summarise_all applies functions to themselves

[flying-sheep]

> dplyr::summarise_all(data.frame(wat = letters), function(col) sum(!is.na(col)))
  wat
1   1
Warning message:
In is.na(col) : is.na() applied to non-(list or vector) of type 'closure'
> dplyr::summarise_all(data.frame(wat = letters), dplyr::funs(function(col) sum(!is.na(col))))
Error in summarise_impl(.data, dots) : 
  Column `wat` is of unsupported type function
Calls: <Anonymous> ... summarise -> summarise.tbl_df -> summarise_impl -> .Call

[JohnMount]

I also see errors. But, I see a different pattern of what works and does not work (following the 3 ways to specify a function mentioned in help(funs): by name, the function, or an expression over "."). Notice the if you store your function in a variable and then refer to the variable, that works (possible work-around, but according to the documentation you should be able to define the function inline as you did).

suppressPackageStartupMessages(library("dplyr"))

dplyr::summarise_all(data.frame(wat = letters), 
                     sum(!is.na(.)))
#> Error in inherits(x, "fun_list"): object '.' not found

f <- function(col) sum(!is.na(col))
dplyr::summarise_all(data.frame(wat = letters), 
                     f)
#>   wat
#> 1  26

f <- function(col) sum(!is.na(col))
dplyr::summarise_all(data.frame(wat = letters), 
                    function(col) sum(!is.na(col)))
#>   wat
#> 1  26


dplyr::summarise_all(data.frame(wat = letters), 
                     dplyr::funs(sum(!is.na(.))))
#>   wat
#> 1  26

f <- function(col) sum(!is.na(col))
dplyr::summarise_all(data.frame(wat = letters), 
                     dplyr::funs(f))
#>   wat
#> 1  26

f <- function(col) sum(!is.na(col))
dplyr::summarise_all(data.frame(wat = letters), 
                     dplyr::funs(function(col) sum(!is.na(col))))
#> Error in summarise_impl(.data, dots): Column `wat` is of unsupported type function

packageVersion("dplyr")
#> [1] '0.7.3'

[lionel-]

@flying-sheep Could you try again on master please? I can't reproduce the issue.

@JohnMount It looks like all these cases work as expected.

[flying-sheep]

With master I now get:

> dplyr::summarise_all(data.frame(wat = letters), function(col) sum(!is.na(col)))
  wat
1  26
> dplyr::summarise_all(data.frame(wat = letters), dplyr::funs(function(col) sum(!is.na(col))))
Fehler in summarise_impl(.data, dots) : 
  Column `wat` is of unsupported type function

so at least the first one was eliminated!

[lionel-]

I don't think the second one is supposed to work. I guess we could special-case function. What do you think @hadley?

[flying-sheep]

the docs say:

     ...: A list of functions specified by:
            • Their name, ‘"mean"’
            • The function itself, ‘mean’
            • A call to the function with ‘.’ as a dummy argument,
              ‘mean(., na.rm = TRUE)’

the second one says functions are accepted.

[lionel-]

yes but funs() is a quoting function so you're not supplying a function but just an expression. The second bullet should be "The function name".

[hadley]

Agreed. We should clarify the docs for this case. Maybe:

A list of functions specified by:

• Their name as a string, "mean"
• The bare function function name, e.g. mean
• A function call using . as a pronoun to refer to the input data, e.g. mean(., na.rm = TRUE)

[krlmlr]

It feels inconsistent that funs(mean) works but funs(function(x) mean(x)) or even funs(mean = function(x) mean(x)) doesn't. For now let's update the docs and explicitly mention that passing an anonymous function won't work.

[lionel-]

If we could go back in time we'd disallow funs(mean) I think.

[krlmlr]

Reference: #3368 (comment).

[hadley]

If we could do back in time, I think we'd disallow funs(mean(., na.rm = TRUE)) and require that all argument be functions (or the formula shortcut)

[lionel-]

We wouldn't need funs() at all then.