Skip to contents

Contributing to sdvmodels

Source: .github/CONTRIBUTING.md

Thanks for helping. sdvmodels is a data package: it re-packages the fitted models that the sportsdataverse-py pipeline trains and publishes, so most changes are either a new bundle or a fix to how the bundle is packaged and documented.

Filing an issue

  • A model will not load, an object is missing, or install fails: use the Bug report template, with a minimal reprex and sessionInfo().
  • A model’s predictions look wrong: that is the training pipeline; open the issue in sportsdataverse-py.
  • A model or table sdvmodels should ship: use the Model request template.

Development setup

install.packages(c("devtools", "roxygen2", "testthat", "pkgdown", "arrow", "jsonlite", "usethis"))
devtools::load_all()
devtools::test()
devtools::document()
devtools::check()

Updating the models

gh release download cfb_model_artifacts --repo sportsdataverse/sportsdataverse-data --dir data-raw/artifacts --clobber
Rscript -e 'source("data-raw/MODELS.R")'
Rscript -e 'devtools::document(); devtools::test()'

Every cfb_*_model must stay the raw UBJ byte vector of the released file (readBin, never re-serialised by the local xgboost), so it is byte-identical to what sportsdataverse-py scores with. Parquet-derived tables must be materialised before saving; see CLAUDE.md for why.

The logo and the site

  • The hex logo, favicons and social card are drawn by data-raw/hex-logo/ (npm install && npm run build); never edit the PNGs by hand.
  • The pkgdown site deploys from main to the gh-pages branch through .github/workflows/pkgdown.yaml. Preview locally with pkgdown::build_site().

Commits and pull requests

  • Use Conventional Commits (data:, feat:, fix:, docs:, test:, chore:, ci:).
  • One logical change per commit, on a branch, through a pull request.
  • Never list AI tools as co-authors; the human author is the sole contributor of record.

Code of conduct

This project follows the Contributor Covenant. By participating you agree to abide by its terms.