给定任何类型的向量，如何获得相同 class 的 NA 值？

Question

Suppose I want to do something like:

mask_values <- function(x, mask) ifelse(mask, x, NA)

The purpose of this function is to take a vector and replace some of its values with NA based on the value of mask . However, this function doesn't guarantee that the return type is always the same as the input x . For example:

date_vec <- rep(lubridate::today(), 10)
my_mask <- rep(c(TRUE, FALSE), length.out = 10)
class(mask_values(date_vec, my_mask))

which yields "numeric" rather than the desired "Date" . So I try switching to dplyr::if_else , which is supposed to preserve types:

mask_values <- function(x, mask) dplyr::if_else(mask, x, NA)
class(mask_values(date_vec, my_mask))

However, if_else also requires the input types to be the same as each other, and NA has type "logical", which means I get this error:

Error: `false` must be a `Date` object, not a logical vector.

So it seems that if I want to use if_else in order to preserve the input type, I need to be able to obtain an NA value with the same class as the input. Is there a reliable way to do this for any class? One possibility seems to be x[NA] , but I'm not sure if that is a universal solution or if it just happens to work with the examples that I've tested. You can assume that the only classes that matter are "vector-like" classes for which NA values exist, such as Date and POSIXct, as well as all the basic R data types (logical, character, numeric, etc.).

Alternatively, is there another way to implement my mask_values function such that the return value always has the same type as x ?

Answer 1

I recommend avoiding ifelse whenever possible. It is quite inefficient and as you have seen also quirky regarding what it returns (although that is well documented). I rarely use it and, if I do use it, only for interactive use and not programmatically.

The canonical and safe way of setting values to NA in base R is is.na<- . (Note that it supports logical and positional indexing. mask could also be a numeric vector.)

mask_values <- function(x, mask) {
  is.na(x) <- mask
  x
}

#or simply this:
#mask_values <- `is.na<-` 
#i.e., `is.na<-` is already what you want.

class(mask_values(date_vec, my_mask))
#[1] "Date"

Alternatively, you can also use simple subset-assignment. NA is a logical value. (If you create it like this. It can be coerced to other types and of course you can specify it as other types with NA_real_ etc.) If you assign a logical vector into any other vector, it will be coerced to that other vector's type (because "logical" is the most primitive type).

mask_values <- function(x, mask) {
  x[mask] <- NA
  x
}

class(mask_values(date_vec, my_mask))
#[1] "Date"

Btw., this subset-assignment is how the is.na<-.default method is defined.

I prefer doing subset-assignment explicitly in my code but occasionally the convenience function replace can be useful.