Ellipsis Trouble: Passing ... to lm - r

Question

I am building a wrapper around lm to do some additional calculations. I'd like the wrapper to pass ... to lm, but I am getting into trouble with lm's weights argument.
LmWrapper <- function(df, fmla, ...) {
est <- lm(fmla, df, ...)
list(model = est)
}
If I call the wrapper with a weights argument,
data(airquality)
LmWrapper(airquality, Ozone ~ Wind, weights = Temp)
R does not know where to look for the weights:
Error in eval(expr, envir, enclos) :
..1 used in an incorrect context, no ... to look in
The lm help page says
All of weights, subset and offset are evaluated in the same way as variables in formula, that is first in data and then in the environment of formula.
but the wrapper seems to change things.
How do I fix this?
The traceback() for the above error looks like this:
8: eval(expr, envir, enclos)
7: eval(extras, data, env)
6: model.frame.default(formula = fmla, data = df, weights = ..1,
drop.unused.levels = TRUE)
5: stats::model.frame(formula = fmla, data = df, weights = ..1,
drop.unused.levels = TRUE)
4: eval(expr, envir, enclos)
3: eval(mf, parent.frame())
2: lm(fmla, df, ...) at #2
1: LmWrapper(diamonds, price ~ carat, weights = depth)
Calling lm directly, works just fine:
lm(Ozone ~ Wind, airquality, weights = Temp)

Answer 1

So the problem is that lm normally looks up those names in argument data but somehow scoping goes wrong. You can fix that by looking up column references and passing them on manually.
LmWrapper <- function(df, fmla, ...) {
# get names of stuff in ...
argNames = sapply(substitute(list(...))[-1L], deparse)
# look for identical names in df
m = match(names(df), argNames, 0L)
# store other arguments from ... in a list
args = list(eval(parse(text = argNames[-m])))
# name the list
names(args) = names(argNames[-m])
# store complete values in args, instead of just references to columns
# the unlist code is rather ugly, the goal is to create a list where every
# element is a column of interest
args[names(argNames)[m]] = unlist(apply(df[, as.logical(m), drop = FALSE],
2, list), recursive = FALSE)
# also put other stuff in there
args$formula = fmla
args$data = df
# do lm
est = do.call(lm, args)
list(model = est)
}
data(airquality)
airquality$subset = airquality$Solar.R > 200
LmWrapper(airquality, Ozone ~ Wind, weights = Temp, subset = subset,
method = 'qr')
The code above is not the most beautiful, but it works for both subset and weights. Alternatively, you could just handle weights and subset as exceptions.

Answer 2

Thanks for this answer #Vandenman. I just implemented it with two changes that I wanted to share, in case someone else comes across this thread:
1) If there are no arguments in ... apart from columns in the data, the code as above creates an NA element in the list, which throws a warning - I added a condition below to get around that.
2) The model object returned by the code above has a very very long call since it includes each weight etc. If you don't care about the ability to use update(), it might be more insightful to replace it by the actual function call to record what happened, as implemented below.
run_lm <- function(df, formula, ...) {
# get names of stuff in ...
argNames = sapply(substitute(list(...))[-1L], deparse)
# look for identical names in df
m = match(names(df), argNames, 0L)
# store other arguments from ... in a list, if any
dot_args <- eval(parse(text = argNames[-m]))
if (is.null(dot_args)) {args <- list()
} else {
args <- list(dot_args)
# name the list
names(args) = names(argNames[-m])
}
# store complete values in args, instead of just references to columns
# the unlist code is rather ugly, the goal is to create a list where every
# element is a column of interest
args[names(argNames)[m]] = unlist(apply(df[, as.logical(m), drop = FALSE],
2, list), recursive = FALSE)
# also put other stuff in there
args$formula = formula
args$data = df
# do lm
mod <- do.call(lm, args)
mod$call <- sys.call()
mod
}