R devel - Jun 2003 - suggestion for make.names

I would like to suggest a modification to the make.names() function.
The current implementation has two problems: 
1. It doesn't check if a name matches an R keyword (like
"function").
2. The uniqueness algorithm is not invariant to concatenation.
   In other words, 
   make.names(c("a","a","a"),unique=T) != 
  
make.names(c(make.names(c("a","a"),unique=T),"a"),unique=T)

The first problem means that you can construct a data frame for which 
there is no valid formula:
  lm(if~then, data.frame("if"=3,"then"=4))

The second problem means that you get funny row names when you build
up a data frame by concatenation:
  rbind(data.frame(x=1),data.frame(x=2),data.frame(x=3)) ! 
rbind(rbind(data.frame(x=1),data.frame(x=2)),data.frame(x=3))

I'm providing a new implementation (and documentation) of make.names which
fixes these problems.  The uniqueness part is handled by make.unique, a
useful function in its own right.  For example, the following code in
rbind.data.frame:
    while(any(xj <- duplicated(rlabs)))
    rlabs[xj] <- paste(rlabs[xj], 1:sum(xj), sep = "")
could be replaced by:
    rlabs = make.unique(rlabs)

Another way to fix the first problem, by the way, would be to allow
quoting in formulas, e.g. lm("if" ~ then, ...).  Currently, when you
do this it gives an error.  This could eliminate the need for
make.names in many cases, by being able to quote names like 
"New Jersey" or "log(x)".

Tom Minka
www.stat.cmu.edu/~minka/

Moved to R-devel, as it seems more appropriate to that audience and I 
think has a slightly more liberal policy on attachments.

You say
> I'm providing a new implementation (and documentation) of make.names 
where, please?


On Tue, 17 Jun 2003, Tom Minka wrote:
> I would like to suggest a modification to the make.names() function.
> The current implementation has two problems: 
> 1. It doesn't check if a name matches an R keyword (like
"function").
> 2. The uniqueness algorithm is not invariant to concatenation.
>    In other words, 
>    make.names(c("a","a","a"),unique=T) != 
>   
make.names(c(make.names(c("a","a"),unique=T),"a"),unique=T)
> 
> The first problem means that you can construct a data frame for which 
> there is no valid formula:
>   lm(if~then, data.frame("if"=3,"then"=4))
> 
> The second problem means that you get funny row names when you build
> up a data frame by concatenation:
>   rbind(data.frame(x=1),data.frame(x=2),data.frame(x=3)) !>  
rbind(rbind(data.frame(x=1),data.frame(x=2)),data.frame(x=3))
> 
> I'm providing a new implementation (and documentation) of make.names
which
> fixes these problems.  The uniqueness part is handled by make.unique, a
> useful function in its own right.  For example, the following code in
> rbind.data.frame:
>     while(any(xj <- duplicated(rlabs)))
>     rlabs[xj] <- paste(rlabs[xj], 1:sum(xj), sep = "")
> could be replaced by:
>     rlabs = make.unique(rlabs)
> 
> Another way to fix the first problem, by the way, would be to allow
> quoting in formulas, e.g. lm("if" ~ then, ...).  Currently, when
you
> do this it gives an error.  This could eliminate the need for
> make.names in many cases, by being able to quote names like 
> "New Jersey" or "log(x)".
Yes, we know and it is planned to allow that.  However, R is a volunteer 
project and that looks far from trivial and rather low priority.  Now if 
you have a patch for that we would be pleased to see it.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595