: #50) Corrects a review finding found after the fact: the prose joined clauses with semicolons, colons and dashes far more often than the published article. Not part of the original issue; do not revert as redundant. - index.qmd, install.qmd, methodology.qmd, the markdown cells of the three notebooks and the prose of README.rst: one idea per sentence, sentence-initial connectors or subordination instead of semicolons, colons only before enumerations, no dashes as punctuation. Meaning, numbers, code, formulas, headings and citations are unchanged. - Notebooks edited only in markdown cells through nbformat (code cells verified identical); quarto/_freeze regenerated. - test_site_prose_style enforces the punctuation limits; CLAUDE.md gets the rule. The test helper now skips Jupyter's .ipynb_checkpoints directories. Verified: executing render, then render without Python with the freeze unchanged, outputs quoted by the text unchanged, full suite 104 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
mgraffg
added a commit
that referenced
this pull request
Nov 21, 2026
* Crear la plantilla del sitio en Quarto (estructura tipo LegalIA) y la página de Introducción Migra quarto/ de un dashboard de una sola página a un proyecto Quarto de tipo website: _quarto.yml con navbar/footer/tema propio, index.ipynb como libreta de inicio con la Introducción basada en el artículo (arXiv:2403.04693) y las tarjetas de instalación, y pages/ como destino para las libretas que agregarán #38, #39 y #40. Closes #43 Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Agregar identificadores únicos a las celdas de markdown y actualizar la información del lenguaje en el notebook de Quarto * Crear la página 'scikit-learn Users' (libreta) con la API actual (closes #38) Migra la sección "scikit-learn Users" del borrador anterior (quarto/CompStats.qmd) a quarto/pages/sklearn_users.ipynb, siguiendo el formato de libreta/website adoptado en #43. Los ejemplos ya usaban la API actual (func/BiB vía CompStats.metrics.f1_score); el flujo se verificó ejecutando con `quarto render`. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Crear la página 'Competition Organizers' (libreta) usando el dataset de INGEOTEC/Delitos (closes #39) Sólo las etiquetas del corpus son redistribuibles (no el texto de los tuits, por los términos de servicio de Twitter/X), así que la libreta descarga las etiquetas de la partición de prueba y simula cuatro "submissions" con distinto nivel de acierto para ilustrar el flujo de un organizador: construir el leaderboard con f1_score, revisar los intervalos de confianza, y calcular difference()/p-values respecto al mejor. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Expandir la libreta 'Competition Organizers' a cinco competidores y agregar corrección por comparaciones múltiples Reemplaza el baseline de clase mayoritaria por un quinto equipo simulado, de modo que Team Alpha y Team Beta comparten el mismo nivel de habilidad (pero con predicciones distintas, no idénticas) y quedan estadísticamente empatados en primer lugar, mientras que Team Gamma, Team Delta y Team Epsilon son competitivos pero con rendimiento menor. Se fija la semilla del remuestreo bootstrap (np.random.seed) para que el resultado sea reproducible: Team Gamma queda significativamente por debajo del mejor con el p-value crudo, Holm y FDR, pero empatado bajo la corrección de Bonferroni, lo que ilustra cómo cambian tanto los p-values como las gráficas (Perf.plot/Difference.plot) según el método de corrección usado. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Crear la página 'Multiple Testing' (libreta) con ejemplo de tres métricas de rendimiento Muestra un Perf multi-medida (precision, recall, F1) sobre tres sistemas sintéticos con distinto trade-off precision/recall, donde cada uno gana en una métrica distinta, y usa Difference.p_value(correction=...) para corregir por comparaciones múltiples entre medidas (closes #40) Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Reestructurar docs/source/index.rst como documentación de desarrollo La página principal de Sphinx deja de ser el tutorial completo (Quickstart Guide con la API funcional antigua de performance()) y pasa a incluir una introducción breve, una nota que remite al sitio de Quarto para el tutorial de uso, y la arquitectura interna (antes en architecture.rst, ahora fusionada en index.rst) como documentación orientada a quien contribuye al código. También se agrega el docstring faltante en StatisticSamples.__getitem__. Closes #44 Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Enlazar el sitio principal (Quarto) con la documentación de la API en ReadTheDocs Se agrega un enlace "API Docs" en el navbar de quarto/_quarto.yml hacia https://compstats.readthedocs.io/en/latest/. El enlace de vuelta desde Sphinx hacia https://ingeotec.github.io/CompStats/ ya existía en docs/source/index.rst (agregado al cerrar #44). Closes #42 Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Agregar enlaces 'Open in Colab' a las libretas del sitio (closes #41) Cada libreta bajo quarto/ (index.ipynb y pages/*.ipynb) recibe un badge que enlaza al notebook crudo en la rama main para abrirlo directamente en Colab. Las tres libretas con código (sklearn_users, competition_organizers, multiple_testing) también reciben una celda `pip install -q CompStats` al inicio, ya que Colab no trae el paquete preinstalado; el resto de sus dependencias (seaborn, statsmodels, pandas) llegan de forma transitiva por ser dependencias declaradas de CompStats en pyproject.toml. