Elaborate on map(.id = vars)

[wlandau]

• Choosing ID variables.
• Advantages and dangers of map(.id = FALSE).
• How to define custom independent IDs for targets.

[wlandau]

Problem: when you add new targets, sometimes it disrupts the indexing and invalidates other targets.

library(drake)
dataset <- c("census", "worldbank")
drake_plan(
  x = target(
    analyze_data(data),
    transform = map(data = !!dataset, .id = FALSE)
  )
)
#> # A tibble: 2 x 2
#>   target command                  
#>   <chr>  <expr>                   
#> 1 x      analyze_data("census")   
#> 2 x_2    analyze_data("worldbank")

# Add new data
dataset <- c("census", "gapminder", "worldbank")

# The World Bank analysis changed from x_2 to x_3,
# which invalidates the target.
drake_plan(
  x = target(
    analyze_data(data),
    transform = map(data = !!dataset, .id = FALSE)
  )
)
#> # A tibble: 3 x 2
#>   target command                  
#>   <chr>  <expr>                   
#> 1 x      analyze_data("census")   
#> 2 x_2    analyze_data("gapminder")
#> 3 x_3    analyze_data("worldbank")

^{Created on 2019-02-27 by the reprex package (v0.2.1)}

Solution: create your own id. Below, the World Bank target is always x_b9e5ea49.

library(drake)
library(tidyverse)
  
my_id <- function(x) {
  map_chr(x, digest::digest, algo = "murmur32", serialize = FALSE) %>%
    rlang::syms()
}

dataset <- c("census", "worldbank")
id_vals <- my_id(dataset)

drake_plan(
  x = target({
      quote(id)
      analyze_data(data)
    },
    transform = map(data = !!dataset, id = !!id_vals, .id = id)
  )
)
#> # A tibble: 2 x 2
#>   target     command                                              
#>   <chr>      <expr>                                               
#> 1 x_365d5563 {     quote(`365d5563`)     analyze_data("census") } 
#> 2 x_b9e5ea49 {     quote(b9e5ea49)     analyze_data("worldbank") }

dataset <- c("census", "gapminder", "worldbank")
id_vals <- my_id(dataset)
drake_plan(
  x = target({
    quote(id)
    analyze_data(data)
  },
  transform = map(data = !!dataset, id = !!id_vals, .id = id)
  )
)
#> # A tibble: 3 x 2
#>   target     command                                                
#>   <chr>      <expr>                                                 
#> 1 x_365d5563 {     quote(`365d5563`)     analyze_data("census") }   
#> 2 x_3d724852 {     quote(`3d724852`)     analyze_data("gapminder") }
#> 3 x_b9e5ea49 {     quote(b9e5ea49)     analyze_data("worldbank") }

^{Created on 2019-02-27 by the reprex package (v0.2.1)}

[wlandau]

So then why is this not already the default behavior of map(.id = FALSE)? Because:

1. Hash suffixes are cryptic and ugly.
2. An automated solution would not completely solve the problem. If your grid of grouping variables has duplicates, we need still need drake:::make_unique() to add suffixes _2, _3, etc.

library(drake)
library(tidyverse)

my_id <- function(x) {
  map_chr(x, digest::digest, algo = "murmur32", serialize = FALSE) %>%
    rlang::syms()
}

y_vals <- c(1, 1)
id_vals <- my_id(as.character(y_vals))
drake_plan(x = target({
    id
    y
  },
  transform = map(y = !!y_vals, id = !!id_vals, .id = id))
)
#> # A tibble: 2 x 2
#>   target       command                 
#>   <chr>        <expr>                  
#> 1 x_9416ac93   {     `9416ac93`     1 }
#> 2 x_9416ac93_2 {     `9416ac93`     1 }

^{Created on 2019-02-28 by the reprex package (v0.2.1)}