Skip to content

Question on substitute.

2 messages · Simon Knapp, Uwe Ligges

Simon Knapp
# 
I have a data frame of around 4500 rows which contain file name
prefixes and other interesting variables. each prefix corresponds to
multiple files and a I only have 4300 complete sets of files. If any
one of the files is missing for a given prefix, then the function I
use fails. The effect is emulated by the function "f" and data
"dummy.data":

f <- function(x) if(runif(1) < 0.1) stop('oops') else rnorm(10)
dummy.data <- data.frame(cbind(paste('a', 1:100, sep=''),
matrix(rnorm(9000),nrow=100, ncol=9)))

I would like the results as a data.frame and if it weren't for the
errors would do:

res <- apply(dummy.data, 1, function(x) try(f(x))) #... though I
wouldn't really need the try would I.
#dimnames(res)[[2]] <- as.character(dummy.data[,1])

but the errors make res a list and the second line fails. I can do:

res <- NULL
for(i in 1:nrow(dummy.data)) if(class(tmp <- try(f(dummy.data[i,])))
!= 'try-error') res <- cbind(res, tmp)

which gives me the correct result, but I don't know which column of
res belongs to which row of dummy.data.

I can get the desired result with with:

#--- desired result --------
res <- NULL
for(i in 1:nrow(dummy.data)) if(class(tmp <- try(f(dummy.data[i,])))
!= 'try-error')
    eval(parse(text=paste("res <- cbind(res,", dummy.data[i,1],
"=tmp)", sep='')))
#------------------------------

but it is sort of ugly and I was hoping I could do something like:

res <- NULL
for(i in 1:nrow(dummy.data)) if(class(tmp <- try(f(dummy.data[i,])))
!= 'try-error')
    res <- eval(substitute(cbind(res, a=b), list(a=dummy.data[i,1], b=tmp)))

but the a is not being substituted (not that this is prettier, but I'm
trying to come to grips with
substitute/do.call/calls/expressions/...). I guess that argument names
are a different beast to argument values in expressions.



My question (finally) is: is there a way of achieving 'desired
result', but using 'quote', 'substitute' or any other similar
functions/idioms I've not run into yet? Alternatively, is there a
completely different way?

Any help appreciated,
Simon Knapp.
Uwe Ligges
#
Simon Knapp wrote:
So continue with, e.g.:

  valid <- !sapply(res, inherits, "try-error")
  res <- res[valid]
  res <- matrix(unlist(res), ncol=length(res))
  colnames(res) <- as.character(dummy.data[valid,1])

I would not try any of your further questions, they result into slower 
and more unreadable code.

Uwe Ligges