feat: gg_roc per_class=TRUE — per-class OvR ROC curves (#88, closes #72)

[ehrlinger]

Summary

• gg_roc() gains per_class = FALSE. When per_class = TRUE on a multi-class forest, returns a long-format gg_roc data frame with a class factor column and a named AUC vector attribute (ordered by descending AUC).
• plot.gg_roc() gains panel = c("overlay", "facet"). Detects the class column and dispatches to the new multi-class overlay or faceted path.
• summary.gg_roc() updated to print named per-class AUC values when the class column is present.
• Binary forests: per_class = TRUE is a silent no-op.
• ROC CIs deferred to v2.9.0 (issue Confidence intervals on ROC curves.  #7 / Meta: gg_roc enhancements (multi-class + confidence intervals) #72-CIs).

Implementation reuses the existing private .rf_prob_matrix / .rf_one_class_roc / calc_auc helpers in R/calc_roc.R — no new abstractions.

Test plan

• devtools::test() — gg_roc suite 79/79 pass; full suite 0 failures
• devtools::check(args="--as-cran") — 0 errors, 0 warnings, 0 notes
• 2 vdiffr snapshots added (overlay, facet) — skip cleanly without VDIFFR_RUN_TESTS

Closes #72

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 98.57143% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 87.31%. Comparing base (c580011) to head (6ad6a9e).

Files with missing lines	Patch %	Lines
R/plot.gg_roc.R	96.96%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #91      +/-   ##
==========================================
+ Coverage   87.11%   87.31%   +0.19%     
==========================================
  Files          38       38              
  Lines        3043     3098      +55     
==========================================
+ Hits         2651     2705      +54     
- Misses        392      393       +1     

Files with missing lines	Coverage Δ
R/gg_roc.R	87.71% <100.00%> (+5.21%)	⬆️
R/summary_methods.R	92.35% <100.00%> (+0.25%)	⬆️
R/plot.gg_roc.R	93.75% <96.96%> (+1.34%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

Adds multi-class one-vs-rest ROC support to gg_roc() for randomForest models via a new per_class mode, and updates plotting/summaries/tests/docs to support per-class output.

Changes:

• gg_roc(..., per_class = TRUE) now returns long-format per-class ROC data (with class) plus a named per-class AUC attribute (sorted by descending AUC) for multi-class randomForest objects.
• plot.gg_roc() adds panel = c("overlay", "facet") and renders per-class ROC objects accordingly; summary.gg_roc() prints named per-class AUCs.
• Adds testthat coverage and vdiffr snapshots; updates NEWS/versioning and Rd docs.

Reviewed changes

Copilot reviewed 8 out of 10 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
R/gg_roc.R	Implements per_class long-format ROC generation + per-class AUC attribute for randomForest.
R/plot.gg_roc.R	Adds panel arg and per-class plotting path.
R/summary_methods.R	Updates ROC summary output to print per-class AUCs when class column is present.
tests/testthat/test_gg_roc.R	Adds unit tests for per_class, plotting modes, and summary output.
tests/testthat/test_snapshots.R	Adds opt-in vdiffr snapshots for per-class overlay/facet plots.
man/gg_roc.rfsrc.Rd	Documents new per_class argument.
man/plot.gg_roc.Rd	Documents new panel argument.
NEWS.md	Announces per-class ROC feature and behavior notes.
DESCRIPTION	Bumps development version.
.Rbuildignore	Ignores a new dotfile.

Files not reviewed (2)

• man/gg_roc.rfsrc.Rd: Language not supported
• man/plot.gg_roc.Rd: Language not supported

Comments suppressed due to low confidence (3)

R/plot.gg_roc.R:117

• When x is a raw multiclass randomForest and the user passes per_class = TRUE via ..., the existing multiclass branch will still iterate over which_outcome and call gg_roc(..., which_outcome = ind, per_class = TRUE). Since gg_roc.randomForest() ignores which_outcome when per_class = TRUE, this produces duplicated long-format ROC objects and then falls into the list-of-curves plotting path (wrong output / repeated messages). plot.gg_roc() should detect per_class = TRUE in ... for raw forests and call gg_roc(x, per_class = TRUE, ...) once, then use the per-class plotting branch.

  ## ---- Accept a raw rfsrc or randomForest object -----------------------
  # If we call this with a forest object instead of a gg_roc object,
  # automatically compute the ROC data for the appropriate class(es).
  if (inherits(gg_dta, "rfsrc")) {
    if (inherits(gg_dta, "class")) {
      # Determine the number of outcome classes
      crv <- dim(gg_dta$predicted)[2]

      if (crv > 2 && is.null(which_outcome)) {
        # Multi-class: compute ROC for every class in parallel
        gg_dta <- mclapply(seq_len(crv), function(ind) {
          gg_roc(gg_dta, which_outcome = ind, ...)
        })
      } else {
        # Binary (or user specified a class): default to second column
        if (is.null(which_outcome)) {
          which_outcome <- 2
        }
        gg_dta <- gg_roc(gg_dta, which_outcome, ...)
      }
    } else {
      stop("gg_roc expects a classification randomForest.")
    }
  } else if (inherits(gg_dta, "randomForest")) {
    if (gg_dta$type == "classification") {
      # Determine the number of outcome classes
      crv <- length(levels(gg_dta$predicted))
      if (crv > 2 && is.null(which_outcome)) {
        # Multi-class: compute ROC for every class in parallel
        gg_dta <- parallel::mclapply(seq_len(crv), function(ind) {
          gg_roc(gg_dta, which_outcome = ind, ...)
        })
      } else {

R/plot.gg_roc.R:135

• In the per-class plotting branch, AUC is read from attr(x, "auc"), but x is not guaranteed to be the per-class gg_roc object (e.g., when plot.gg_roc() is called with a raw forest and internally computes gg_dta). Use attr(gg_dta, "auc") so the caption works whenever gg_dta carries the AUC attribute.

    # Rendering is delegated to a helper to keep this method's cyclomatic
    # complexity within the project lint budget.
    if ("class" %in% names(gg_dta)) {
      return(.plot_gg_roc_per_class(gg_dta, attr(x, "auc"), panel))

R/plot.gg_roc.R:88

• The new per-class plotting behavior is only exercised when plot() is called on an already-computed gg_roc object. Given that plot.gg_roc() also accepts raw forest objects and forwards ... to gg_roc(), adding a test that plot(rf, per_class = TRUE, panel = ...) returns a ggplot (and does not emit repeated ignore-messages) would help prevent regressions in this important entry point.

plot.gg_roc <- function(x, which_outcome = NULL,
                        panel = c("overlay", "facet"), ...) {
  panel <- match.arg(panel)
  gg_dta <- x

  ## ---- Accept a raw rfsrc or randomForest object -----------------------
  # If we call this with a forest object instead of a gg_roc object,
  # automatically compute the ROC data for the appropriate class(es).
  if (inherits(gg_dta, "rfsrc")) {

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.