Scatterpie plot with patterns

I'm using the R scatterpie package for plotting pie charts on an xy plane. Each one of the pies/groups also has a class assignment which I would also like to convey in the plot and was thinking of using patterns for that.

Here's the example data.frame:

library(ggplot2)
library(ggpattern)
library(scatterpie)
library(dplyr)
set.seed(1)
df <- data.frame(x = runif(50, 0, 100), y = runif(50, 0, 100), id = paste0("id",1:50),
                 g1 = as.integer(runif(50, 0, 100)), g2 = as.integer(runif(50, 0, 100)), g3 = as.integer(runif(50, 0, 100)),
                 class = sample(c("c1", "c2"), 50, replace = T)) %>%
  dplyr::mutate(size = 0.02*(g1 + g2 + g3))
df$class <- factor(df$class)

Here's the code I tried with the ggpattern package:

ggplot() + geom_scatterpie(data = df, aes(x = x, y = y, group = id, r = size),cols = paste0("g",1:3),color = "black", alpha = 0.8, size = 0.1) +
  scale_fill_manual(values=c("#FFC9F8", "#ee9e77", "#895A44")) +
  geom_col_pattern(aes(pattern_fill = class))+
  xlim(-10,110) + ylim(-10,110) + coord_equal() + theme_void()

Which gives the error:

Don't know how to automatically pick scale for object of type function. Defaulting to continuous.
Error in `f()`:
! Aesthetics must be valid data columns. Problematic aesthetic(s): pattern_fill = class. 
Did you mistype the name of a data column or forget to add after_stat()?

I tried playing around with the pattern part of the command but with no success.

Currently, the only way I am able to encode the class of each pie is with the linetype aesthetics argument:

ggplot() + geom_scatterpie(data = df, aes(x = x, y = y, group = id, r = size, linetype = class),cols = paste0("g",1:3), color = "black", size = 1) +
  scale_fill_manual(values=c("#FFC9F8", "#ee9e77", "#895A44")) +
  xlim(-10,110) + ylim(-10,110) + coord_equal() + theme_void()

But it's not ideal.

Any idea how to make this work or alternative ways to encode class?

I'm not convinced about the visualization, though it is possible via ggplot and ggpattern. I don't think you can use geom_scatterpie, which, as Axeman points out, doesn't really work the way other ggplot geoms do.

If I had to do this, I would probably just make my own pie function and draw the result with geom_polygon_pattern.

You could do that with the following helper function:

make_pie <- function(x, y, size, groups, n, class, rownum) {
  angles <- c(0, 2*pi * cumsum(n)/sum(n))
  do.call("rbind", Map(function(a1, a2, g) {
    xvals <- c(0, sin(seq(a1, a2, len = 30)) * size, 0) + x
    yvals <- c(0, cos(seq(a1, a2, len = 30)) * size, 0) + y
    data.frame(x = xvals, y = yvals, group = g, class = class, rownum = rownum)
  }, head(angles, -1), tail(angles, -1), groups))
}

You can then plot using your data as follows:

df %>%
  mutate(r = row_number()) %>%
  rowwise() %>%
  group_map(~ with(.x, make_pie(x, y, size, c("g1", "g2", "g3"),
                                c(g1, g2, g3), class, r))) %>%
  bind_rows() %>%
  ggplot(aes(x, y, fill = group, group = interaction(group, rownum))) +
  geom_polygon_pattern(aes(pattern = class), pattern_fill = "black",
                       pattern_angle = 45, pattern_spacing = 0.01,
                       pattern_density = 0.1) +
  scale_fill_manual(values = c("#FFC9F8", "#ee9e77", "#895A44")) +
  guides(pattern = guide_legend(override.aes = list(fill = "white"))) +
  coord_equal() +
  theme_void()

You could map the alpha channel as Axeman suggests like this:

df %>%
  mutate(r = row_number()) %>%
  rowwise() %>%
  group_map(~ with(.x, make_pie(x, y, size, c("g1", "g2", "g3"),
                                c(g1, g2, g3), class, r))) %>%
  bind_rows() %>%
  ggplot(aes(x, y, fill = class, group = interaction(group, rownum),
             alpha = group)) +
  geom_polygon() +
  scale_fill_manual(values = c("red4", "green4")) +
  scale_alpha_manual(values = c(0.3, 0.7, 1)) +
  coord_equal() +
  theme_void()

Or just declare each fill color individually:

df %>%
  mutate(r = row_number()) %>%
  rowwise() %>%
  group_map(~ with(.x, make_pie(x, y, size, c("g1", "g2", "g3"),
                                c(g1, g2, g3), class, r))) %>%
  bind_rows() %>%
  mutate(fill_class = paste0(group, " (class ", class, ")")) %>%
  mutate(fill_class = factor(fill_class, unique(fill_class))) %>%
  ggplot(aes(x, y, fill = fill_class, group = interaction(group, rownum))) +
  geom_polygon() +
  scale_fill_manual("Group (class)",
                    values = c("red3", "hotpink", "red", 
                               "navy", "blue", "dodgerblue")) +
  coord_equal() +
  theme_void()