nested ifelse in R so close to working - r

Question

I'm working with the following four columns of raw weight measurement data and a very nearly-functioning nested ifelse statement that results in the 'kg' vector.
Id G4_R_2_4 G4_R_2_5 G4_R_2_5_option2 kg
219 13237 16.0 NA NA 16.0
220 139129 8.50 55.70 47.20 8.50
221 139215 28.9 NA NA 28.9
222 139216 NA 46.70 8.50 46.70
223 139264 12.40 NA NA 12.40
224 139281 13.60 NA NA 13.60
225 139366 16.10 NA NA 16.10
226 139376 61.80 NA NA 61.80
227 140103 NA 48.60 9.10 48.60
The goal is to merge the three 'G4' columns into kg based on the following conditions:
1) If G4_R_2_4 is not NA, print its value
2) If G4_R_2_4 is NA, print the lesser of the values appearing in G4_R_2_5 and G4_R_2_5_option2 (sorry for lame variable names)
I've been working with the following statement (big dataset called 'child'):
> child$kg <- ifelse(child$G4_R_2_4 == 'NA' & child$G4_R_2_5 < child$G4_R_2_5_option2,
child$G4_R_2_5, ifelse(child$G4_R_2_4 == 'NA' & child$G4_R_2_5 > child$G4_R_2_5_option2,
child$G4_R_2_5_option2, child$G4_R_2_4))
Which results in the 'kg' vector I have now. It seems to satisfy the G4_R_2_4 condition (is/is not NA) but always prints the value from G4_R_2_5 for the NA cases. How do I get it to incorporate the greater than/less than condition?

Answer 1

It's not clear from your example, but I think the problem is you're handling NA's incorrectly and\or using wrong type for data.frame's columns. Try rewriting your code like that:
#if your columns are of character type (warnings are ok)
child$G4_R_2_4<-as.numeric(child$G4_R_2_4)
child$G4_R_2_5<-as.numeric(child$G4_R_2_5)
child$G4_R_2_5_option2<-as.numeric(child$G4_R_2_5_option2)
#correct NA handling
child$kg<-ifelse(is.na(child$G4_R_2_4) & child$G4_R_2_5 <
child$G4_R_2_5_option2, child$G4_R_2_5, ifelse(is.na(child$G4_R_2_4) &
child$G4_R_2_5 > child$G4_R_2_5_option2, child$G4_R_2_5_option2, child$G4_R_2_4))

Answer 2

Here is an alternative version that might be interesting, assuming that the values are stored in numerical form (else the column entries should be converted into numerical values, as suggested in the other answers):
get_kg <- function(x){
if(!is.na(x[2])) return (x[2])
return (min(x[3], x[4], na.rm = T))}
child$kg <- apply(child,1,get_kg)
#> child
# Id G4_R_2_4 G4_R_2_5 G4_R_2_5_option2 kg
#219 13237 16.0 NA NA 16.0
#220 139129 8.5 55.7 47.2 8.5
#221 139215 28.9 NA NA 28.9
#222 139216 NA 46.7 8.5 8.5
#223 139264 12.4 NA NA 12.4
#224 139281 13.6 NA NA 13.6
#225 139366 16.1 NA NA 16.1
#226 139376 61.8 NA NA 61.8
#227 140103 NA 48.6 9.1 9.1

Answer 3

We could do this using pmin. Assuming that your 'G4' columns are 'character' class, we convert those columns to 'numeric' class and use pmin on that columns.
indx <- grep('^G4', names(child))
child[indx] <- lapply(child[indx], as.numeric)
d1 <- child[indx]
child$kgN <- ifelse(is.na(d1[,1]), do.call(pmin, c(d1[-1], na.rm=TRUE)), d1[,1])
child$kgN
#[1] 16.0 8.5 28.9 8.5 12.4 13.6 16.1 61.8 9.1
Or without using ifelse
cbind(d1[,1], do.call(pmin, c(d1[-1], na.rm=TRUE)))[cbind(1:nrow(d1),
(is.na(d1[,1]))+1L)]
#[1] 16.0 8.5 28.9 8.5 12.4 13.6 16.1 61.8 9.1
Benchmarks
set.seed(24)
child1 <- as.data.frame(matrix(sample(c(NA,0:50), 1e6*3, replace=TRUE),
ncol=3, dimnames=list(NULL, c('G4_R_2_4', 'G4_R_2_5',
'G4_R_2_5_option2'))) )
cyberj0g <- function(){
with(child1, ifelse(is.na(G4_R_2_4) & G4_R_2_5 <
G4_R_2_5_option2, G4_R_2_5, ifelse(is.na(G4_R_2_4) &
G4_R_2_5 > G4_R_2_5_option2, G4_R_2_5_option2, G4_R_2_4)))
}
get_kg <- function(x){
if(!is.na(x[2])) return (x[2])
return (min(x[3], x[4], na.rm = T))}
RHertel <- function() apply(child1,1,get_kg)
akrun <- function(){cbind(child1[,1], do.call(pmin, c(child1[-1],
na.rm=TRUE)))[cbind(1:nrow(child1), (is.na(child1[,1]))+1L)]}
system.time(cyberj0g())
# user system elapsed
# 0.451 0.000 0.388
system.time(RHertel())
# user system elapsed
# 11.808 0.000 10.928
system.time(akrun())
# user system elapsed
# 0.000 0.000 0.084
library(microbenchmark)
microbenchmark(cyberj0g(), akrun(), unit='relative', times=20L)
#Unit: relative
# expr min lq mean median uq max neval cld
# cyberj0g() 3.750391 4.137777 3.538063 4.091793 2.895156 3.197511 20 b
# akrun() 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000 20 a

Answer 4

I'm pretty sure the problem is that you're not testing whether the values are NAs, you're testing whether they are equal to the string "NA", which they never are. This should work:
child$kg <- ifelse(is.na(child$G4_R_2_4) &
child$G4_R_2_5 < child$G4_R_2_5_option2,
child$G4_R_2_5,
ifelse(is.na(child$G4_R_2_4) &
child$G4_R_2_5 > child$G4_R_2_5_option2,
child$G4_R_2_5_option2,
child$G4_R_2_4))