Automatically creating new variables through interaction between two pre-existing variables

Question

Suppose I own the following set of dados:

dados
#>    letras numeros  cores valor
#> 1       a       1 branco     2
#> 2       a       1  preto     1
#> 3       a       2 branco     9
#> 4       a       2  preto     4
#> 5       a       3 branco     8
#> 6       a       3  preto     4
#> 7       a       4 branco     3
#> 8       a       4  preto     6
#> 9       b       1 branco     3
#> 10      b       1  preto     1
#> 11      b       2 branco    10
#> 12      b       2  preto     5
#> 13      b       3 branco     7
#> 14      b       3  preto    10
#> 15      b       4 branco    10
#> 16      b       4  preto    10
#> 17      c       1 branco    10
#> 18      c       1  preto     2
#> 19      c       2 branco     8
#> 20      c       2  preto     7
#> 21      c       3 branco     5
#> 22      c       3  preto     5
#> 23      c       4 branco     5
#> 24      c       4  preto     3


dados <- 
structure(list(letras = c("a", "a", "a", "a", "a", "a", "a", 
"a", "b", "b", "b", "b", "b", "b", "b", "b", "c", "c", "c", "c", 
"c", "c", "c", "c"), numeros = c(1L, 1L, 2L, 2L, 3L, 3L, 4L, 
4L, 1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L, 1L, 1L, 2L, 2L, 3L, 3L, 4L, 
4L), cores = c("branco", "preto", "branco", "preto", "branco", 
"preto", "branco", "preto", "branco", "preto", "branco", "preto", 
"branco", "preto", "branco", "preto", "branco", "preto", "branco", 
"preto", "branco", "preto", "branco", "preto"), valor = c(2L, 
1L, 9L, 4L, 8L, 4L, 3L, 6L, 3L, 1L, 10L, 5L, 7L, 10L, 10L, 10L, 
10L, 2L, 8L, 7L, 5L, 5L, 5L, 3L)), class = "data.frame", row.names = c(NA, 
-24L))

I would like to get new columns for it, with the combination between the pre-existing columns. For example, if I want to join the columns letras and numeros in a column, just use the function unite:

library(tidyverse)

dados %>% 
  unite(letras_numeros, c("letras", "numeros"), remove = FALSE)
#>    letras_numeros letras numeros  cores valor
#> 1             a_1      a       1 branco     2
#> 2             a_1      a       1  preto     1
#> 3             a_2      a       2 branco     9
#> 4             a_2      a       2  preto     4
#> 5             a_3      a       3 branco     8
#> 6             a_3      a       3  preto     4
#> 7             a_4      a       4 branco     3
#> 8             a_4      a       4  preto     6
#> 9             b_1      b       1 branco     3
#> 10            b_1      b       1  preto     1
#> 11            b_2      b       2 branco    10
#> 12            b_2      b       2  preto     5
#> 13            b_3      b       3 branco     7
#> 14            b_3      b       3  preto    10
#> 15            b_4      b       4 branco    10
#> 16            b_4      b       4  preto    10
#> 17            c_1      c       1 branco    10
#> 18            c_1      c       1  preto     2
#> 19            c_2      c       2 branco     8
#> 20            c_2      c       2  preto     7
#> 21            c_3      c       3 branco     5
#> 22            c_3      c       3  preto     5
#> 23            c_4      c       4 branco     5
#> 24            c_4      c       4  preto     3

How could I automate this process so that all combinations between two and three columns could be created? That is, how to get at the end a data frame that has columns letras_numeros, letras_cores, numeros_cores, letras_numeros_cores, letras, numeros, cores and valor?

Answer 1

This base R solution is not a function, but gives an idea of how to get the desired result. It uses combn to apply interaction combinations 2 and 2 of the columns "letras", "numeros" and "cores". After that apply the same function to the three columns. Then it is only cbind the results.

fun <- function(x, envir = as.environment(dados)){
  X <- mget(x, envir = envir)
  as.character(interaction(X, sep = "_"))
}

variaveis <- names(dados)[-4]
fac1 <- combn(variaveis, 2, fun, simplify = FALSE)
names(fac1) <- sapply(combn(variaveis, 2, simplify = FALSE), paste, collapse = "_")
fac2 <- fun(variaveis)

But now the column corresponding to fac2 has the wrong name:

names(res)
#[1] "letras_numeros" "letras_cores"   "numeros_cores"  "fac2"           "letras"         "numeros"       
#[7] "cores"          "valor"

You fix that and that’s it.

names(res)[names(res) == "fac2"] <- paste(variaveis, collapse = "_")
head(res)
#  letras_numeros letras_cores numeros_cores letras_numeros_cores letras numeros  cores valor
#1            a_1     a_branco      1_branco           a_1_branco      a       1 branco     2
#2            a_1      a_preto       1_preto            a_1_preto      a       1  preto     1
#3            a_2     a_branco      2_branco           a_2_branco      a       2 branco     9
#4            a_2      a_preto       2_preto            a_2_preto      a       2  preto     4
#5            a_3     a_branco      3_branco           a_3_branco      a       3 branco     8
#6            a_3      a_preto       3_preto            a_3_preto      a       3  preto     4

This function is easily generalizable. And automatically solve the column problem with the wrong name, the right name is also assigned within the cycle lapply.

funCombinacoes <- function(data, vars, n){
  fun <- function(x, envir = as.environment(data)){
    X <- mget(x, envir = envir)
    as.character(interaction(X, sep = "_"))
  }
  res <- lapply(n, function(m){
    fac1 <- combn(vars, m, fun, simplify = FALSE)
    names(fac1) <- sapply(combn(vars, m, simplify = FALSE), paste, collapse = "_")
    fac1
  })
  res <- do.call(cbind, unlist(res, recursive = FALSE))
  cbind.data.frame(res, data)
}

res2 <- funCombinacoes(dados, variaveis, n = 2:3)
identical(res, res2)
#[1] TRUE