Do tidyr::crossing works with curly-curly within a function?

Question

f <- function(db, entity, trait) {
  
  db %>%
    summarise(i = crossing({{ trait }},{{ trait }}),
              .by = {{ entity }}) 
}

db <- data.frame(
  entity = c("A", "A", "B", "B"),
  beta = c("X", "Y", "X", "Z")
)

f(db,entity,beta)

Is there a way to make this work? I have the impression that tidyr does not work very well with functional programming in general. Should I resort to expand_grid?

Answer 1

Option 1: dplyr::cross_join

You can achieve it entirely in the dplyr scope.

db <- data.frame(
  entity = c("A", "A", "B", "B"),
  beta = c("X", "Y", "X", "Z")
)

library(dplyr)

f1 <- function(db, entity, trait) {
  db %>%
    reframe(cross_join(pick({{ trait }}), pick({{ trait }})),
            .by = {{ entity }}) 
}

f1(db, entity, beta)

#   entity beta.x beta.y
# 1      A      X      X
# 2      A      X      Y
# 3      A      Y      X
# 4      A      Y      Y
# 5      B      X      X
# 6      B      X      Z
# 7      B      Z      X
# 8      B      Z      Z

---

Option 2: base::expand.grid

You can also use base::expand.grid(). Note that you cannot replace expand.grid() with tidyr::expand_grid() here.

f2 <- function(db, entity, trait) {
  db %>%
    reframe(expand.grid({{ trait }}, {{ trait }}),
            .by = {{ entity }})
}

f2(db, entity, beta)

#   entity Var1 Var2
# 1      A    X    X
# 2      A    Y    X
# 3      A    X    Y
# 4      A    Y    Y
# 5      B    X    X
# 6      B    Z    X
# 7      B    X    Z
# 8      B    Z    Z

---

Option 3: tidyr::expand

To involve tidyr, you should use expand() instead of expand_grid() and crossing(). Note that expand() will de-duplicate and sort its inputs.

f3 <- function(db, entity, trait) {
  db %>%
    group_by({{ entity }}) %>%
    tidyr::expand({{ trait }}, {{ trait }}) %>%
    ungroup()
}

f3(db, entity, beta)

# # A tibble: 8 × 3
#   entity beta...1 beta...2
#   <chr>  <chr>    <chr>   
# 1 A      X        X       
# 2 A      X        Y       
# 3 A      Y        X       
# 4 A      Y        Y       
# 5 B      X        X       
# 6 B      X        Z       
# 7 B      Z        X       
# 8 B      Z        Z

---

Benchmark on a larger dataset

Efficiency: base::expand.grid > dplyr::cross_join > tidyr::expand

db_large <- data.frame(
  entity = rep(LETTERS[1:3], each = 26),
  beta = rep(LETTERS, 3)
)

bench::mark(
  `dplyr::cross_join` = f1(db_large, entity, beta),
  `base::expand.grid` = f2(db_large, entity, beta),
  `tidyr::expand` = f3(db_large, entity, beta),
  check = FALSE
)

# # A tibble: 3 × 13
#   expression           min  median `itr/sec` mem_alloc `gc/sec` n_itr  n_gc total_time
#   <bch:expr>       <bch:t> <bch:t>     <dbl> <bch:byt>    <dbl> <int> <dbl>   <bch:tm>
# 1 dplyr::cross_jo…  8.14ms  8.42ms     117.      123KB     5.31    44     2      377ms
# 2 base::expand.gr…  3.19ms  3.63ms     272.      128KB     6.43   127     3      467ms
# 3 tidyr::expand    11.46ms 11.71ms      83.7     157KB     6.79    37     3      442ms