Problem saving targets object to data/ in a user-defined R package

[luke-a-rogers]

Hello!

I absolutely love targets. I want to use it inside my own R packages where I like to run my analyses. I ran into a problem saving targets objects to data/ via usethis::use_data(). Can you point me in the right direction to solve this problem?

Thanks so much!
Luke

Motivation

I want to write my analysis as an R package, and save the output object to data/ so that it is exported in a manner similar to other package data (mentioned here).

Problem

When I used the usethis::use_data() function inside my write_to_data() function, I got an error message:

> tar_make()
• start target my_data
• built target my_data
• start target write_my_data
...
x error target write_my_data
• end pipeline
Error : object ‘my_data’ not found

Question

What is the preferred way to save a targets object to data/ inside a user-defined R package?

[wlandau]

The difficulty comes from how targets handles in-memory data. To prevent accidental side-effects, in-memory targets are kept in a special environment that inherits from tar_option_get("envir"). To get around this, you can make sure each element of your data exists as a variable in the call frame of write_to_data().

write_to_data <- function (...) {
  args <- list(...)
  envir <- environment()
  purrr::walk2(
    .x = names(args),
    .y = args,
    ~assign(x = .x, value = .y, envir = envir)
  )
  usethis::use_data(..., overwrite = TRUE)
  file.path("data", fs::path_ext_set(names(args)[1], ".rda"))
}

This assumes you name your arguments, e.g. write_to_data(my_data = my_data).

I would also advise tar_option_set(packages = "usedatargets", imports = "usedatargets") instead of the older tar_option_set(envir = getNamespace("usedatargets")): https://books.ropensci.org/targets/practices.html#packages-based-invalidation

[luke-a-rogers]

Wonderful! This works exactly as I hoped. Thank you so much!

[luke-a-rogers]

I got curious how to make sure my data existed as a variable in the call frame even when the argument was not named. I opted to have it throw an error if more than one object was passed to write_to_data(). Working through this helped me understand where usethis::use_data() looks to match the dots. This script worked for me, riffing off your ideas above. I'm curious if this may have any shortcomings that I haven't anticipated:

write_to_data <- function (...) {
  if (...length() != 1) stop("write_to_data() takes exactly 1 argument")
  args <- list(...)
  if (is.null(names(args))) {
    name <- as.character(substitute(...))
  } else {
    name <- names(args)[1]
  }
  assign(x = name, value = ..1)
  usethis::use_data(..., overwrite = TRUE)
  file.path("data", fs::path_ext_set(name, ".rda"))
}

Thanks again!

[wlandau]

Nice, thanks for sharing.

If you only save one object, it may be simpler to do something like this. I have not tested it though.

write_to_data <- function (value) {
  name <- deparse(substitute(value))
  assign(x = name, value = value)
  call <- as.call(
    list(
      quote(usethis::use_data),
      value = as.symbol(name),
      overwrite = TRUE
    )
  )
  eval(call)
  file.path("data", fs::path_ext_set(name, ".rda"))
}