NEWS

modeltime.resample 0.3.0 (2025-09-03)

Compatibility with tune 2.0.0
- Updated internals to support new tune::fit_resamples() behavior.
- Improved handling of .predictions column reconstruction when missing.
- Added robust fallback logic to normalize truth/prediction columns across versions.
Improved Resampling Functions
- modeltime_fit_resamples() now ensures predictions are always saved.
- Added deterministic seeding (withr::with_seed()) for reproducible resample fits.
- Better error handling: failed resample fits now produce .notes with clear error messages.
Plotting Enhancements
- plot_modeltime_resamples() now standardizes truth/estimate detection.
- Improved facetting and summary-line consistency.
- More graceful error messages if truth/pred columns cannot be identified.
- Optional interactive output improved with Plotly checks.
Utility Upgrades
- unnest_modeltime_resamples() more robust:
  - Ensures .predictions column exists (reconstructed if missing).
  - Clear actionable errors when predictions are unavailable.
  - Guarantees .row_id assignment for comparing models across resamples.

tune (>= 2.0.0) is now required.
Added withr to Imports.
Retained compatibility with older tune versions (<2.0.0) via conditional handling.

Standardized use of rlang::abort(message=...) for consistent error messages.
Reduced reliance on tictoc/progressr in favor of clearer progress reporting.
Expanded utils::globalVariables() for safer NSE handling across dplyr/tidyr.

modeltime_resample_accuracy() (#1): When user specifies summary_fns = NULL, returns unsummarized resample metrics with ".resample_id"