[R] using names with functions..

Duncan Murdoch murdoch at stats.uwo.ca
Wed Nov 28 16:56:58 CET 2007

On 11/28/2007 9:52 AM, vito muggeo wrote:
> Dear all,
> I have the following (rather) strange problem..
> For some reasons, I finally work with a variable whose name includes an 
> R function, "a.log(z)", say. And that is a problem when I call it in a 
> formula, for instance:
>  > myname<-"a.log(z)"
>  > dd<-data.frame("a.log(z)"=1:10,y=rnorm(10))
>  > o<-lm(y~1,data=dd)
>  > fo<-as.formula(paste(".~.+",paste(myname, collapse = "+")))
>  > fo
> . ~ . + a.log(z)
>  > update(o,formula=fo)
> Error in eval(expr, envir, enclos) : could not find function "a.log"
>  >
> How can fit the model? namely how can I use "a.log(z)" in the example above?

You can make just about anything into a name by quoting in in backticks, 
e.g. `a.log(z)` should always be seen as a name.  However, there is a 
fair bit of code in contributed packages that converts things to 
character vectors and then reparses it (like your example above!), and 
that code may well get confused by such a name.

For example,

 > deparse(quote(`a.log(z)`))
[1] "a.log(z)"

The package writer could avoid this by using

 > deparse(quote(`a.log(z)`), backtick=T)
[1] "`a.log(z)`"

You also need to worry about the fact that by default, data.frame() will 
convert the name of the first column to a.log.z.

So you could get everything you want with this variation on your example:

 > myname<-"`a.log(z)`"
 > dd<-data.frame("a.log(z)"=1:10,y=rnorm(10), check.names=FALSE)
 > o<-lm(y~1,data=dd)
 > fo<-as.formula(paste(".~.+",paste(myname, collapse = "+")))
 > fo
. ~ . + `a.log(z)`
 > update(o,formula=fo)

lm(formula = y ~ `a.log(z)`, data = dd)

(Intercept)   `a.log(z)`
     -0.3931       0.0753

However, I'd suggest avoiding the use of such a name.

Duncan Murdoch

More information about the R-help mailing list