Skip to contents

Changelog

Source: NEWS.md

huerd (development version)

Bug Fixes

  • plot_palette_analysis() no longer advances the global random number generator state: the point jitter of the comparative panels is now drawn under withr::with_preserve_seed(), so rendering a dashboard no longer silently changes downstream random results
  • plot_palette_analysis() now treats an OKLAB matrix input like the equivalent hex vector: the reference-palette panel selects by color count instead of length() of the matrix (3 cells per color previously pushed 3+ color matrices into the large-palette branch), and a single-row matrix now returns early like a single hex color. Undetermined (NA) pairwise distances render as neutral grey cells in the distance heatmap instead of transparent ones

Improvements

  • The pairwise distance heatmap of the plot_palette_analysis() dashboard now colors cells on a fixed scale (from the distinctness threshold to the theoretical maximum OKLAB distance of ~1.0) instead of normalizing to each palette’s own maximum; cell colors now encode absolute distances and are directly comparable across palettes and dashboards
  • The dashboard panel descriptions in the “Introduction to huerd” vignette now match the actual panels (the long-removed “Nearest Neighbor Distances” panel is no longer listed, and the two boxplot panels are described)
  • is_cvd_safe()’s min_cvd_distance default now derives from the package-wide distinctness threshold constant instead of a duplicated 0.08 literal, so the two can no longer drift apart silently (default value unchanged)
  • The DESCRIPTION Title is now in title case, resolving an R CMD check --as-cran NOTE

Deprecated

  • optimizer = "nlopt_direct" in generate_palette() is now deprecated and emits a warning; it will be removed in a future release. The DIRECT algorithm’s center-lattice sampling cannot reliably find configurations with all pairwise-separated colors in this parameterization, so it returns degenerate palettes (all colors collapsing to a single hex) for most palette sizes; increasing max_iterations does not help. Use the default "nloptr_cobyla" or "nlopt_neldermead" instead.

Internal Changes

  • Removed more dead internal code and doc drift: the unused .calculate_luminance() helper and the empty R/data.R; the text-only color print helper is renamed .print_color_with_swatch() -> .print_color_label() and its roxygen (plus print_color_vector()’s) no longer claims swatches; .handle_no_free_colors() roxygen documents generation_metadata and drops the nonexistent optimize_for
  • initialize_kmeans_plus_plus() candidate-center distance computation vectorized: the nested apply() loops over candidates and centers are replaced by a per-center rowSums()/pmin() accumulation. Results are bit-identical (verified with identical() across 80 seed x configuration combinations and at the generate_palette() level), with a large speedup for palettes with many free colors: the n_free = 100 tests drop the helpers-init file from ~224 s to ~4 s
  • Test suite pruned of non-behavioral slop: 65 tests removed (function-existence and formals() signature checks, requireNamespace()-style dependency checks, constant/export existence assertions, and a “gamut penalty” test that re-implemented arithmetic from an implementation that no longer exists); no behavioral coverage lost. The .merge_aesthetic_config() version-mismatch snapshot now records an input reachable through the public API (config_version = 999) instead of a string version and an unknown key, both of which validate_inputs() rejects before the merge ever sees them
  • Deduplicated the five optimizer implementations (optimize_colors_constrained(), optimize_colors_sann(), optimize_colors_nlopt_direct(), optimize_colors_nlopt_neldermead(), optimize_colors_lbfgs()) into a single shared internal driver; roughly 1,000 lines removed with no behavior change (verified byte-identical outputs across all optimizers)
  • Removed the unreachable optimization-state-tracking machinery: the track_states, save_every, and return_states parameters of the optimizers and the corresponding optimization_states plumbing; no public API or caller ever used them
  • Removed dead internal code: the unused validate_color_input_smart() / validate_hex_colors() / validate_oklab_matrix() validation subsystem, the legacy multi-objective objective_function_aggregator(), the unused .validate_balance_weights(), .validate_aesthetic_weights(), and .normalize_weights() helpers, and the unused track-free print indirection .print_color_item()
  • Removed three aesthetic_init_config keys that were accepted but never used by any code (kmeans_C_sd_multiplier, harmony_hcl_L_min_sd, harmony_hcl_C_min_sd); passing them now errors with an “Invalid name” message instead of being silently ignored
  • Cleaned unused NAMESPACE imports (colorspace::simulate_cvd, stats::setNames, stats::var, grDevices::rgb)
  • Removed all redundant NAMESPACE importFrom directives (every call site uses explicit pkg:: qualification); the last two unqualified calls (dist(), hcl.colors()) were namespace-qualified

