Skip to contents

Changelog

Source: NEWS.md

golfastr 0.3.0

Win probability

  • load_holes() files now carry win_prob, win_prob_before, wpa (win probability added) and skill_rating for every hole of every PGA Tour stroke play event since 2004.
  • add_win_probability() computes the same columns for an event in progress or for your own hole data.
  • The model plays out the rest of the event from every hole: each hole is drawn from how the field played it that day, shifted for each player’s skill rating and for how the player has done this week and this round. Cuts and playoffs are handled, and the result is exact once no cut is left. Fit on 2004-2019, it is calibrated on 2020-2026 events it never saw.

Database

  • update_db() builds a local SQLite database (or writes to any DBI connection) from the hosted data, in one schema: every table has a season column, ShotLink tables carry both PGA Tour and ESPN IDs, and dates are text. Later runs add missing seasons and refresh the latest.

Hole history

  • load_holes() covers 2004 to now (was 2025 to now), including playoff holes as an extra round.
  • Hole pars come from each player’s course card for the round, which fixes ESPN par labels that were off (for example a par 6 on a par 4 hole) and the scores to par built from them.

IDs

Logos

  • load_logos() returns logo image URLs for tours, tournaments, ShotLink courses and clubs (every host of a men’s or women’s major, the Ryder Cup, Presidents Cup or Solheim Cup since 1960), with PGA Tour, ESPN and ShotLink IDs to join on. Only URLs are provided, not images.

What’s in the bag

  • load_witb() returns player equipment from the what’s in the bag database (WITBDB): one row per club with head, loft, shaft and grip specs, parsed from GolfWRX and PGA Tour Club Tracker bag posts, with the source URL of every bag. load_witb_switches() lists each player’s equipment changes by slot.
  • load_shots() returns every stroke from PGA Tour ShotLink, 2012 to now: distance, distance to the hole, lie before and after, play-by-play text, and, in recent seasons, shot coordinates and launch monitor readings (ball speed, spin, launch angle, apex height).
  • load_tee_times() returns pairings, tee times and starting holes for every round.
  • load_course_setup() returns par, yardage and pin position by hole and round.
  • All three read season files from GitHub releases.

Strokes gained

  • calculate_strokes_gained() scores every shot against a bundled baseline of expected strokes by lie and distance, fit from ShotLink strokes.
  • summarise_strokes_gained() totals shots by round, tournament or season, adjusted to the field. Season totals track the PGA Tour’s official strokes gained total closely (correlation 0.97 in 2025 and 0.98 in 2026) and cover every player in the field, not only those with enough measured rounds.

Analysis

  • Analysis and plot functions now accept a data argument, so they work directly on tibbles from load_leaderboard(): player_summary("Scheffler", data = load_leaderboard(2025)).
  • Column names from hosted/live loaders (player_name, tournament_id, score_to_par) and older saved files (display_name, event_id, score_display) are aligned automatically.
  • save_to_rds() / save_to_parquet() deduplicate on tournament_id as well as the older event_id column.
  • load_strokes_gained() gains a year argument to filter by season.

Website and docs

  • Function reference hides internal helpers.
  • README, getting-started article, and field descriptions match the current loaders. score_to_par of "E" (even par) is documented.
  • DESCRIPTION lists all supported tours.

Other

  • ESPN requests time out after 30 seconds instead of hanging.
  • Bundled strokes gained snapshot is updated through the 2026 season.

Website

  • golfastr.com is now a full pkgdown site: the README front page, a grouped function reference, a getting-started guide, and a field-descriptions article replace the old placeholder page.

Data releases

  • Season assets are now published in three formats — rds, parquet, and csv — so the hosted data can be used directly from Python and other tools.
  • Hole-by-hole files gain derived columns computed at build time: to_par, field_avg (field scoring average on that hole in that round), vs_field (strokes vs the field average), and cume_to_par (running tournament score through each hole, playoff holes included).

Deprecations

Other improvements

  • Errors and messages are formatted with the cli package.
  • Hosted-data downloads can be cached across R sessions: set options(golfastr.cache = "filesystem") to keep downloads for 24 hours in the package cache directory, or "off" to disable caching. The default ("memory") caches for the current session as before.
  • R CMD check now runs on GitHub Actions for every push and pull request.

Hosted Season Data

  • load_leaderboard() and load_holes() now load completed PGA Tour tournaments from pre-built files hosted on GitHub releases (updated daily) instead of querying ESPN tournament by tournament. Loading a full season of leaderboards drops from minutes to seconds, and hole-by-hole data now covers the full field rather than requiring one API request per player.
  • Both functions gain a live argument to bypass the hosted data and fetch directly from ESPN (e.g., for in-progress tournaments). The live API is also used automatically as a fallback when hosted data is unavailable.
  • load_holes() output now includes position (final leaderboard standing) and year columns.
  • New data-raw/update_season_data.R pipeline builds and uploads the hosted files; runs daily via GitHub Actions.

golfastr 0.2.0

CRAN release: 2026-02-18

  • Added load_strokes_gained() with a bundled PGA Tour strokes gained dataset.
  • Added LIV Golf support via tour = "liv".

golfastr 0.1.5

CRAN release: 2026-02-12

CRAN resubmission with documentation fixes.

  • Changed ESPN API URLs from HTTP to HTTPS
  • Added missing @return tags to legacy functions
  • Fixed README logo to use absolute URL instead of relative path
  • Clarified \dontrun{} vs \donttest{} usage in examples

golfastr 0.1.4

This release focuses on CRAN compliance and simplifying the storage backend.

Breaking Changes

  • Removed save_to_db() and load_from_db() - DuckDB storage has been replaced with simpler file-based storage
  • All functions using db_path parameter now use file_path instead
  • Removed DBI dependency

New Functions

Function Updates

Backend Changes

  • Cache directory now uses tools::R_user_dir() for CRAN compliance
  • Replaced all cat() calls with message() for suppressible output
  • arrow package moved to Suggests (optional, only needed for Parquet)

golfastr 0.1.0

Initial release.

Data Loading

Analysis Functions

Visualization

Storage

Caching

Data Sources