---
title: "Analysis of Palette Colors"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Analysis of Palette Colors}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  fig.width = 9,
  fig.height = 3,
  comment = "#>"
)
```

```{r setup}
library(peRsian)
library(dplyr)
```

While there is always a qualitative aspect to choosing colors for data visualizations, it's also possible to quantitatively assess how distinguishable the colors within a palette are.

We can use the `colorblindcheck` package to evaluate how distinguishable the colors in each palette are, both under normal vision and various forms of colorblindness, which is particularly important for ensuring accessibility in visualizations.

## `peRsian` Palettes

Let's start by listing all available palettes in the `peRsian` package.

```{r}
names(persian_palettes)
```

Next, we can have a look at an evaluation of each palette using `colorblindcheck`.

```{r results='asis'}
# Generate a subjeading and evaluation for each palette
for(palette_name in names(persian_palettes)) {
  cat("\n\n### ", palette_name, "\n\n")
  
  colors <- persian_palette(palette_name)
  check_results <- colorblindcheck::palette_check(colors, plot = TRUE)
  
  check_results |>
    knitr::kable(digits = 2) |>
    print()
}
```

## Quantitative Comparison

With the scores produced by `colorblindcheck`, we can now perform a proper quantitative analysis to check which palettes meet certain thresholds and which might need adjustments.

### Colorblind-Safety

Let's compare `peRsian` with the popular `viridis` palette, which has specifically been designed to handle colorblindness well, to put its the color difference scores into perspective.

```{r}
# Helper function to conveniently calculate scores and assign a palette name
calculate_palette_scores <- function(colors, palette_name, ...) {
  check_results <- colorblindcheck::palette_check(colors, ...)
  cbind(
    data.frame(palette = palette_name),
    check_results
  )
}
```

First, we calculate the scores for `viridis`.

```{r}
viridis_scores_df <- calculate_palette_scores(
  viridisLite::viridis(8),
  "viridis",
  plot = TRUE
)
viridis_scores_df |>
  knitr::kable(digits = 2)
```

Let's use the smallest difference in `viridis` as the threshold to identify which palettes in `peRsian` we deem "colorblind-safe". 

```{r}
viridis_min_score <- min(viridis_scores_df$min_dist)
viridis_min_score
```

```{r}
# Create a dataframe with scores for all palettes in the package
persian_scores_df <- do.call(rbind, lapply(names(persian_palettes), function(palette_name) {
  calculate_palette_scores(
    persian_palette(palette_name),
    palette_name,
    tolerance = viridis_min_score
  )
}))
```

Now we can use this information to only mark palettes as colorblind-safe, where the smallest difference is above the threshold across ALL different forms of visual perception.

```{r}
# Keep only palettes where all entries are above threshold
palettes_above_threshold <- persian_scores_df |>
  group_by(palette) |> 
  # Number of comparisons == Number of comparisons above threshold
  filter(all(ncp == ndcp))

colorblind_safe_palettes <- palettes_above_threshold$palette |>
  unique()

colorblind_safe_palettes
```

Let's also collect the names of all palettes that didn't make it (use these with caution!).

```{r}
non_colorblind_safe_palettes <- names(persian_palettes)[!names(persian_palettes) %in% palettes_above_threshold$palette]

non_colorblind_safe_palettes
```

### Ordering Palettes

Finally, we can order all palettes by their minimum color difference (under full vision) to get a better overview of which ones are most distinguishable and which ones may need more work.

```{r}
persian_scores_df |>
  filter(name == "normal") |>
  arrange(desc(min_dist))
```

Palettes which may need adjustments:

```{r}
persian_scores_df |>
  filter(name == "normal") |>
  arrange(min_dist)
```