Skip to content

AGENTS.md

Latest commit

聽

History

History
177 lines (134 loc) 路 6.88 KB

AGENTS.md

File metadata and controls

177 lines (134 loc) 路 6.88 KB
Raw

AGENTS.md

Project-level instructions for AI coding agents working on this repository. GitHub Copilot Code Review, Copilot coding agent, Codex, and other AGENTS.md-aware tools read this file directly.

Package overview

statsExpressions is an R package that creates tidy data frames and plotmath expressions with details from statistical tests. It serves as the statistical backend for ggstatsplot.

Architecture

Main functions (R/)

  • Statistical tests: oneway_anova(), two_sample_test(), one_sample_test(), corr_test(), contingency_table(), meta_analysis(), pairwise_comparisons(), and pairwise_contingency_table().
  • Supporting public helpers include centrality_description(), add_expression_col(), tidy_model_expressions(), tidy_model_parameters(), extract_stats_type(), stats_type_switch(), and long_to_wide_converter().
  • The package standardizes parametric, nonparametric, robust, and Bayesian type labels, but supported choices vary by function. Do not assume every exported function has this interface; for example, pairwise_contingency_table() does not accept type.
  • Statistical functions return tibbles with a special expression column for plotmath output where applicable.

Key internal helpers

  • tidy_model_effectsize(): Convert effect-size output to tidy conventions.
  • extract_estimate_type() and extract_statistic_text(): Choose plotmath labels for estimates and test statistics.
  • prior_switch(): Choose prior labels for Bayesian output.

Dependencies

Core dependencies include the tidyverse stack (dplyr, purrr, tidyr, and rlang) and the easystats ecosystem (insight, parameters, performance, effectsize, bayestestR, datawizard, and correlation). Treat DESCRIPTION as the source of truth for dependency constraints.

The minimum supported R version is 4.5. CI covers R-devel, the current R release, and the previous R release; keep README support wording independent of specific version numbers.

Developer workflow

Use the repository Makefile for routine package tasks:

make install_deps # Install dependencies declared in DESCRIPTION
make build        # Build the package tarball
make check        # Build and run R CMD check --no-manual
make install      # Build and install the package locally
make document     # Regenerate roxygen docs and render README.Rmd
make lint         # Run lintr::lint_package()
make format       # Run styler::style_pkg()
make hooks        # Run all prek hooks
make clean        # Remove package build and check artifacts
make update_deps  # Refresh dependency constraints, docs, and codemeta

make update_deps is a maintenance operation that can rewrite dependency constraints and generated metadata. Do not use it merely to install the current dependency set.

Versioning and changelog

  • Development versions use a fourth-component .9000 suffix.
  • Keep the version in DESCRIPTION, codemeta.json, and the first NEWS.md heading synchronized.
  • Record user-facing compatibility changes in NEWS.md; omit routine dependency updates and internal lint or CI maintenance.

Repository skills

  • Use .agents/skills/create-release/SKILL.md only when asked to prepare, submit, resume, or publish a CRAN release.

Testing

  • The package uses testthat edition 3 with parallel execution.
  • make check is the canonical full local validation command.
  • Snapshot tests are used extensively for both tidy statistical output and the expression column.
  • Tests cover source areas, but helper and shared source files may be exercised by broader test files rather than a one-to-one filename match.
  • Set seeds before Bayesian or otherwise stochastic tests.
  • Use skip_if_not_installed() for optional dependencies.
  • Suppress warnings only when a test intentionally exercises a warning-producing path.
  • Codecov requires 100% project and patch coverage.

Follow the existing snapshot style:

test_that("descriptive name", {
  df <- function_under_test(data = dataset, x = var1, y = var2)
  expect_snapshot(dplyr::select(df, -expression))
  expect_snapshot(df[["expression"]])
})

Code conventions

  • Use lintr for linting and styler for formatting.
  • Use snake_case for functions and variables.
  • Use the base R pipe (|>), not the magrittr pipe (%>%).
  • Preserve tidy evaluation for unquoted column arguments.

Roxygen documentation

  • Roxygen uses Markdown and the pkgapi and roxyglobals roclets configured in DESCRIPTION.
  • Use @autoglobal from roxyglobals where appropriate.
  • Shared extended examples live in man/rmd-fragments/.
  • After changing roxygen comments, run make document and commit the generated NAMESPACE or man/*.Rd changes. Do not edit generated .Rd files by hand.

Common function parameters

  • data: Input data frame.
  • x, y: Unquoted column names using tidy evaluation.
  • type: Usually one of "parametric", "nonparametric", "robust", or "bayes" where supported.
  • paired: Whether the design is paired or within-subjects.
  • digits: Number of decimal places.
  • conf.level: Confidence level between 0 and 1.

Important patterns

Statistical method selection

Functions normalize supported type values with extract_stats_type() and then select the appropriate statistical and effect-size functions. Follow the existing per-function branching and argument construction; do not assume one shared switch() shape applies to every analysis.

Expression generation

Statistical results are standardized and passed to add_expression_col() or the specialized expression helpers. Keep returned columns and attributes stable because ggstatsplot consumes them.

Easystats integration

Use tidy_model_parameters() and tidy_model_effectsize() to normalize easystats output rather than duplicating column-renaming and confidence-interval logic.

Files to update together

When modifying a function, consider all relevant surfaces:

  1. R/<function>.R or its helper file.
  2. The corresponding files under tests/testthat/.
  3. Generated man/<function>.Rd after roxygen regeneration.
  4. man/rmd-fragments/<function>.Rmd when that fragment exists.
  5. NEWS.md for user-facing changes.

CI/CD

Workflows under .github/workflows/ run standard and hard R CMD checks, coverage, documentation and extra checks, formatting, linting, prek hooks, pkgdown builds, and deployment tasks. Most jobs call reusable workflows from IndrajeetPatil/workflows; update the callers rather than copying those workflows into this repository.

The shared R CMD check matrix intentionally covers R-devel, release, and oldrel. Do not reintroduce oldrel-2 unless the package support policy changes.

Open pull requests as ready for review rather than as drafts. Unless explicitly requested, do not wait for CI/CD checks to finish after pushing; report that the checks were triggered and include the pull request or workflow link.