R all.equal中可能存在的错误

Question

I have faced some strange behavior in the R's all.equal function. Basically, I create two same data.frames differently and then call the all.equal function (checking data and attributes as well).

The code to reproduce the behavior is as follows:

var.a <- data.frame(cbind(as.integer(c(1,5,9)), as.integer(c(1,5,9))))
colnames(var.a) <- c("C1", "C2")
rownames(var.a) <- c("1","5","9")

var.b <- data.frame(matrix(NA, nrow = 10, ncol = 2))
var.b[, 1] <- 1:10
var.b[, 2] <- 1:10
colnames(var.b) <- c("C1", "C2")
var.b <- var.b[seq(1, nrow(var.b), 4), ]

all.equal(var.a, var.b)

Is this a bug or am I just missing something? I did quite some debugging of the all.equall function and it appears the problem is the the rownames of the data.frames (once them being a character the other time a numeric vector). The response of the all.equall function:

[1] "Attributes: < Component 2: Modes: character, numeric >"
[2] "Attributes: < Component 2: target is character, current is numeric >"

However,

typeof(rownames(var.a)) == typeof(rownames(var.b))

returns TRUE , which confuses me.

PS: The structure of the objects seems the same:

> str(var.a)
'data.frame':   3 obs. of  2 variables:
$ C1: int  1 5 9
$ C2: int  1 5 9
> str(var.b)
'data.frame':   3 obs. of  2 variables:
$ C1: int  1 5 9
$ C2: int  1 5 9

I would appreciate if someone could shed some light on this.

Answer 1

(I'm not exactly clear what bug you are thinking you have found. The data frames were not created the same way.) There are two differences in the structures of var.a and var.b: The mode of the elements in the columns: numeric in 'var.a' and integer in 'var.b'; and the mode of the rownames: integer for 'var.a' and character in 'var.b':

> dput(var.b)
structure(list(C1 = c(1L, 5L, 9L), C2 = c(1L, 5L, 9L)), .Names = c("C1", 
"C2"), row.names = c(1L, 5L, 9L), class = "data.frame")
> dput(var.a)
structure(list(C1 = c(1, 5, 9), C2 = c(1, 5, 9)), .Names = c("C1", 
"C2"), row.names = c("1", "5", "9"), class = "data.frame")

> mode(attr(var.b, "row.names"))
[1] "numeric"
> storage.mode(attr(var.b, "row.names"))
[1] "integer"
> mode(attr(var.a, "row.names"))
[1] "character"

Added note: If you wanted to check for numerical equality you should use the 'check.attributes' switch:

> all.equal(var.a, var.b, check.attributes=FALSE)
[1] TRUE

If you look at var.b with dput , you can see that the rownames are numeric:

> dput(var.b)
structure(list(C1 = c(1L, 5L, 9L), C2 = c(1L, 5L, 9L)), .Names = c("C1", 
"C2"), row.names = c(1L, 5L, 9L), class = "data.frame")

Answer 2

However,

typeof(rownames(var.a)) == typeof(rownames(var.b))

returns TRUE, which confuses me.

In addition to the most voted answer, note that the attributes are stored as "character" for var.a and as "numeric" for var.b :

> attr(var.a, "row.names")
[1] "1" "5" "9"
> attr(var.b, "row.names")
[1] 1 5 9

Whereas the rownames() function will coerce its output value to "character" :

> rownames(var.a)
[1] "1" "5" "9"
> rownames(var.b)
[1] "1" "5" "9"

This is why you get TRUE in the command above. As per ?rownames :

For a data frame, value for rownames should be a character vector of non-duplicated and non-missing names (this is enforced), and for colnames a character vector of (preferably) unique syntactically-valid names. In both cases, value will be coerced by as.character, and setting colnames will convert the row names to character.

A more pertinent check would be:

> typeof(attr(var.a, "row.names")) == typeof(attr(var.b, "row.names"))
[1] FALSE

This said, I believe that all.equal() messages are cryptic at best...

Answer 3

One is of mode numeric and the other is of mode integer. You can see this with:

str(var.a); str(var.b)


> str(var.a); str(var.b)
'data.frame':   3 obs. of  2 variables:
 $ C1: num  1 5 9
 $ C2: num  1 5 9
'data.frame':   3 obs. of  2 variables:
 $ C1: int  1 5 9
 $ C2: int  1 5 9