huerd 0.6.2 (2026-08-16)

Bug Fixes

  • interpret_palette_quality() no longer errors on palettes with more than 40 colors, where the theoretical-maximum distance lookup is undefined and performance_ratio is NA; the summary now reports that the percentage is not available
  • validate_color_input_smart() now reports the actual problem (“Colors must be a character vector (hex codes) or an OKLAB matrix”) for wrong-type inputs instead of an unrelated “no base colors provided” message
  • export_palette() documentation now correctly states that "hex" returns a character vector, not a single string

New Features

  • generate_palette() gains a cvd_safe argument (default TRUE): when TRUE, the objective maximizes the worst-case perceptual distance across deuteranopia, protanopia, and tritanopia simulations; when FALSE, it optimizes normal-vision distance only. Applies to all derivative-free optimizers; documented as having no effect for "nlopt_lbfgs", whose smooth objectives are normal-vision only
  • The previously ignored cvd_safe arguments of quick_palette() and brand_palette() are now forwarded to generate_palette() and take effect
  • reproduce_palette() forwards the stored cvd_safe setting; palettes generated before 0.6.2 reproduce with the previous default behavior

Infrastructure

  • Moved withr from Suggests to Imports: reproduce_palette() calls withr::with_preserve_seed() on its default code path, which must not fail for installations without suggested packages
  • Removed the committed R/.tldr tool cache from version control (already excluded from the built tarball via .Rbuildignore)
  • Removed redundant Author/Maintainer fields from DESCRIPTION (regenerated automatically from Authors@R)

huerd 0.6.1 (2026-01-26)

Bug Fixes

  • plot_palette_analysis() now returns invisible(NULL) without warning when fewer than two colors are supplied.
  • simulate_palette_cvd() returns empty results silently when input contains no valid colors.
  • plot_cvd_comparison() now preserves the fallback border color when color conversion fails.
  • Optimizers validate inputs and return structured results instead of failing on invalid inputs (optimize_colors_constrained(), optimize_colors_sann()).

Internal Changes

  • Improved robustness in calculate_perceptual_distances() for NULL/empty/NA inputs.
  • Strengthened initialization parameter validation for aesthetic initialization bounds and influence settings.
  • Expanded .Rbuildignore to exclude additional development artifacts.

huerd 0.6.0 (2026-01-18)

Bug Fixes

  • Fixed hex color validation regex in evaluate_palette() to properly reject literal “NA” strings while allowing actual NA values
  • Fixed flaky test in test-optimization-core.R for nlopt_neldermead optimizer performance expectations
  • Fixed reproduce_palette() to properly restore RNG state using withr::with_preserve_seed() instead of direct .Random.seed manipulation
  • Fixed validate_oklab_matrix() test errors by properly initializing validation_result structure

Code Quality Improvements

  • Added comprehensive documentation to internal helper functions with @param descriptions
  • Improved error handling and validation in export_palette() for different export formats (JSON, CSV, CSS, Sass)
  • Enhanced L-BFGS optimizer state consistency by adding colors_oklab field and defensive clamping
  • Updated all tests to use testthat 3rd edition describe/it BDD pattern (1,006+ tests)
  • Fixed CVD analysis tests to handle list structure return from simulate_palette_cvd()

