remapping categorical variables in base and tidy

[markvanderloo]

Hi Norm,

I just read your base-vs-tidy document. Very insightful, thank you! I'm sharing it with our internal R users.

Regarding the last section on relabeling classes. Have you considered this option?

map <- c()
map[3:5] <- c("three", "four", "five")
mtcars$gear <- map[mtcars$gear]

Using named vectors as a way to map values onto each other has been an extremely useful trick for me in many cases, although I admit that it may not be very intuitive for beginning users.

A more common use case seems to me is mapping character to character, in which case one simply does something like

map <- c("a"="alpha", "b"="beta")
map[letters]

you even get NA for unmapped values, which seems to me desired behavior in many cases.

Thanks again for writing all this up!
Best,
Mark

[danielreispereira]

I still think that the nested ifelse() calls is preferable over the case_when() construct. After all, that is what an average human would do on Excel: recode the levels manually with IF()s.

Teaching case_when() is suggesting that instead of sitting down and thinking for 10min one should go for the documentation of a yet another function.

The point is that the R-base documentation is not as sexy, and does not appeal to newcomers.

[BroVic]

@markvanderloo that's a very useful trick! I think we need a function in Base-R that does this efficiently.

[dusadrian]

Another solution in base R:

library(admisc)
recode(mtcars$gear, "3 = three; 4 = four; 5 = five")
#  [1] "four"  "four"  "four"  "three" "three" "three" "three" "four"  "four"  "four"  "four" 
# [12] "three" "three" "three" "three" "three" "three" "four"  "four"  "four"  "three" "three"
# [23] "three" "three" "three" "four"  "five"  "five"  "five"  "five"  "five"  "four"

Or, since this is a categorical variable, why not properly declare it as categorical:

library(declared)
gear <- declared(mtcars$gear, label = "Number of gears", labels = c("Three" = 3, "Four" = 4, "Five" = 5))
gear
# <declared<integer>[32]> Number of gears
#  [1] 4 4 4 3 3 3 3 4 4 4 4 3 3 3 3 3 3 4 4 4 3 3 3 3 3 4 5 5 5 5 5 4
# 
# Labels:
#  value label
#      3 Three
#      4  Four
#      5  Five

w_table(gear)

#       fre    rel   per   cpd
#       ----------------------
# Three  15  0.469  46.9  46.9 
#  Four  12  0.375  37.5  84.4 
#  Five   5  0.156  15.6 100.0 
#       ----------------------
#        32  1.000 100.0

[BroVic]

This is another nice approach but I think this is so basic that we ought to have it natively in the language. Or does anyone know of such a solution? I remember trying to get this done with the replace function.

[dusadrian]

@BroVic: that requires a question about what "base" R means: is it anything which is not using tidy, or is it the R package base?

I believe this is about the difference between the "normal" R language (including contributed R packages that use the normal R language), and the "tidy" R language / dialect.

It would be pointless to expect solving anything using just the base package: this is why contributed packages exist, in the first place.

[BroVic]

@dusadrian Yes, I agree with you. But if you ask me, an operation as basic as this should be available by default.

[matloff]

Once again, keep it simple. I think use of named vectors falls into the realm, yes.