In R, I have 600,000 categorical variables, each of which is classified as "0", "1", or "2".
What I would like to do is collapse "1" and "2" and leave "0" by itself, such that after re-categorizing "0" = "0"; "1" = "1" and "2" = "1". In the end I only want "0" and "1" as categories for each of the variables.
Also, if possible, I would rather not create 600,000 new variables, if I can replace the existing variables with the new values that would be great!
What would be the best way to do this?
I find this is even more generic using factor(new.levels[x]):
> x <- factor(sample(c("0","1","2"), 10, replace=TRUE))
> x
[1] 0 2 2 2 1 2 2 0 2 1
Levels: 0 1 2
> new.levels<-c(0,1,1)
> x <- factor(new.levels[x])
> x
[1] 0 1 1 1 1 1 1 0 1 1
Levels: 0 1
The new levels vector must the same length as the number of levels in x, so you can do more complicated recodes as well using strings and NAs for example
x <- factor(c("old", "new", NA)[x])
> x
[1] old <NA> <NA> <NA> new <NA> <NA> old
[9] <NA> new
Levels: new old
recode()'s a little overkill for this. Your case depends on how it's currently coded. Let's say your variable is x.
If it's numeric
x <- ifelse(x>1, 1, x)
if it's character
x <- ifelse(x=='2', '1', x)
if it's factor with levels 0,1,2
levels(x) <- c(0,1,1)
Any of those can be applied across a data frame dta to the variable x in place. For example...
dta$x <- ifelse(dta$x > 1, 1, dta$x)
Or, multiple columns of a frame
df[,c('col1','col2'] <- sapply(df[,c('col1','col2'], FUN = function(x) ifelse(x==0, x, 1))
There is a function recode in package car (Companion to Applied Regression):
require("car")
recode(x, "c('1','2')='1'; else='0'")
or for your case in plain R:
> x <- factor(sample(c("0","1","2"), 10, replace=TRUE))
> x
[1] 1 1 1 0 1 0 2 0 1 0
Levels: 0 1 2
> factor(pmin(as.numeric(x), 2), labels=c("0","1"))
[1] 1 1 1 0 1 0 1 0 1 0
Levels: 0 1
Update: To recode all categorical columns of a data frame tmp you can use the following
recode_fun <- function(x) factor(pmin(as.numeric(x), 2), labels=c("0","1"))
require("plyr")
catcolwise(recode_fun)(tmp)
I liked the function in dplyr that can quickly recode values.
library(dplyr)
df$x <- recode(df$x, old = "new")
Hope this helps :)
You could use the rec function of the sjmisc package, which can recode a complete data frame at once (given, that all variables have at least the same recode-values).
library(sjmisc)
mydf <- data.frame(a = sample(0:2, 10, T),
b = sample(0:2, 10, T),
c = sample(0:2, 10, T))
> mydf
a b c
1 1 1 0
2 1 0 1
3 0 2 0
4 0 1 0
5 1 0 0
6 2 1 1
7 0 1 1
8 2 1 2
9 1 1 2
10 2 0 1
mydf <- rec(mydf, "0=0; 1,2=1")
a b c
1 1 1 0
2 1 0 1
3 0 1 0
4 0 1 0
5 1 0 0
6 1 1 1
7 0 1 1
8 1 1 1
9 1 1 1
10 1 0 1
A solution with forcats package from tidyverse
library(forcats)
> x <- factor(sample(c("0","1","2"), 10, replace=TRUE))
> x
[1] 1 1 1 0 1 0 2 0 1 0
Levels: 0 1 2
> fct_collapse(x, "1" = c("1", "2"))
[1] 1 1 1 0 1 0 1 0 1 0
Levels: 0 1
Note that if you just want the results to be 0-1 binary variables, you can forego factors altogether:
f <- sapply(your.data.frame, is.factor)
your.data.frame[f] <- lapply(your.data.frame[f], function(x) x != "0")
The second line can also be written more succinctly (but possibly more cryptically) as
your.data.frame[f] <- lapply(your.data.frame[f], `!=`, "0")
This turns your factors into a series of logical variables, with "0" mapping to FALSE and anything else mapping to TRUE. FALSE and TRUE will be treated as 0 and 1 by most code, which in turn should give essentially the same result in an analysis as using a factor with levels "0" and "1". In fact, if it doesn't give the same result, that would cast doubt on the correctness of the analysis....