New Features

Infrastructure

  • Added withr to Suggests for proper RNG state management
  • Updated .gitignore with .venv/ and pkgdown/ patterns
  • Updated .Rbuildignore with corrected .venv exclusion pattern
  • Added comprehensive vignettes for data scientist, designer, and package developer workflows

huerd 0.5.3

Breaking Changes

  • simulate_palette_cvd() with a single cvd_type now returns a named list instead of a character vector; access the simulated colors via result$simulated. The print method was updated accordingly

Code Quality Improvements

This release focuses on code quality, maintainability, and consistency improvements identified through comprehensive code review.

User Messaging

  • Replaced cat() calls with semantic cli messaging functions (cli_alert_info(), cli_alert_warning(), cli_alert_success(), cli_inform()) for better user experience
  • Added cli to package Imports

Type Safety & Performance

  • Replaced sapply() with vapply() for type-stable return values
  • Replaced is.na() with anyNA() for more efficient NA detection
  • Fixed style issues: = → <- for assignment, 1:n → seq_len(n) for safer indexing, T/F → TRUE/FALSE

Code Organization

  • Added named constants for magic numbers: .CANDIDATE_POOL_BASE, .MIN_DISTANCE_THRESHOLD, .OKLAB_TOLERANCE
  • Extracted repeated clamping code to .clamp_to_bounds() helper function
  • Added @param documentation to internal helper functions
  • Removed dead/commented-out code

API Consistency

  • Added ... parameter to exported functions for future extensibility
  • simulate_palette_cvd() return structure unified across single and multiple cvd_type values (see Breaking Changes above)

Bug Fixes

  • Corrected gradient formulas in smooth optimization objectives (gradient_smooth_repulsion() and gradient_smooth_logsumexp())
  • Added input validation to gradient functions
  • Added numerical gradient verification tests to ensure correctness

huerd 0.5.2 (2025-07-21)

Bug Fixes

  • Corrected distance matrix calculation in smooth optimization objectives where diagonal masking affected only element [1,1] rather than the full diagonal
  • Eliminated redundant distance calculations by using dist() output directly instead of symmetric matrix conversion
  • Applied log-sum-exp numerical stability technique for robust computation under extreme parameter values
  • Added regression tests for distance calculation correctness

huerd 0.5.0 (2025-07-07)

New Features

Smooth Optimization Support

  • Added L-BFGS optimizer for gradient-based optimization
  • Implemented two smooth objective functions: smooth_repulsion and smooth_logsumexp
  • Added analytical gradient computation for faster convergence
  • Speed improvements of 5-20x for larger palettes (8+ colors)

Multi-Objective Framework Updates

  • Extended framework to support smooth and discrete objectives
  • Support for weighted combinations of smooth objectives

API Changes

  • Added optimizer = "nlopt_lbfgs" option for L-BFGS optimization
  • Added weights = c(smooth_repulsion = 1) and weights = c(smooth_logsumexp = 1) for smooth objectives

Bug Fixes

  • Fixed critical bug where L-BFGS optimizer always used smooth_repulsion objective regardless of weights parameter
  • Fixed crash in evaluate_palette_quality() when called directly with hex colors (“argument is of length zero” error)
  • Both smooth objectives (smooth_repulsion and smooth_logsumexp) now work correctly and produce different optimization results

Internal Changes

  • Added analytical gradient functions for smooth objectives
  • Extended multi-objective framework to handle smooth and discrete objectives
  • Improved input validation in evaluate_palette_quality() function

huerd 0.4.2 (2025-07-07)

New Features

  • Added L-BFGS optimizer (optimizer = "nlopt_lbfgs") for gradient-based optimization
  • Added support for smooth optimization using analytical gradients

Internal Changes

  • Added optimize_colors_lbfgs() function using NLOPT_LD_LBFGS algorithm

huerd 0.4.1 (2025-07-07)

Internal Changes

  • Added smooth differentiable objective functions for future gradient-based optimization

huerd 0.4.0 (2025-07-06)

Major Changes

New Features

  • Added reproducibility system for palette generation with metadata storage.