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Configure issue-pilot for this repository (#47) * Formalize documentation, CI and PyPI publishing (#48, #49, #50) (#51) * Replace legacy CI with pytest and a tag-driven PyPI publish workflow (issue #48) - test.yml runs pytest on Python 3.10-3.12, Coveralls coverage and the PyPI version gate; publish-pypi.yml verifies the v* tag against __version__, builds sdist+wheel, runs twine check and uploads. - Add scripts/check_package_versions.py, [build-system] and a `test` extra. - Add CompStats/tests/test_packaging.py; update badge URLs and CLAUDE.md (new Releasing section); drop the stale nosetests.xml ignore line. Verification notes: `python -m build` is not installed in .venv (the issue assumed it was) and nothing may be installed, so the build was checked by calling setuptools.build_meta.build_sdist/build_wheel on a scratch copy: compstats-0.2.0.tar.gz and compstats-0.2.0-py3-none-any.whl built. check_package_versions.py printed "OK: ok" (pypi 0.1.15, local 0.2.0). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Make the Sphinx developer reference formal and strict-build clean (issue #49) - Rewrite docs/source/conf.py (furo, no sys.path, intersphinx for python, numpy, pandas, scipy, sklearn, statsmodels, seaborn, matplotlib); add docs/requirements.txt and utils_api.rst; drop .readthedocs.yaml. - Restructure index.rst (introduction, statistical background, architecture, invariants, development, citing with both papers); update metrics_api.rst and performance_api.rst. - Fix all 258 strict-build warnings by editing docstring text only: parameter names and .measure factories become inline literals (including the metrics_docs template), unqualified/abbreviated type names are qualified, malformed reST in performance.py is repaired; improve the weak docstrings listed in the issue. No executable code changed (verified by comparing the ASTs with docstrings stripped; only the metrics_docs template string differs). - Add the `docs` job (strict build) to test.yml and CompStats/tests/test_docs.py; update CLAUDE.md. `python -m sphinx -n -W --keep-going -b html docs/source docs/build/html` ends with "build succeeded." and no warning. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Formalize the Quarto user site and README, and publish both doc sites to GitHub Pages (issue #50) - Quarto: new navbar, home page as index.qmd, new Install, Methodology and Citing pages; the Colab install cells are shown but not executed (#| eval: false); sklearn_users is seeded and ends with the comparison against the best system; quarto/_freeze refreshed so CI renders without Python. - README.rst rewritten (short, formal, no images); pyproject URLs point at the GitHub Pages site; legacy docs/*.ipynb removed. - website.yml replaces publish.yml: renders the site from the freeze, builds the Sphinx reference into /dev/ and deploys with actions/deploy-pages from main. - CLAUDE.md gets a Documentation section; docs/source/index.rst documents the user-site preview; test_docs.py gains the freeze, navbar, README, workflow and project-URL tests. Verified: the executing render, the render with QUARTO_PYTHON=/nonexistent (no kernel started, freeze unchanged), the strict Sphinx build into quarto/_site/dev, docutils on README.rst, and the full pytest suite. Closes #45 Closes #46 Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Align the documentation prose with the published article (review fix-1: #48 #49 #50) Corrects a review finding found after the fact; these edits are not part of the original issues and must not be reverted as redundant. - Remove the "developed by INGEOTEC" attribution from the home page and the INGEOTEC mention from the site footer (authors named instead). - Replace "validated on eight competitions" with the article's wording (designed to be universally applicable, illustrated with eight cases). - State why classical tests do not apply (Dietterich 1998, Demsar 2006) and the purpose of the methodology on the home page and in the README. - Methodology page: hypotheses and partial-overlap reading, 0.05 decision rule and rationale of the paired bootstrap (Koehn 2004), reasons for multiple comparison correction, competitiveness restructured into three aspects with PPI separated as potential for improvement; add three references. - docs/source/index.rst: same competitiveness correction. - test_packaging.py no longer contains the legacy strings it checks for; add test_site_follows_the_article. Verified: strict Sphinx build, README with docutils, render from the freeze without Python (freeze unchanged), full suite 100 passed (run with MPLBACKEND=Agg, as CI does; the macOS GUI backend blocks in plt.show()). