Recode only certain values keep others as it is while preserving multiple data types in different co-CodePudding

I have a data frame with more than 1000 character columns. I want to recode certain character columns as numeric but keep other columns as characters. I hope to achieve this without manually including/excluding columns for recoding (aka the operation needs to be on the entire data frame). I tried using dplyr::recode and dplyr::case_when but the unrecoded columns were all coerced into NA or I had to convert all columns into numeric.

A toy dataset:

 df = tibble(name = c('Tom', 'Sarah'), wealth = c('poor', 'wealthy'), health = c('poor', 'good'))

I think dplyr::recode and dplyr::case require all columns to be the same type after recoding, yet I don't want the 'name' column to be numeric.

I tried using:

df_recoded = df %>% mutate(across(everything(),~recode(
  ., 
  'poor' = 1, 
  'good' = 2,
  'wealthy' = 2,
  .default = .    #doesn't work
)))

 df_recoded = df %>% mutate(across(everything(),~case_when(
      . == 'poor' ~ 1, 
      . == 'good' ~ 2,
      . == 'wealthy' ~ 2, 
      TRUE ~ .    # doesn't work. 
    )))

The desired output is:

#> # A tibble: 2 x 3
#>   name  wealth health
#>   <chr> <dbl>  <dbl> 
#> 1 Tom   1      1     
#> 2 Sarah 2      2

CodePudding user response：

Issue is that 1 and 2 needs to be characters when you do this operation.

library(tidyverse)

df = tibble(name = c('Tom', 'Sarah'), 
            wealth = c('poor', 'wealthy'), 
            health = c('poor', 'good')) 

df %>%  
  mutate(across(everything(), 
                ~ case_when(.x == "poor" ~ "1", 
                                 .x %in% c("good", "wealthy") ~ "2",
                                 TRUE ~ .x)))
#> # A tibble: 2 x 3
#>   name  wealth health
#>   <chr> <chr>  <chr> 
#> 1 Tom   1      1     
#> 2 Sarah 2      2

^{Created on 2022-06-29 by the reprex package (v2.0.1)}

CodePudding user response：

Try this

library(tidyverse)

df = tibble(name = c('Tom', 'Sarah'), 
            wealth = c('poor', 'wealthy'), 
            health = c('poor', 'good')) 

df %>%  
    mutate(across(everything(), 
                  ~ case_when(.x == "poor" ~ "1", 
                              .x %in% c("good", "wealthy") ~ "2",
                              TRUE ~ .x))) -> df_recoded

df_recoded[] <- lapply(names(df_recoded) , 
\(x) if(x != "name") as.numeric(df_recoded[[x]]) else df_recoded[[x]])

you can use base R to simplify the solution

l <- lapply(df , \(x) ifelse(x == "poor" , 1 , ifelse(x %in% c("good", "wealthy") , 2 , x)))

l$wealth <- as.numeric(l$wealth)
l$health <- as.numeric(l$health)

as_tibble(l)

output

df_recoded

# A tibble: 2 × 3
  name  wealth health
  <chr>  <dbl>  <dbl>
1 Tom        1      1
2 Sarah      2      2