semantic color encoding w/ mall

[EvaMaeRey]

I shared this experiment with @edgararuiz and @cararthompson but I thought it also might be of interest for this group. I think the set-up is okay for a one-off, but might be interesting to think about pitfalls ... 🎨

[cararthompson]

I'm intrigued about what happened to the kangaroo (not a sentence I expected to type here, hehe). This is a fun experiment!

[EvaMaeRey]

I just searched 'kangaroo' in my repo to find the 'mall-w-colors7' is the kangaroo-including study! ... https://evamaerey.github.io/mytidytuesday/2026-04-25-mall-w-colors/mall-w-colors7.html

This is the code that precedes the featured code:

library(mall)
options(.mall_chat = ellmer::chat_google_gemini())

library(tidyverse)

my_palette <- c("grey9", "violet", "darkgreen", "midnightblue",
"turquoise", "darkorange", "grey", "brown",
"darkred", "burlywood", "goldenrod2", "slateblue")

my_palette |>
tibble(pal_colors = _) |>
ggplot() +
aes(x = pal_colors, y = "all",
fill = I(pal_colors)) +
geom_tile()

I kind of wondered, I think with Edgar, if kangaroo has less strong an association with any particular color, compared w/ other animals. But this is just a guess.

The same situation is in this other experiment - and warnings aren't supressed there, so possibly more informative...
https://evamaerey.github.io/mytidytuesday/2026-04-25-mall-w-colors/mall-w-colors10.html

I think a note from you @cararthompson was on retrieving the names (or hex) of colors being used in the plot?

[cararthompson]

I think my concern was that if you had a goldfish and a frog in one graph (which I would associate with darkorange and green) and say a fox, a goldfish and a frog in another, would the goldfish get the same colour as it did originally, or would the colours jump around unhelpfully between animals because we need a second dark orange-ish colour. I may have missed something, but my question really is: what how do we keep things stable across graphs if new colours are needed (extracting the hex codes and assigning them within a named vector that we could subsequently add to would be my manual approach).

[EvaMaeRey]

Yeah - it does seem like the color selection should be a pre-step and cached and not just in the plotting pipeline. And I think you could even have color 'categorizations' for the same input, if you have more disaggregated data - like if there are 20 rows of 'goldfish', and you are using geom_bar for aggregation - not sure of this. I this it could be fun to think about interplay w/ gg color and fill scales and llms, but haven't done too much thinking through.

Maybe:

animals |> 
  ggplot() + 
  aes(x = animal, fill = animal) + 
  geom_bar() + 
  scale_fill_semantic()  # unique (animal) values are breaks, allows user-defined palette input (like the 'labels' arg from llm_classify), and maybe quote back the colors that do get selected along with the value it represent in a message...