Programming with measure()

[markfairbanks]

The new measure() has some very useful functionality, but the design being based around symbols makes creating custom functions with it difficult.

Simple example:

library(data.table)

df <- data.table(
  x_1 = 1,
  x_2 = 2,
  y_1 = 1,
  y_2 = 2
)

# Goal output
melt(df, measure.vars = measure(group, id, sep = "_"))
#>    group id value
#> 1:     x  1     1
#> 2:     x  2     2
#> 3:     y  1     1
#> 4:     y  2     2

my_melt <- function(data, group_name, id_name) {
  melt(
    data,
    measure.vars = measure(group_name, id_name, sep = "_")
  )
}

my_melt(df, group, id)
#>    group_name id_name value
#> 1:          x       1     1
#> 2:          x       2     2
#> 3:          y       1     1
#> 4:          y       2     2

Is there a possibility of adding an argument to measure() that is character based (or some other solution)?

In the meantime is there a recommended workaround? I think I'm missing something simple, but I couldn't figure out a good way to do this.

[tdhock]

Hi I wrote the measure() function so I can help!
A simple easy and method is to use as.symbol and as.call,

> (measure.call <- as.call(list(
+   as.symbol("measure"),
+   as.symbol("group"),
+   as.symbol("id"),
+   sep="_")))
measure(group, id, sep = "_")
> (melt.call <- as.call(list(
+   as.symbol("melt"),
+   as.symbol("df"),
+   measure.vars=measure.call)))
melt(df, measure.vars = measure(group, id, sep = "_"))
> eval(melt.call)
   group id value
1:     x  1     1
2:     x  2     2
3:     y  1     1
4:     y  2     2
> 

As this is a straightforward use of base R I think we don't need to add an argument to measure() that is character based (or some other solution). If that is ok with you @markfairbanks can you please close?

[markfairbanks]

Though this workaround works, I still think that some other solution should be implemented to make this easier. For a few reasons:

1. As far as I know no other data.table functions require you to use as.call() or as.symbol() just to create a custom function, so it creates an inconsistency with the rest of data.table.
2. as.call() and as.symbol() are not common functions used/understood by the vast majority of R users.
3. All NSE functions currently in data.table have simple workarounds.
   • Ex1: setorder has setorderv (and other set*/set*v functions)
   • Ex2: [.data.table adding the env argument in v1.14.1

With those in mind here are a few proposals:

• Add an argument that is character based to measure()
• Add a measurev()
• Add an env argument to measure()

[tdhock]

Hi @markfairbanks thanks for the suggestion, please try my new measurev function in measurev branch, #5022

[tdhock]

well #5022 was recently merged into master so if the new measurev function works for you @markfairbanks can you please close this issue?

[markfairbanks]

Works great, thanks @tdhock 👍

library(data.table)

df <- data.table(
  x_1 = 1,
  x_2 = 2,
  y_1 = 1,
  y_2 = 2
)

my_melt <- function(data, fun_list) {
  melt(
    data,
    measure.vars = measurev(fun.list = fun_list, sep = "_")
  )
}

my_melt(df, list(group = NULL, id = NULL))
#>    group id value
#> 1:     x  1     1
#> 2:     x  2     2
#> 3:     y  1     1
#> 4:     y  2     2