我想在我的数据框中添加一个新变量(N_notNAs),它定义是否有任何其他变量是NA.
x y z N_notNAs
2 3 NA NA
NA 1 3 NA
2 3 5 1
4 4 3 1
Dav*_*urg 11
不知道为什么这是你想要的输出,但实现这一目标的一种可能方法是将NA每行相加并加上它的功效NA- 这样NA ^ 0将返回1,其他一切都将变为NA
NA^rowSums(is.na(df))
# [1] NA NA 1 1
@David Arenburg的解决方案非常漂亮但不管怎么说还有一些.
前4个假设列df是数字并使用相同的基本思想,而后3个不需要数字列.他们也使用一个共同的想法.
该ifelse解决方案在字符数方面最长,但可能更具可读性且最不棘手.
正如下面的@alexis_laz所提到的,apply(...)可以!complete.cases(df)在最后3个解决方案中替换,并且我们会得到一些简化ifelse(complete.cases(df), 1, NA),c(NA, 1)[complete.cases(df) + 1]并且match(complete.cases(df), TRUE).
rowSums(0*df) + 1
max.col(0*df) + 1
do.call(pmin, 0*df) + 1
do.call(pmax, 0*df) + 1
ifelse(apply(df, 1, anyNA), NA, 1)
c(NA, 1)[apply(df, 1, anyNA) + 1]
match(apply(df, 1, anyNA), FALSE)
有一些问题为什么你想要一个NA或1的列.如果你想要一个逻辑TRUE/FALSE结果那么apply(df, 1, anyNA)就可以了.
免责声明:此处基准测试的一些答案产生TRUE/FALSE向量和NOT NA或1向量.
为了便于获取信息,这里是我的机器上的基准测试,用于3列data.frame,包含1e6行:
Unit: milliseconds
expr min lq mean median uq max neval cld
alexis_laz(df) 12.87138 13.30044 15.46142 13.49258 13.80019 29.59228 10 a
akrun(df) 46.06203 48.31564 49.82198 49.94947 51.05219 53.91161 10 ab
GGrothendieck3(df) 55.42513 56.59798 69.37274 59.16803 64.44442 155.62797 10 ab
GGrothendieck4(df) 54.88489 58.08043 69.54111 58.63820 65.36838 149.21380 10 ab
GGrothendieck2(df) 60.26961 62.37184 97.93301 69.80034 158.39302 193.03562 10 bc
By989(df) 115.30531 118.81843 133.44343 123.17356 130.36815 223.22601 10 c
GGrothendieck1(df) 123.99504 128.61030 140.62055 132.31073 137.83856 220.33666 10 c
David(df) 131.42639 131.66415 143.03384 133.50082 136.29453 225.17487 10 c
GGrothendieck7(df) 1100.69319 1109.60500 1147.25668 1142.83955 1156.37090 1270.32547 10 d
GGrothendieck6(df) 1060.97719 1124.85486 1148.54833 1140.91949 1170.62952 1247.80220 10 d
GGrothendieck5(df) 1218.79235 1251.03109 1287.47851 1285.20543 1311.82753 1364.89158 10 e
PaulHiemstra(df) 1436.31149 1461.14340 1511.42476 1502.34413 1552.09517 1608.22418 10 f
对于具有1e3行的1e5列的df:
Unit: milliseconds
expr min lq mean median uq max neval cld
alexis_laz(df) 356.1987 360.8647 366.2464 364.4488 368.9666 391.5828 10 a
David(df) 1387.1657 1415.7325 1530.0748 1436.9192 1542.1830 1968.9455 10 a
akrun(df) 1773.5728 1800.9288 1880.9201 1868.3143 1965.7862 2018.0870 10 a
GGrothendieck5(df) 4891.3247 5385.9903 8206.9116 9065.2893 9890.5795 10284.7369 10 b
GGrothendieck6(df) 5034.4408 9089.9334 9099.5746 9785.7042 10221.1537 11905.3997 10 b
GGrothendieck7(df) 5142.7372 9635.2558 9711.4691 9861.5164 10524.7317 11651.6198 10 b
PaulHiemstra(df) 5326.8807 9959.3951 10079.1672 10175.4814 11048.6218 12659.1130 10 b
By989(df) 9941.5236 10015.6652 10090.2076 10067.7127 10123.5885 10300.4110 10 b
GGrothendieck2(df) 25715.5451 25840.3138 26686.3386 26453.6770 26982.5627 29689.6019 10 c
GGrothendieck3(df) 26065.7005 26343.5734 27112.4387 26470.7166 27267.7166 31374.5133 10 c
GGrothendieck4(df) 25911.6476 26179.3999 27121.3442 26361.2242 27335.2762 31941.6339 10 c
GGrothendieck1(df) 34979.3212 35162.3589 36254.1681 35685.4975 36470.3027 41130.0531 10 d
源代码:
David <-function(df) {
NA^rowSums(is.na(df))
}
By989 <- function(df) {
rowSums(df) & rowSums(df, na.rm = T)
}
PaulHiemstra <- function(df) {
ifelse(apply(is.na(df), 1, any), NA, 1)
}
akrun <- function(df) {
NA^Reduce(`|`, lapply(df, is.na))
}
GGrothendieck1 <- function(df) {
rowSums(0*df) + 1
}
GGrothendieck2 <- function(df) {
max.col(0*df) + 1
}
GGrothendieck3 <- function(df) {
do.call(pmin, 0*df) + 1
}
GGrothendieck4 <- function(df) {
do.call(pmax, 0*df) + 1
}
GGrothendieck5 <- function(df) {
ifelse(apply(df, 1, anyNA), NA, 1)
}
GGrothendieck6 <- function(df) {
c(NA, 1)[apply(df, 1, anyNA) + 1]
}
GGrothendieck7 <- function(df) {
match(apply(df, 1, anyNA), FALSE)
}
alexis_laz <- function(df) {
complete.cases(df)
}
set.seed(5)
n<-function(x) sample(c(1:5,NA),1e6,replace=TRUE)
df<-data.frame(A=n(),B=n(),C=n())
results<-microbenchmark(David(df),
By989(df),
PaulHiemstra(df),
akrun(df),
GGrothendieck1(df),
GGrothendieck2(df),
GGrothendieck3(df),
GGrothendieck4(df),
GGrothendieck5(df),
GGrothendieck6(df),
GGrothendieck7(df),
alexis_laz(df),
times=10)
print(results,order="mean")