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Cite works on the user site through Quarto's bibliography (review fix-2: #50) Corrects a review finding found after the fact: citations were plain text and the reader had no link to the full reference. Not part of the original issue; do not revert as redundant. - Add quarto/references.bib and `bibliography: references.bib`; citations on the home and methodology pages are now `[@key]` links to complete entries (authors, title, venue, volume, pages, DOI or URL), and the hand-written reference list is replaced by the generated one. - Add volume and pages of the 2024 article to the BibTeX in citing.qmd and README.rst. - test_docs.py: methodology test checks citation through the bibliography; new test_site_bibliography. Verified: render from the freeze without Python (freeze unchanged), citation targets present, README valid, full suite 101 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Carry the preprint link in the bibliography with a site citation style (review fix-3: #50) Corrects a review finding found after the fact: the arXiv link of the 2024 article was typed in the home page prose instead of being part of its reference. Not part of the original issue; do not revert as redundant. - Add quarto/compstats.csl (author-date, prints the BibTeX `note` field, which Quarto's default Chicago style does not) and `csl: compstats.csl`. - references.bib: `note` with the preprint URL for Nava-Munoz2024; double braces around the Efron1994 publisher. - index.qmd: drop the hand-typed preprint link. - test_site_bibliography checks the style, the note and the absence of the link in the prose. Verified: render from the freeze without Python (freeze unchanged), preprint link inside the reference entry on both pages, full suite 101 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Show all code on the user site without a fold link (review fix-4: #50) Corrects a review finding found after the fact: every code cell of the three notebook pages was collapsed behind a "Code" link. Not part of the original issue; do not revert as redundant. - Remove `code-fold: true` from quarto/_quarto.yml (Quarto's default shows the code unfolded). - Add test_site_code_is_visible. Verified: render from the freeze without Python (freeze unchanged), no code-fold in the three notebook pages and the code still present, full suite 102 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Improve the scikit-learn tutorial and the Colab install cell (review fix-5: #50) Corrects a review finding found after the fact; these edits are not part of the original issue and must not be reverted as redundant. - Colab install cell of the three notebooks: `try: import CompStats / except ImportError:` guard that installs with `uv pip install -q --system`, falling back to pip; still `#| eval: false`. - sklearn_users.ipynb: explain `statistic` and `se` of the first output, explain the figure of "Rest of the classifiers", state that the package is not limited to classification and add a Regression section (diabetes, mean_absolute_error, plot, Holm-adjusted p-values). - quarto/_freeze regenerated; test_docs.py and CLAUDE.md follow the new install cell; new test_sklearn_users_covers_regression. Verified: executing render, then render without Python with the freeze unchanged, inserted sentences checked against the regenerated outputs, full suite 103 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Actualizar metadatos en el cuaderno de scikit-learn para evitar colapsar las celdas * Rewrite the site prose in the punctuation style of the article (review fix-6: #50) Corrects a review finding found after the fact: the prose joined clauses with semicolons, colons and dashes far more often than the published article. Not part of the original issue; do not revert as redundant. - index.qmd, install.qmd, methodology.qmd, the markdown cells of the three notebooks and the prose of README.rst: one idea per sentence, sentence-initial connectors or subordination instead of semicolons, colons only before enumerations, no dashes as punctuation. Meaning, numbers, code, formulas, headings and citations are unchanged. - Notebooks edited only in markdown cells through nbformat (code cells verified identical); quarto/_freeze regenerated. - test_site_prose_style enforces the punctuation limits; CLAUDE.md gets the rule. The test helper now skips Jupyter's .ipynb_checkpoints directories. Verified: executing render, then render without Python with the freeze unchanged, outputs quoted by the text unchanged, full suite 104 passed (MPLBACKEND=Agg). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com> * Publish to PyPI only on manual dispatch Remove the tag trigger of publish-pypi.yml, and the tag/version check that depended on it, so that creating a tag or a GitHub release no longer tries to upload to PyPI. Update the release instructions and the workflow test accordingly. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Formalizes the project's documentation and release pipeline in three steps, followed by six review fixes on the same branch.
