How to apply "outer" on data frames in Tidyverse?

Question

I have a data frame that contains measurements for multiple groups. I want to apply outer to each pair of groups and compute a summary. Can it be done in Tidyverse style? Below is an example of how to do it with two loops. I have played around with expand_grid and group_by, but without success.

df <- iris
groups = as.character(levels(df$Species))
n <- length(groups)
out <- matrix(numeric(n*n), nrow = n, ncol = n)
colnames(out) <- groups
rownames(out) <- groups
for (sp1 in 1:n) {
  for (sp2 in 1:n) {
    out[sp1, sp2] <- sum(outer(
      unlist(subset(df, Species == groups[sp1], Sepal.Length)),
      unlist(subset(df, Species == groups[sp2], Sepal.Length)),
      ">"))
  }
}
out

           setosa versicolor virginica
setosa       1128        139        34
versicolor   2302       1169       479
virginica    2457       1927      1165

Answer 1

1) Base R With these sorts of calculations base R is probably more appropriate. We can split Sepal.Length into a 3 element list s, one component per Species, and then use outer or a double sapply.

s <- with(iris, split(Sepal.Length, Species))
outer(s, s, Vectorize(\(x, y) sum(outer(y, x, "<"))))

##            setosa versicolor virginica
## setosa       1128        139        34
## versicolor   2302       1169       479
## virginica    2457       1927      1165

# alternately replace the line with outer with this
sapply(s, \(u) sapply(s, \(v) sum(outer(u, v, "<"))))

2) magrittr We could make this into a magrittr pipeline like this:

library(magrittr)
iris %>%
  with(split(Sepal.Length, Species)) %>%
  outer(., ., Vectorize(\(x, y) sum(outer(y, x, "<"))))

3) tidyverse or to go full tidyverse

library(dplyr)
library(tidyr)

iris %>%
  select(Sepal.Length, Species) %>%
  expand_grid(a = ., b = .) %>%
  unnest(c(a, b), names_sep = ".") %>%
  summarize(value = sum(a.Sepal.Length < b.Sepal.Length), 
            .by = ends_with("Species")) %>%
  pivot_wider(id_cols = b.Species, names_from = a.Species) %>%
  rename(Species = b.Species)

##   # A tibble: 3 × 4
##   Species    setosa versicolor virginica
##   <fct>       <int>      <int>     <int>
## 1 setosa       1128        139        34
## 2 versicolor   2302       1169       479
## 3 virginica    2457       1927      1165

4) tapply Another base R approach is via tapply like this:

with(iris, tapply(kronecker(Sepal.Length, Sepal.Length, "<"), 
  expand.grid(Species = Species, Species = Species), sum))
##             Species
## Species      setosa versicolor virginica
##   setosa       1128        139        34
##   versicolor   2302       1169       479
##   virginica    2457       1927      1165