Easy way to split one field into n facets

[rjake]

I often create a heatmap that is too long to display on one chart. I have to scratch my head each time to figure out how to break it into columns. I eventually end up with some integer division solution x %/% y like here:

Code

library(tidyverse)
library(scales)

df <- 
  USJudgeRatings |> 
  as_tibble(rownames = "y") |> 
  gather(x, fill, -y) |> 
  mutate(
    x = fct_reorder(x, fill, mean),
    y = fct_reorder(y, fill, sum, .desc = FALSE)
  ) |> 
  print()

# number of columns for plot
n_col <- 2

# separate into columns and keep factor order
plot_prep <- 
  df |> 
  mutate(
    y_order = as.integer(y),
    max_rows = ceiling(n_distinct(y) / n_col),
    facets = (-y_order %/% max_rows) # needs something to facet on, need '-' to keep highest values on left
  ) |> 
  print()

# plot
p <- 
  plot_prep |> 
  ggplot(aes(x, y, fill = fill)) +
  facet_wrap(~facets, ncol = n_col, scales = "free_y") +
  geom_tile(color = "white", size = 0.5) +
  scale_fill_stepsn(breaks = breaks_pretty(6), colours = viridis_pal()(5)) +
  theme(
    strip.background.y = element_blank(),
    strip.text = element_blank(),
    axis.text = element_text(size = 7)
  )

p + theme(aspect.ratio = 1)

# identify aspect ratio
aspect <- 
  plot_prep |> 
  select(x, y) |> 
  summarise_all(n_distinct) |> 
  mutate(ratio = y / x / n_col) |> 
  print() 

p + theme(aspect.ratio = aspect$ratio) # <------ fixed 1:1 size

Could a solution be implemented like one of these?

• facet_wrap(facets = ~., ncol = 2) -- currently just keeps chart as-is (no facets)
• a helper function facet_wrap(facets = split_facet(ncol = 2))
• a new function facet_split(ncol = 2)

---

Additionally, related to #4584, I often struggle to get the ratios correct.
coord_fixed() throws an error and aspect.ratio affects each panel rather than "unit of y per unit of x".
I included code ☝️ to show how I do it: [# unique y] / [# unique x] / [# of cols]

[hadley]

This would be a nice feature, but since it could be implemented in a separate package, I think it should be. (That makes much easier to evolve rapidly because it's not strongly connected to the very slow ggplot2 development process). You might consider shopping this idea around the authors of the various ggplot2 extension packages.