Targets, tar_map, Directories with dates

[decat7]

First of all thanks Will for an awesome package. I am issues with directories with dates being converted into as.numeric dates only in the context of running the pipeline.

Reprex:

R/make_dataset.R

make_dataset <- function(path,data,date_fun){
  require("data.table")

  if(!dir.exists(file.path(path,date_fun))){
    dir.create(file.path(path,date_fun),recursive = TRUE)
  }

  fwrite(data,file.path(path,date_fun,"data.csv"))

  return(file.path(path,date_fun))
}

Run outside of target:

dat = data.frame(x = sample.int(100), y = sample.int(100),sample(letters[1:2],100,replace = TRUE))

make_dataset(path = "./data/preexist/data_source1",data = dat,date_fun = as.Date("2022-11-04"))

Output:
"./data/preexist/data_source1/2022-11-04"

make_dataset(path = "./data/preexist/data_source1",data = dat,date_fun = "2022-11-04")
Output:
"./data/preexist/data_source1/2022-11-04"

The output is the same for as.Date and no as.Date and is the desired effect.

Run with of targets and tar_map with static branching:
_targets.R

library(targets)
library(tarchetypes)
library(tidyr)
library(data.table)

source("R/make_dataset.R")
tar_option_set(packages = c("dplyr","data.table"))

dat = data.frame(x = sample.int(100), y = sample.int(100),sample(letters[1:2],100,replace = TRUE))

values <- tibble(
  target_id = c("data_source1_2022_11_04","data_source1_2022_11_11"),
  make_dataset_function = rlang::syms(c("make_dataset", "make_dataset")),
  summary_function = rlang::syms(c("make_summary", "make_summary")),
  date_fun = c("2022-11-04","2022-11-11")%>%as.Date(),
  input_path = c("./data/preexist/data_source1", "./data/preexist/data_source1")
)

targets <- tar_map(
  values = values,
  names = "target_id",
  tar_target(input,input_path,format = "file"),
  tar_target(data, make_dataset_function(input, dat,date_fun))
  #,tar_target(summary, summary_function(data))
)
list(targets)

tar_make()

The folders are created as as.numeric(as.Date("2022-11-04")) which is 19300 instead of "2022-11-04"

It is a requirement that my function arg date_fun be a date data type and not a character. date_fun = c("2022-11-04","2022-11-11")%>%as.Date(), in the values tibble. Changing date_fun to character will cause problems with other date calculations in more complex function steps. date_fun is would be used for more than just directory creation.

It is unclear if this is a tibble dates issues or targets issues in write date as a folder names. Any thoughts on what area is forcing as.numeric(as.Date("2022-11-04"))?

Thanks in advance!

[wlandau]

This is an instance of #94 (comment). You could try substituting expressions instead of the actual data values.

values <- tibble(
  ...,
  date_fun = purrr::map(c("2022-11-04","2022-11-11"), ~rlang::call2("as.Date", .x))
)

It is a requirement that my function arg date_fun be a date data type and not a character.

There are several workarounds in R that may allow you to relax this requirement: for example, a wrapper around make_dataset() that converts a supplied character string to a date before calling make_dataset().

[decat7]

Thanks for explaining. This workaround is fine.