A friend of mine has been doing a lot of work analyzing hydrologic time series from a large number of stream gauges in California. She’s been doing all kinds of statistical analyses with various transformations and combinations of the dataset, which has inevitably led to working with nested lists of horrifying proportions. While nested lists are a pretty handy structure for programming, they lose their appeal pretty quickly once it’s time for a human to start accessing their contents.

She approached me yesterday with what sounds like a simple question:

“How can I loop through a nested list of dataframes and write each dataframe element to a directory structure that matches the list structure?”

In other words, if she has a list of three sublists, and each sublist contains two dataframes, she wants to end up with a folder that contains three folders, each of which contains two CSV files. For a normal list, this would be a trivial application of lapply and write.csv. For a nested list it gets trickier, but if you know the structure beforehand it’s just a matter of writing some nested loops. But what if you don’t know the list structure beforehand? And what if some of these (sub)lists also contain elements other than dataframes that you want to ignore?

My solution consists of three steps: first, I generate a flat list of element identifiers by sending a recursive function to lapply that only returns a value for elements in the nested list that are dataframes. By default, lapply uses the list names for the returned object; for a nested list I end up with names like “foo.A.one”, “foo.A.two”, “foo.B.one”, etc. Then, I convert the list of identifiers to a list of target filepaths by using gsub to replace “.” with “/”. I can then extract folder names from the list of filepaths using dirname and create the directory structure via dir.create. Finally, I convert the element identifiers to a list of expressions that evaluate write.csv for each list element, and use eval and lapply to write the files to the appropriate folders.

At the request of my friend, I wrapped this methodology into the poorly-named function magicCSV, with an additional argument that lets you specify the top-level directory to write the directory structure and files to.

magicCSV = function(startlist, targetdir){
ff = function(x){
if (class(x) == "list")
lapply(x, ff)
else if(class(x) == "data.frame")
TRUE
else
NULL
}
lnames = names(unlist(lapply(startlist, ff)))
fnames = file.path(file.path(targetdir), paste0(gsub(".", "/",
lnames, fixed = TRUE), ".csv"))
dirnames = unlist(lapply(fnames, dirname))
varnames = paste0("startlist$", gsub(".", "$", lnames,
fixed = TRUE))
evalstrings = paste0('write.csv(', varnames, ', file ="',
fnames, '")')
exprs = lapply(evalstrings, function(x) parse(text = gsub("\\",
"/", x, fixed = TRUE)))

suppressWarnings(lapply(dirnames, dir.create, recursive = TRUE))

invisible(lapply(exprs, eval, envir = environment()))
}


Note the use of file.path to ensure that it doesn’t matter if a user specifies a target directory as C:/foo or C:/foo/, and the weirdness of replacing \\ with / when creating the expressions so that eval doesn’t break (I don’t know why we have to do this). While it is ordinarily better to use Recall to call functions recursively, we can’t do that here because Recall doesn’t work when passed as a function argument (such as to lapply).

Here’s some code you can use to test out magicCSV:

# function to create an arbitrary dataframe
rdf = function(){
data.frame(replicate(sample(seq(2, 6), 1), rnorm(10)))
}
# make a nested list of dataframes for testing
test = list(
foo = list(
A = rdf(),
B = rdf(),
roo = factor(sample(LETTERS, 100, replace = TRUE))),
bar = list(
C = rdf(),
D = rdf()),
baz = list(
boo = list(
E = rdf(),
F = rdf()),
far = list(
gar = list(
G = rdf(),
H = rdf()),
gaz = list(
I = rdf(),
J = rdf()),
gab = factor(sample(LETTERS, 100, replace = TRUE))
)
)
)

# create a temporary directory
mydir = tempdir()
# test it out
magicCSV(test, mydir)
dir(mydir, recursive = TRUE)
read.csv(file.path(mydir, "bar", "C.csv"), row.names = 1)


And there you have it! It would be pretty easy to generalize this function further by e.g. passing a function to allow the user to create a custom eval statement, or an argument specifying what class to search the nested loop for. I’ll leave that to you.

EDIT

It turns out there’s a better way to do the above that avoids the messiness of eval and parse. The [[]] bracketing is actually way more powerful than I originally realized; you can actually access nested lists by using vectors! Below is a slightly rewritten version of magicCSV that takes advantage of this.

magicCSV = function(startlist, targetdir){
ff = function(x){
if (class(x) == "list")
lapply(x, ff)
else if(class(x) == "data.frame")
TRUE
else
NULL
}
lnames = names(unlist(lapply(startlist, ff)))

fnames = file.path(file.path(targetdir), paste0(gsub(".", "/",
lnames, fixed = TRUE), ".csv"))
dirnames = unlist(lapply(fnames, dirname))

varnames = strsplit(lnames, split = ".", fixed = TRUE)

suppressWarnings(lapply(dirnames, dir.create, recursive = TRUE))
invisible(lapply(seq_along(varnames), function(i)
write.csv(startlist[[varnames[[i]]]], file = fnames[[i]])))
}


That’s much better!