rbind on data.frame that contains a column that is also a data.frame

-----Original Message-----
From: r-devel-bounces at r-project.org
[mailto:r-devel-bounces at r-project.org] On Behalf Of Nicholas
L Crookston
Sent: Friday, August 06, 2010 8:35 AM
To: Michael Lachmann
Cc: r-devel-bounces at r-project.org; r-devel at r-project.org
Subject: Re: [Rd] rbind on data.frame that contains a column
that is also a data.frame

OK...I'll put in my 2 cents worth.

It seems to me that the problem is with this line:

b$a=a , where "s" is something other than a vector with
length equal to nrow(b).

I had no idea that a dataframe could hold a dataframe. It is not just
rbind(b,b) that fails, apply(b,1,sum) fails and so does plot(b). I'll
bet other R commands fail as well.

My point of view is that a dataframe is a list of vectors
of equal length and various types (this is not exactly what the help
page says, but it is what it suggests to me).

Hum, I wonder how much code is based on the idea that a
dataframe can hold
a dataframe.

I used to think that non-vectors in data.frames were
pretty rare things but when I started looking into
the details of the modelling code I discovered that
matrices in data.frames are common.  E.g.,

  > library(splines)
  > sapply(model.frame(data=mtcars, mpg~ns(hp)+poly(disp,2)), class)

  $mpg
  [1] "numeric"

  $`ns(hp)`
  [1] "ns"     "basis"  "matrix"

  $`poly(disp, 2)`
  [1] "poly"   "matrix"
You may not see these things because you don't call model.frame()
directly, but most modelling functions (e.g., lm() and glm())
do call it and use the grouping provided by the matrices to encode
how the columns of the design matrix are related to one another.

If matrices are allowed, shouldn't data.frames be allowed as well?

Bill Dunlap
Spotfire, TIBCO Software
wdunlap tibco.com

15 years of using R just isn't enough! But, I can
say that not
one
line of code I've written expects a dataframe to hold a dataframe.

Hi,

The following was already a topic on r-help, but after

understanding
what is

going on, I think it fits better in r-devel.

The problem is this:
When a data.frame has another data.frame in it, rbind

doesn't work well.

Here is an example:
--

a=data.frame(x=1:10,y=1:10)
b=data.frame(z=1:10)
b$a=a
b

z a.x a.y
1   1   1   1
2   2   2   2
3   3   3   3
4   4   4   4
5   5   5   5
6   6   6   6
7   7   7   7
8   8   8   8
9   9   9   9
10 10  10  10

rbind(b,b)

Error in `row.names<-.data.frame`(`*tmp*`, value = c("1",

"2", "3", "4",
 :

duplicate 'row.names' are not allowed
In addition: Warning message:
non-unique values when setting 'row.names': ?1?, ?10?, ?2?,

?3?, ?4?,
?5?,

?6?, ?7?, ?8?, ?9?
--

Looking at the code of rbind.data.frame, the error comes from the
lines:
--
xij <- xi[[j]]
if (has.dim[jj]) {
value[[jj]][ri, ] <- xij
rownames(value[[jj]])[ri] <- rownames(xij)   # <--  problem is here
}
--
if the rownames() line is dropped, all works well. What this line
tries to do is to join the rownames of internal elements of the
data.frames I try to rbind. So the result, in my case should have a
column 'a', whose rownames are the rownames of the original

column 'a'.
It

isn't totally clear to me why this is needed. When would a

data.frame

have different rownames on the inside vs. the outside?

Notice also that rbind takes into account whether the

rownames of the

data.frames to be joined are simply 1:n, or they are something else.
If they are 1:n, then the result will have rownames 1:(n+m). If not,
then the rownames might be kept.

I think, more consistent would be to replace the lines above with
something like:
if (has.dim[jj]) {
value[[jj]][ri, ] <- xij
rnj = rownames(value[[jj]])
rnj[ri] = rownames(xij)
rnj = make.unique(as.character(unlist(rnj)), sep = "")
rownames(value[[jj]]) <- rnj
}

In this case, the rownames of inside elements will also be

joined, but

in case they overlap, they will be made unique - just as

they are for

the overall result of rbind. A side effect here would be that the
rownames of matrices will also be made unique, which till now didn't
happen, and which also doesn't happen when one rbinds matrices that
have rownames. So it would be better to test above if we are dealing
with a matrix or a data.frame.

But most people don't have different rownames inside and outside.
Maybe it would be best to add a flag as to whether you care or don't
care about the rownames of internal data.frames...

But maybe data.frames aren't meant to contain other data.frames?

If instead I do
b=data.frame( z=1:10, a=a)
then rbind(b,b) works well. In this case the data.frame was

converted to
its

columns. Maybe
b$a = a
should do the same?

Michael
--
View this message in context: http://r.789695.n4.nabble.com/rbind-
on-data-frame-that-contains-a-column-that-is-also-a-data-frame-
tp2315682p2315682.html
Sent from the R devel mailing list archive at Nabble.com.

      [[alternative HTML version deleted]]