  • Added hue order reversal detection for CVD analysis.

Breaking Changes

  • Removed new_device, device_width, and device_height parameters from plot_palette_analysis().

  • Moved gridExtra from Suggests to Imports, added grid to Imports.

  • Removed helper functions: .create_graphics_device, .is_ragg_available, .is_interactive_ide, calculate_safe_margins.

huerd 0.3.1 (2025-07-04)

New Features

  • Added automated font size scaling to plot_palette_analysis() that automatically adjusts text sizes based on device dimensions.

huerd 0.3.0 (2025-07-04)

Major Changes

  • Multi-optimizer framework: The generate_palette() function now supports four different optimization algorithms through the new optimizer parameter:
    • "nloptr_cobyla" (default): Deterministic optimization with constraint handling
    • "sann": Stochastic simulated annealing
    • "nlopt_direct": Deterministic global optimization using DIRECT algorithm
    • "nlopt_neldermead": Derivative-free local optimization using Nelder-Mead simplex
  • Added automatic font scaling to plot_palette_analysis() that prevents text overlap and out-of-frame issues.

Minor Changes

  • Added internal utility functions .hex_to_oklab() and .oklab_to_hex() to eliminate repetitive color conversion patterns.

  • Fixed failing visualization tests and added test coverage for new utility functions.

  • Updated test files to use utility functions.

huerd 0.1.1 (2025-07-01)

Minor Changes

  • Removed unused cli and crayon dependencies.

huerd 0.1.0 (2025-07-01)

This is a major refactoring and simplification of huerd, focusing the package on a single, scientifically-grounded objective: pure minimax color palette generation.

Major Changes

  • Pure minimax optimization: The core generate_palette() function has been streamlined to use a pure minimax objective, maximizing the minimum perceptual distance between colors in the OKLAB space. All complex multi-objective parameters (optimize_for, balance_weights, aesthetic_penalty_weights_LC) have been removed, simplifying the API and aligning the package with a clear, defensible scientific goal.

  • Comprehensive diagnostic dashboard: A new plot_palette_analysis() function has been added. It provides a comprehensive, six-panel diagnostic dashboard inspired by scicomap for in-depth palette analysis. This feature uses only base R graphics and introduces zero new dependencies.

  • Automatic brightness sorting: All palettes generated by generate_palette() are now automatically sorted by their OKLAB lightness value, ensuring intuitive and consistent ordering from darkest to lightest.

  • Simplified evaluation: The evaluate_palette() function is now a pure data provider, returning raw, objective metrics (distances, CVD safety, color distribution) without any subjective heuristic scoring.

Minor Changes

  • The is_cvd_safe() function now uses the more robust evaluate_palette() function for its calculations.
  • The print methods for huerd_palette and huerd_evaluation have been updated to reflect the new, simplified data structures.

Removals

The following experimental and non-essential features have been removed to streamline the package:

  • Force-field and repulsion-based optimization algorithms.
  • Palette animation framework.
  • Algorithm comparison tools.
  • Post-hoc CVD refinement steps.
  • 3D color space visualizations.
  • All functionality related to the CAM16 and CIECAM16 color models.

huerd 0.0.3 (2025-06-22)

  • Added a post-hoc cvd_adjustment step to improve palettes for color vision deficiency.
  • Added a README.md with basic usage instructions.

huerd 0.0.2 (2025-01-12)

  • Switched the core optimization algorithm to an inverse-square law repulsion model.
  • Removed all functionality related to the CIECAM16 color model.
  • Added a swatchplot to the animation feature.

huerd 0.0.1 (2025-01-04)

  • Initial release.
  • Core functionality for palette generation using the CAM16 and CIECAM16 color models.
  • Features for color space conversion and animation of the optimization process.