NaN is removed when using na.rm=TRUE

Question

This reproducible example is a very simplified version of my code:

x <- c(NaN, 2, 3)

#This is fine, as expected
max(x)
> NaN

#Why does na.rm remove NaN?
max(x, na.rm=TRUE) 
> 3

To me, NA (missing value) and NaN (not a number) are two completely different entities, why does na.rm remove NaN? How can I ignore NA and not NaN?

ps:I am using 64-bit R version 3.0.0 on Windows7.

Edit: Upon some more study I found that is.na returns true for NaN too! This is the cause of confusion for me.

is.na(NaN)
> TRUE

score 8 · Accepted Answer · answered Apr 16 '13 at 03:51

8

It's a language decision:

> is.na(NaN)
[1] TRUE

is.nan differentiates:

> is.nan(NaN)
[1] TRUE
> is.nan(NA)
[1] FALSE

So you may need to call both.

answered Apr 16 '13 at 03:51

Matthew Lundberg

42,009
6
90
112

Yes, I just noticed this behaviour. My idea of missing value needs to change. – Nishanth Apr 16 '13 at 03:55
@e4e5f4 I probably would not have chosen this behavior, were I designing the language. But it is what it is. – Matthew Lundberg Apr 16 '13 at 03:56
1

It is interesting to note that the help for `sum` states that `na.rm = TRUE` will remove both `NaN` and `NA` (whereas `?Extremes`) does not. – mnel Apr 16 '13 at 04:02
3

Also of interest: class(NA) == "logical"; class(NaN) == "numeric". – neilfws Apr 16 '13 at 04:12
@neilfws That's a convenience, and another language decision. Try `class(NA_real_)`. – Matthew Lundberg Apr 16 '13 at 04:14

score 3 · Answer 2 · answered Apr 16 '13 at 03:53

na.rm arguments in functions generally use is.na() or an analogous function.
And since is.na(NaN) == TRUE, you then get the behavior you're observing.

Now should NaN be treated as also NA? That is a different question ;)

The best way around this is to explicitly tell R how to handle NaN One example:

ifelse(any(is.nan(x)), NaN, min(x, na.rm=TRUE))

NaN is removed when using na.rm=TRUE

2 Answers2

Linked