Fix aes_string() deprecation warning in diagnostics plots (#582)

[kalra-mohit]

Summary

ggplot2 >= 3.0.0 hard-deprecates aes_string(). Both plot.model_diagnostics() and plot.predict_diagnostics() called it directly, so every diagnostics plot emitted a lifecycle deprecation warning:

Warning message:
`aes_string()` was deprecated in ggplot2 3.0.0.
ℹ Please use tidy evaluation idioms with `aes()`.

Reproduced exactly using the reprex from #582:

library(DALEX)
library(randomForest)
set.seed(72)

model_apart_rf <- randomForest(m2.price ~ ., data = apartments)
explain_apart_rf <- DALEX::explain(
  model = model_apart_rf, verbose = FALSE,
  data = apartments_test[, -1], y = apartments_test$m2.price,
  label = "Random Forest"
)

diag_rf <- model_diagnostics(explain_apart_rf)
plot(diag_rf)  # <- warning fired here, pre-fix

Root cause / fix: both functions built their aes() mapping with variable-name strings (column names passed in as function arguments, e.g. variable = "y_hat"), which is exactly what aes_string() was for before tidy evaluation existed. The fix swaps in aes() + .data[[...]], which is the direct replacement ggplot2's own deprecation message points to, and keeps the "map by column-name string" behavior these functions depend on (variable/yvariable are user-facing string parameters).

While tracing this, I found plot.predict_diagnostics()'s histogram branch (used when variables = NULL) has the exact same latent bug — not mentioned in the issue title, but same root cause and same file area, so fixed it too rather than leaving a near-identical bug in place.

flowchart LR
    A["model_diagnostics(explainer)"] --> B["plot.model_diagnostics()"]
    B --> C["ggplot(all_models, aes_string(...))\n⚠ deprecated in ggplot2 >= 3.0.0"]
    C -->|"this PR"| D["ggplot(all_models, aes(x = .data[[variable]], ...))"]

    E["predict_diagnostics(explainer, ..., variables = NULL)"] --> F["plot.predict_diagnostics()\n(histogram branch)"]
    F --> G["ggplot(df, aes_string(...))\n⚠ same deprecation"]
    G -->|"this PR"| H["ggplot(df, aes(x = .data[[\"Var1\"]], ...))"]

Loading

Why it matters

model_diagnostics()/plot() and predict_diagnostics()/plot() are core, commonly-used entry points (residual diagnostics, instance-level diagnostics). The warning fires on every single call, which is noisy for users and — since lifecycle warnings can be promoted to errors by options(lifecycle_verbosity = "error") or under strict CRAN checks on newer ggplot2 — a forward-compatibility risk, not just cosmetic noise.

Test plan

• Reproduced the exact warning from plot.model_diagnostics() gives a warning for using deprecated aes_string() with ggplot2 >=3.0.0 #582's reprex against current master.
• Added a regression test to tests/testthat/test_model_diagnostics.R and tests/testthat/test_predict_diagnostics.R that asserts the plotting functions' source no longer references aes_string().
  • Checked this structurally (source inspection) rather than via expect_warning/expect_no_warning, because lifecycle deprecation warnings are only emitted once per R session — whichever test calls plot.model_diagnostics() first "uses up" the warning, so a warning-capturing test would silently pass on later calls in the same test run regardless of whether the underlying bug is fixed. Confirmed this empirically before choosing the structural check.
• Verified both new tests fail against the pre-fix source (failed: 1) and pass after the fix (failed: 0), by checking out the pre-fix version of the two R/ files, rebuilding, and rerunning testthat::test_file() on just the two affected test files, then restoring the fix and rerunning.
• Ran the full existing test_model_diagnostics.R and test_predict_diagnostics.R suites (not just the new tests) after the fix — all pass, and the pre-existing "Plot" tests no longer emit the aes_string() warning either.
• Verified the fixed plots produce identical data mappings (gg object, correct columns bound to x/y/color/fill) — this is a call-site swap, not a behavior change.

Ran only the two affected test files locally (per this session's scoped-testing convention); full R CMD check / CI will run the complete suite.

Closes #582

[mayer79]

color = label etc will do