dplyr：为什么有些操作在不调用 rowwise() 的情况下“按行”工作，而其他操作却不行？

Question

I am still trying to figure out, how rowwise works exactly in R/dplyr.

For example I have this code:

library(dplyr)
df = data.frame(
  group = c("a", "a", "a", "b", "b", "c"),
  var1 = 1:6,
  var2 = 7:12
)

df %>%
  mutate(
    concatNotRW = paste0(var1, "-", group), # work on rows
    meanNotRW = mean(c(var1, var2)), # works not on rows
    charsNotRW = strsplit(concatNotRW, "-") # works on rows
  ) %>%
  rowwise() %>%
  mutate(
    concatRW = paste0(var1, "-", group), # all work on rows
    meanRW = mean(c(var1, var2)),
    charsRW = strsplit(concatRW, "-")
  ) -> res

The res dataframe looks like this:

  group  var1  var2 concatNotRW meanNotRW charsNotRW concatRW meanRW chars    
  <chr> <int> <int> <chr>           <dbl> <list>     <chr>     <dbl> <list>   
1 a         1     7 1-a               6.5 <chr [2]>  1-a           4 <chr [2]>
2 a         2     8 2-a               6.5 <chr [2]>  2-a           5 <chr [2]>
3 a         3     9 3-a               6.5 <chr [2]>  3-a           6 <chr [2]>
4 b         4    10 4-b               6.5 <chr [2]>  4-b           7 <chr [2]>
5 b         5    11 5-b               6.5 <chr [2]>  5-b           8 <chr [2]>
6 c         6    12 6-c               6.5 <chr [2]>  6-c           9 <chr [2]>

What I do not understand is why paste0 can take each cell of a row and pastes them together (essentially performing a rowwise-operation), yet mean can't do that. What am I missing and are there any rules on what already works rowwise without the call to rowwise() ? I did not find so much info in the rowwise()-vi.nette here https://dplyr.tidyverse.org/articles/rowwise.html

Answer 1

paste can take vectors as input in the variadic argument ( ... ) and return the same length as vector whereas mean takes the variadic argument for other inputs ( trim etc) and return a single value. Here we need rowMeans . Regarding strsplit , it returns a list of split elements

library(dplyr)
df %>%
  mutate(
    concatNotRW = paste0(var1, "-", group),
    meanNotRW = rowMeans(across(c(var1, var2))),
    charsNotRW = strsplit(concatNotRW, "-") 
  )

---

> mean(c(1:5, 6:10))
[1] 5.5

Note that the vector we are passing is a single vector by c oncatenating both vectors 1:5 and 6:10

whereas

> paste(1:5, 6:10)
[1] "1 6"  "2 7"  "3 8"  "4 9"  "5 10"

are two vectors passed into paste

---

For splitting the column into two columns, we can use separate

library(tidyr)
 df %>%
  mutate(
    concatNotRW = paste0(var1, "-", group),
    meanNotRW = rowMeans(across(c(var1, var2)))) %>% 
    separate(concatNotRW, into = c("ind", "chars"))
 group var1 var2 ind chars meanNotRW
1     a    1    7   1     a         4
2     a    2    8   2     a         5
3     a    3    9   3     a         6
4     b    4   10   4     b         7
5     b    5   11   5     b         8
6     c    6   12   6     c         9

---

Why some operations work on rowwise depends on the function. If the function is vectorized, it works on the whole column and doesn't need rowwise . Here, both functions paste and mean are vectorized except that paste is vectorized for variadic input and mean is only vectorized to take a single vector and return a single value as output. Suppose, we have a function that checks each value with if/else , then it is not vectorized as if/else expects a single logical value. In that case, can use either rowwise or Vectorize the function