Canonical way to include one id column into all elements of resulting list from split.default

user_2884859 0
Viewed 117

I am splitting a data.frame into a list on the basis of its column names. What I want is to include a id column (id) to not just one item but into all elements of the resulting list.

Presently I am doing it through subsequent binding of id column to all items of list through map and bind_cols (alternatives through Map/do.call/mapply etc. I can do similarly myself). What I want to know is there any canonical way of doing it directly, maybe with a function argument of split.default or through some other function directly and thus saving two or three extra steps.

Reproducible example

df <- data.frame(
  stringsAsFactors = FALSE,
                id = c("A", "B", "C"),
             nm1_a = c(928L, 476L, 928L),
             nm1_b = c(61L, 362L, 398L),
             nm2_a = c(965L, 466L, 369L),
             nm2_b = c(240L, 375L, 904L),
             nm3_a = c(429L, 730L, 788L),
             nm3_b = c(99L, 896L, 540L),
             nm3_c = c(463L, 143L, 870L)
      )

df
#>   id nm1_a nm1_b nm2_a nm2_b nm3_a nm3_b nm3_c
#> 1  A   928    61   965   240   429    99   463
#> 2  B   476   362   466   375   730   896   143
#> 3  C   928   398   369   904   788   540   870

What I am doing presently

library(tidyverse)

split.default(df[-1], gsub('^(nm\\d+).*', '\\1', names(df)[-1])) %>%
  map(~ .x %>% bind_cols('id' = df$id, .))
#> $nm1
#>   id nm1_a nm1_b
#> 1  A   928    61
#> 2  B   476   362
#> 3  C   928   398
#> 
#> $nm2
#>   id nm2_a nm2_b
#> 1  A   965   240
#> 2  B   466   375
#> 3  C   369   904
#> 
#> $nm3
#>   id nm3_a nm3_b nm3_c
#> 1  A   429    99   463
#> 2  B   730   896   143
#> 3  C   788   540   870

What I want is exactly the same output, but is there any way to do it directly or a more canonical way?

dataframe r list split
3 Answers

Just for a diversity of options, here's what you said you didn't want to do. The pivot / split / pivot method can help scale better and adapt beyond keeping an ID based just on column position. It also makes use of the ID in order to do the reshaping, so it might also be more flexible if you have other operations to do in the intermediate steps and don't know for sure that your row order will stay the same—that's one of the reasons I sometimes avoid binding columns. It also (at least for me) makes sense to split data based on some variable rather than by groups of columns.

library(tidyr)

df %>%
  pivot_longer(-id) %>%
  split(stringr::str_extract(.$name, "^nm\\d+")) %>%
  purrr::map(pivot_wider, id_cols = id, names_from = name)
#> $nm1
#> # A tibble: 3 x 3
#>   id    nm1_a nm1_b
#>   <chr> <int> <int>
#> 1 A       928    61
#> 2 B       476   362
#> 3 C       928   398
#> 
#> $nm2
#> # A tibble: 3 x 3
#>   id    nm2_a nm2_b
#>   <chr> <int> <int>
#> 1 A       965   240
#> 2 B       466   375
#> 3 C       369   904
#> 
#> $nm3
#> # A tibble: 3 x 4
#>   id    nm3_a nm3_b nm3_c
#>   <chr> <int> <int> <int>
#> 1 A       429    99   463
#> 2 B       730   896   143
#> 3 C       788   540   870
user_53258620

This one should be just two steps, split and replace.

Map(`[<-`, split.default(df[-1], substr(names(df)[-1], 1, 3)), 'id', value=df[1])
# $nm1
#   nm1_a nm1_b id
# 1   928    61  A
# 2   476   362  B
# 3   928   398  C
# 
# $nm2
#   nm2_a nm2_b id
# 1   965   240  A
# 2   466   375  B
# 3   369   904  C
# 
# $nm3
#   nm3_a nm3_b nm3_c id
# 1   429    99   463  A
# 2   730   896   143  B
# 3   788   540   870  C
user_65740380

You can make use of a temporary variable so that the code is cleaner and easy to understand.

common_cols <- 1
tmp <- df[-common_cols]

lapply(split.default(tmp, sub('^(nm\\d+).*', '\\1', names(tmp))), 
       function(x) cbind(df[common_cols], x))

#$nm1
#  id nm1_a nm1_b
#1  A   928    61
#2  B   476   362
#3  C   928   398

#$nm2
#  id nm2_a nm2_b
#1  A   965   240
#2  B   466   375
#3  C   369   904

#$nm3
#  id nm3_a nm3_b nm3_c
#1  A   429    99   463
#2  B   730   896   143
#3  C   788   540   870
user_39629140
Related