#48 — Replace legacy CI with pytest and a tag-driven PyPI publish workflow
.github/workflows/test.ymlrunspytest CompStatson Python 3.10–3.12, collects coverage and runs the version gatescripts/check_package_versions.py(the local__version__must be exactly one patch or one minor release ahead of PyPI)..github/workflows/publish-pypi.ymlpublishes to PyPI when av<__version__>tag is pushed, after checking that the tag equals__version__.test.yaml,pip.yaml) are removed;CompStats/tests/test_packaging.pycovers the packaging metadata.#49 — Make the Sphinx developer reference formal and strict-build clean
docs/source/builds withsphinx -n -W --keep-goingwithout warnings; CI runs that same command.CompStats/(no executable code); the shared docstring template inmetrics_docswas fixed first..readthedocs.yamlremoved); the developer reference is published on GitHub Pages under/dev/.#50 — Formalize the Quarto user site and README, and publish both doc sites to GitHub Pages
quarto/) gets a home page, Install, Methodology and Citing pages written from the published article, and the three tutorial notebooks use thePerf-based API only..github/workflows/website.ymlrenders the site from the committedquarto/_freeze(CI never executes notebooks), builds the Sphinx reference into/dev/and deploys both withactions/deploy-pageson pushes tomain.README.rstis rewritten to match;CompStats/tests/test_docs.pychecks the site, the freeze and the README.Review fixes
f4b315a5c4f897d7b6a3e95fe5b3statistic/seand of the figure, regression example) and the Colab install cell (uv, installed only when missing) (issue named: Formalize the Quarto user site and README, and publish both doc sites to GitHub Pages #50) —0763e705abc121Commits
7830a8aReplace legacy CI with pytest and a tag-driven PyPI publish workflow (issue Replace legacy CI with pytest and a tag-driven PyPI publish workflow #48)7d0dfd1Make the Sphinx developer reference formal and strict-build clean (issue Make the Sphinx developer reference formal and strict-build clean #49)8608cbbFormalize the Quarto user site and README, and publish both doc sites to GitHub Pages (issue Formalize the Quarto user site and README, and publish both doc sites to GitHub Pages #50)f4b315aAlign the documentation prose with the published article (review fix-1)5c4f897Cite works on the user site through Quarto's bibliography (review fix-2)d7b6a3eCarry the preprint link in the bibliography with a site citation style (review fix-3)95fe5b3Show all code on the user site without a fold link (review fix-4)0763e70Improve the scikit-learn tutorial and the Colab install cell (review fix-5)a57a565Actualizar metadatos en el cuaderno de scikit-learn para evitar colapsar las celdas (maintainer's own commit)5abc121Rewrite the site prose in the punctuation style of the article (review fix-6)Verification
MPLBACKEND=Agg .venv/bin/python -m pytest CompStats: 104 passed (run on the branch tip before opening this pull request). On macOS the suite needsMPLBACKEND=Agg; without itplt.show()blocks.#| eval: false), and the GitHub Pages deployment only runs onmain.After merging
develop, which is not the default branch, so GitHub will not close the issues on merge.main.v0.2.0after the merge tomain.Closes #48
Closes #49
Closes #50
Closes #45
Closes #46
🤖 Generated with Claude Code