Suprising behaviour of if_any() inside case_when

[iagogv3]

I am not sure if I am doing something wrong, but If that is the case I cannot see what in next reproducible example:

data <- structure(list(x = c(1, 1, -1, -1, -1, -1, 1, -1, -1, -1, 1, 
-1, -1, -1, -1, 1, -1, 1, -1, -1, -1, 1, -1, 1, -1, -1, -1, -1, 
-1, -1), y = c(-1, -1, 1, 1, 2, 1, -1, 1, 1, 1, -1, 1, 1, 1, 
1, -1, 1, -1, 1, 1, 2, -1, 1, -1, 1, 1, 1, 1, 1, 1)), row.names = c(NA, 
-30L), class = c("tbl_df", "tbl", "data.frame"))

data %>%
    mutate(dich = case_when(if_any(c(x, y), function(.x) {.x == 1}) ~ "green",
                             x==2 | y==2 ~ "blue",
                             TRUE ~ NA_character_)) %>%
    pull(dich) %>% table(useNA = "ifany")
blue green 
    2    28 

# but

data %>%
    mutate(dich = case_when(if_any(c(x, y), function(.x) {.x == 1}) ~ "green",
                             if_any(c(x, y), function(.x) {.x == 2}) ~ "blue",
                             TRUE ~ NA_character_)) %>%
    pull(dich) %>% table(useNA = "ifany")
green  <NA> 
   28     2 

Thank you!

[hadley]

Can you please provide a minimal reprex (reproducible example)? The goal of a reprex is to make it as easy as possible for me to recreate your problem so that I can fix it: please help me help you!

If you've never heard of a reprex before, start by reading "What is a reprex", and follow the advice further down the page. Please make sure your reprex is created with the reprex package as it gives nicely formatted output and avoids a number of common pitfalls.

[iagogv3]

@hadley Here a reprex:

library(dplyr)
iris %>% 
  head() %>%
  mutate(G.Width = case_when(if_any(ends_with("Width"), ~ . > 3.5) ~ "Big", 
                             if_any(ends_with("Width"), ~ . > 3.1) ~ "Medium",
                             TRUE ~ "Small"))
#>   Sepal.Length Sepal.Width Petal.Length Petal.Width Species G.Width
#> 1          5.1         3.5          1.4         0.2  setosa   Small
#> 2          4.9         3.0          1.4         0.2  setosa   Small
#> 3          4.7         3.2          1.3         0.2  setosa   Small
#> 4          4.6         3.1          1.5         0.2  setosa   Small
#> 5          5.0         3.6          1.4         0.2  setosa     Big
#> 6          5.4         3.9          1.7         0.4  setosa     Big

^{Created on 2021-03-03 by the reprex package (v1.0.0)}

[hadley]

Can you please make a smaller reprex? Is it necessary to use 150 rows to illustrate the problem?

[iagogv3]

@hadley Reprex updated

[hadley]

Simpler reprex:

library(dplyr, warn.conflicts = FALSE)

df <- data.frame(
  x = 1:3
)
df %>% mutate(
  y1 = case_when(
    if_any(everything(), ~ . >= 3) ~ "big",
    if_any(everything(), ~ . >= 2) ~ "medium",
    TRUE ~ "small"
  ),
  y2 =  case_when(
    x >= 3 ~ "big",
    x >= 2 ~ "medium",
    TRUE ~ "small"
  )
)
#>   x    y1     y2
#> 1 1 small  small
#> 2 2 small medium
#> 3 3   big    big

^{Created on 2021-03-03 by the reprex package (v1.0.0)}

[hadley]

Even simpler reprex:

library(dplyr, warn.conflicts = FALSE)

df <- data.frame(
  x = 1:3
)

df %>% mutate(
  y3 = if_any(x, ~ . >= 3),
  y2 = if_any(x, ~ . >= 2),
  
  data.frame(
    z3 = if_any(x, ~ . >= 3),
    z2 = if_any(x, ~ . >= 2)
  )
) %>% 
  select(y2, z2)
#>      y2    z2
#> 1 FALSE FALSE
#> 2  TRUE FALSE
#> 3  TRUE  TRUE

^{Created on 2021-03-03 by the reprex package (v1.0.0)}

[romainfrancois]

It has to do with the across() caching mechanism that is in place so that the tidy selection is not evaluated each time.

library(dplyr, warn.conflicts = FALSE)

df <- data.frame(
  x = 1:3
)

trace(across, tracer = quote({ print(key); print(setup$fns) }), at = 4)
#> Tracing function "across" in package "dplyr"
#> [1] "across"
df %>% mutate(
  y1 = case_when(
    if_any(everything(), ~ . >= 3) ~ "big",
    if_any(everything(), ~ . >= 2) ~ "medium",
    TRUE ~ "small"
  )
)
#> Tracing across({ .... step 4 
#> [1] "across({"
#> $`1`
#> <lambda>
#> function (..., .x = ..1, .y = ..2, . = ..1) 
#> . >= 3
#> <environment: 0x7fbbf89bc2e8>
#> attr(,"class")
#> [1] "rlang_lambda_function" "function"             
#> 
#> Tracing across({ .... step 4 
#> [1] "across({"
#> $`1`
#> <lambda>
#> function (..., .x = ..1, .y = ..2, . = ..1) 
#> . >= 3
#> <environment: 0x7fbbf89bc2e8>
#> attr(,"class")
#> [1] "rlang_lambda_function" "function"
#>   x    y1
#> 1 1 small
#> 2 2 small
#> 3 3   big

^{Created on 2021-03-04 by the reprex package (v0.3.0)}

if_any() is currently implemented as:

if_any <- function(.cols = everything(), .fns = NULL, ..., .names = NULL) {
  df <- across({{ .cols }}, .fns = .fns, ..., .names = .names)
  n <- nrow(df)
  df <- vec_cast_common(!!!df, .to = logical())
  .Call(dplyr_reduce_lgl_or, df, n)
}

So the call to across() is always across({{ .cols }}, .fns = .fns, ..., .names = .names) which gives "across({" as the key.

I'll fix if_any() and if_all() right now, however this probably needs some more thinking around the caching because e.g.

library(dplyr, warn.conflicts = FALSE)
times <- function(cols, multiplier = 1) {
  across({{ cols }}, function(x) {
    print(multiplier)
    x * multiplier
  })
}
data.frame(x = 1) %>% 
  summarise(
    times(x, 1)$x + times(x, 2)$x
  )
#> [1] 1
#> [1] 1                                                # <-------- should be 2
#>   times(x, 1)$x + times(x, 2)$x
#> 1                             2

^{Created on 2021-03-04 by the reprex package (v0.3.0)}