performance - Alternative of slower ifelse in R data table

Question

I am writing a function where multiple ifelse are being used for data table operation. Although I am using data tables for speed but multiple ifelse making my code slow and this function is for large data set. Hence I was wondering if there is an alternative to iflese. One example iflese from the function(there are close to 15 iflese ), in this example flag is set to 1 if x is blank else 0.

    dt<-dt[,flag:=ifelse(is.na(x)|!nzchar(x),1,0)]

Answer 1

The fastest approach will probably depend on what your data looks like. Those mentioned in the comments are all comparable for this example:

(twice was mentioned by @DavidArenburg; and onceadd by @akrun. I'm not really sure how to benchmark these with replications > 1, since the objects are actually modified during the benchmark.)

DT <- data.table(x=sample(c(NA,"",letters),1e8,replace=TRUE))

DT0 <- copy(DT)
DT1 <- copy(DT)
DT2 <- copy(DT)
DT3 <- copy(DT)
DT4 <- copy(DT)
DT5 <- copy(DT)
DT6 <- copy(DT)
DT7 <- copy(DT)

library(rbenchmark)
benchmark(
ifelse  = DT0[,flag:=ifelse(is.na(x)|!nzchar(x),1L,0L)],
keyit   = {
    setkey(DT1,x)   
    DT1[,flag:=0L]
    DT1[J(NA_character_,""),flag:=1L]
},
twiceby = DT2[, flag:= 0L][is.na(x)|!nzchar(x), flag:= 1L,by=x],
twice   = DT3[, flag:= 0L][is.na(x)|!nzchar(x), flag:= 1L],
onceby  = DT4[, flag:= +(is.na(x)|!nzchar(x)), by=x],
once    = DT5[, flag:= +(is.na(x)|!nzchar(x))],
onceadd = DT6[, flag:= (is.na(x)|!nzchar(x))+0L],
oncebyk = {setkey(DT7,x); DT7[, flag:= +(is.na(x)|!nzchar(x)), by=x]},
replications=1
)[1:5]
#      test replications elapsed relative user.self
# 1  ifelse            1   19.61   31.127     17.32
# 2   keyit            1    0.63    1.000      0.47
# 6    once            1    3.26    5.175      2.68
# 7 onceadd            1    3.24    5.143      2.88
# 5  onceby            1    1.81    2.873      1.75
# 8 oncebyk            1    0.91    1.444      0.82
# 4   twice            1    3.17    5.032      2.79
# 3 twiceby            1    3.45    5.476      3.16

Discussion. In this example, keyit is the fastest. However, it's also the most verbose and it changes the sorting of your table. Also, keyit is very specific to the OP's question (taking advantage of the fact that exactly two character values fit the condition is.na(x)|!nzchar(x)), and so might not be as great for other applications, where it would need to be written something like

keyit   = {
    setkey(DT1,x)
    flagem = DT1[,some_other_condition(x),by=x][(V1)]$x
    DT1[,flag:=0L]
    DT1[J(flagem),flag:=1L]
}