Wrong results in "summarise" when characters are used as variable identifiers

[patrickroocks]

The following code

summarise(mtcars, n_distinct("mpg"))

doesn't rise any errors or warning but returns (randomly) wrong results (but approximately the same magnitude) between 16 and 24, whereas the correct use of the variable.

summarise(mtcars, n_distinct(mpg))

returns 25.

Indicated by the stackoverflow question "Randomness in dplyr" a similar behavior occurs on larger datasets too.

[romainfrancois]

The problem is there somewhere:https://github.com/hadley/dplyr/blob/424b304ae235d9ff33b6fbcd7860e8f94ec7e9f2/inst/include/dplyr/Result/Count_Distinct.h

@hadley : Should I make n_distinct("mpg" ):

• illegal
• return 1

By the same token, would we allow expressions inside of n_distinct, e.g. n_distinct(mpg*2)

[hadley]

Ideally n_distinct("mpg") would return 1 and expressions would also work.

[romainfrancois]

Sure. Just wanted to check first, forbid stuff is easier to do.

[romainfrancois]

Ah. What about this:

z <- 1:2
df <- data.frame( x = c(1,1,1, 2,2,2) )
summarise( group_by(df,x), n_distinct(z) )

z comes from outside the data frame. Currently I get bogus results like this:

Source: local data frame [2 x 2]

  x n_distinct(z)
1 1             2
2 2             3

[romainfrancois]

This is because the underlying class Count_Distinct expects it is given something to visit that is as long as there are rows in the data, which is enough of an assumption when we know n_distinct only manages symbols.

I'm not trying to push towards forbidding things, but it would be easier to name whatever expression that would go inside n_distinct in some mutate call before and only accept symbols in n_distinct.

Also n_distinct("mpg") seems like a time bomb, probably not what was expected from the SO question, making it an hard error at least gives a better clue that it's wrong ?

[hadley]

Ok, that's reasonable - can you just make Count_Distinct throw an error message like "Input to n_distinct() must be a single variable name` ?

[romainfrancois]

Done. Only thing allowed is a variable name that must be from the data frame.