I() does not preserve a vector's names in `data.table` unlike `data.frame`

[dcaseykc]

According to the documentation of data.frame, I() can be used to prevent data.frame from stripping the names of a vector. By contrast, data.table strips the names still but does preserve the "asis" class. Is this difference expected?

library('data.table')
fake = setNames(1:3, c('a','b','c'))

data.frame(v1 = I(fake))[, 'v1']
#> a b c 
#> 1 2 3
class(data.frame(v1 = I(fake))[, 'v1'])
#> [1] "AsIs"
data.table(v1 = I(fake))[, v1]
#> [1] 1 2 3
class(data.table(v1 = I(fake))[, v1])
#> [1] "AsIs"

sessionInfo()
#> R version 4.0.2 (2020-06-22)
#> Platform: x86_64-w64-mingw32/x64 (64-bit)
#> Running under: Windows 10 x64 (build 18363)
#> 
#> Matrix products: default
#> 
#> locale:
#> [1] LC_COLLATE=English_United States.1252 
#> [2] LC_CTYPE=English_United States.1252   
#> [3] LC_MONETARY=English_United States.1252
#> [4] LC_NUMERIC=C                          
#> [5] LC_TIME=English_United States.1252    
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] data.table_1.13.0
#> 
#> loaded via a namespace (and not attached):
#>  [1] compiler_4.0.2  magrittr_1.5    tools_4.0.2     htmltools_0.5.0
#>  [5] yaml_2.2.1      stringi_1.4.6   rmarkdown_2.3   highr_0.8      
#>  [9] knitr_1.29      stringr_1.4.0   xfun_0.16       digest_0.6.25  
#> [13] rlang_0.4.7     evaluate_0.14

^{Created on 2020-11-02 by the reprex package (v0.3.0)}

[jangorecki]

I never seen use of named vector in data.table. I would suggest to put them in another column.
If they should work is of course a valid question.

[dcaseykc]

In case it helps, my use case is around computing the relative fractions of factors and I was hoping to store the factor levels in names. Here I've named the function relative_fraction, but the main approach would be to have some general function (e.g. calc) and use s3 dispatching to handle different classes. Because I'd like to use the same entry point for different types of variables, returning two columns sometimes and one column other times (particularly when J includes ./list) seems like it'll get tricky.

library('data.table')

#fake data
dt = data.table(factor(sample(1:3, 100, replace = T), 1:3, c('a','b','c')))

relative_fraction = function(x){
  r = as.numeric(table(x)/length(x))
  names(r) = levels(x)
  setattr(r, 'lvls', levels(x))
  
  return(r)
}

dt[, relative_fraction(V1)] #names and attribute are preserved
#>    a    b    c 
#> 0.26 0.32 0.42 
#> attr(,"lvls")
#> [1] "a" "b" "c"
r1 = dt[, .(relative_fraction(V1))]
r1[, V1] #only the attribute is preserved
#> [1] 0.26 0.32 0.42
#> attr(,"lvls")
#> [1] "a" "b" "c"


dt2 = data.table(factor(sample(1:3, 100, replace = T), 1:3, c('d','e','f')))
dt2[, relative_fraction(V1)] #names and attributes
#>    d    e    f 
#> 0.42 0.29 0.29 
#> attr(,"lvls")
#> [1] "d" "e" "f"
r2 = dt2[, .(relative_fraction(V1))]
r2[, V1] #only the attributes
#> [1] 0.42 0.29 0.29
#> attr(,"lvls")
#> [1] "d" "e" "f"

#Ideally these would be the same
c(dt[, relative_fraction(V1)], dt2[, relative_fraction(V1)]) #names, no attribute
#>    a    b    c    d    e    f 
#> 0.26 0.32 0.42 0.42 0.29 0.29
c(r1[,V1], r2[,V1]) #not names and also not attribute
#> [1] 0.26 0.32 0.42 0.42 0.29 0.29
rbind(r1, r2)[,V1] #only the attributes of the first one (not ideal, but I think thats just how rbind works)
#> [1] 0.26 0.32 0.42 0.42 0.29 0.29
#> attr(,"lvls")
#> [1] "a" "b" "c"


rbind(data.frame(v = I(setNames(1, 'a'))), data.frame(v = I(setNames(1, 'b'))))[, 'v'] #names make it through rbind for data.frames
#> a b 
#> 1 1

[jangorecki]

I think the following could do the job

dt[, relative_fraction(setNames(V1, V2))]

or if needed to be very verbose

dt[, {
  named_vec = setNames(V1, V2)
  relative_fraction(named_vec)
}]

---

wait a moment, now I see you want to return named vector rather than take it as input.
I would suggest then to store them in a list, because inside a list you can nest whatever you want, including named vector.

data.table(id=1:2, v=list(c("a"=1L,"b"=2L), c("z"=9L)))$v

But for performance it still seems better to produce 2 columns, one with a names and another with values.

[MichaelChirico]

I agree we shouldn't really support named vectors in data.tables. It would be very complicated to keep track of that during e.g. grouping, what should happen when we group by c(a=1, b=1), should there be 1 or two groups? If one group, which name should be kept?

Now with #7136, we can conveniently extract the names to a column during conversion: as.data.table(fake, keep.rownames=TRUE).