Feature Request: Map one Factor to Another

[billdenney]

I often need to map a factor to another variable that either is already a factor or I need to make into a factor.

Would you consider including something like the following?

fct_mirror <- function(.f, .x) {
  factor(.x, levels=levels(.f), ordered=is.ordered(.f))
}

(This would solve an issue similar to tidyverse/dplyr#3731 by allowing fct_mirror(my_factor, case_when(...)).)

[hadley]

I think the problem is a bit underspecified currently. I can see at least three things you might want to do:

• create a new factor (e.g. original motivation from SO)
• change levels of existing factor
• change values of existing factor

Your proposal would help with the last, but not the others. I'd likely to fully explore the problem space before thinking about a solution.

[billdenney]

You're right, the original SO question was different. The original SO question seems like a mix of fct_relevel and case_when.

To change the levels of an existing factor, I would think that is the work of fct_recode or fct_expand, but maybe you have a different use case in mind.

My code above intends to allow remapping of one factor to another as is a common case working with clinical trial data:

library(dplyr)                                                                            
#> Warning: package 'dplyr' was built under R version 3.4.4
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
d <- expand.grid(Treatment_Original=factor(c("Treatment 1", "Treatment 2"), ordered=TRUE),
Day=c(-1, 1))                                                                             
                                                                                          
# Fails because "Baseline" and `Treatment_Original` are different classes                 
d %>%                                                                                     
mutate(Treatment_Baseline=case_when(Day == -1~"Baseline",                                 
TRUE~Treatment_Original))                                                                 
#> Warning: package 'bindrcpp' was built under R version 3.4.4
#> Error in mutate_impl(.data, dots): Evaluation error: must be type character, not integer.
# Succeeds, but loses the fact that I want a factor for `Treatment_Baseline`              
d %>%                                                                                     
mutate(Treatment_Baseline=case_when(Day == -1~"Baseline",                                 
TRUE~as.character(Treatment_Original)))                                                   
#>   Treatment_Original Day Treatment_Baseline
#> 1        Treatment 1  -1           Baseline
#> 2        Treatment 2  -1           Baseline
#> 3        Treatment 1   1        Treatment 1
#> 4        Treatment 2   1        Treatment 2
# Succeeds, but is cumbersome                                                             
d %>%                                                                                     
mutate(Treatment_Baseline=factor(                                                         
case_when(Day == -1~"Baseline",                                                           
TRUE~as.character(Treatment_Original)),                                                   
levels=c("Baseline", levels(Treatment_Original)),                                         
ordered=is.ordered(Treatment_Original)))                                                  
#>   Treatment_Original Day Treatment_Baseline
#> 1        Treatment 1  -1           Baseline
#> 2        Treatment 2  -1           Baseline
#> 3        Treatment 1   1        Treatment 1
#> 4        Treatment 2   1        Treatment 2

(This use case is made more relevant if #138 is implemented.)