有没有比这更好的方法
DT <- DT[,!apply(DT,2,function(x) all(is.na(x))), with = FALSE]
Run Code Online (Sandbox Code Playgroud)
仅在未完全用NAs填充的列上对数据表进行子集化?
谢谢
基本思想是使用以下内容找到所有NA列:
na_idx = sapply(DT, function(x) all(is.na(x)))
Run Code Online (Sandbox Code Playgroud)
要将其应用于对表进行子集化,答案取决于您是要从表中删除这些列,还是计划创建一个单独的派生表;
在前一种情况下,您应该将这些列设置为NULL:
DT[ , which(sapply(DT, function(x) all(is.na(x)))) := NULL]
Run Code Online (Sandbox Code Playgroud)
在后一种情况下,有几种选择:
idx = sapply(DT, function(x) !all(is.na(x)))
DT = DT[ , idx, with = FALSE] # or DT = DT[ , ..idx]
DT = DT[ , lapply(.SD, function(x) if (all(is.na(x))) NULL else x)]
Run Code Online (Sandbox Code Playgroud)
apply和colSums方法将涉及矩阵转换,这可能是低效的。
这是此处以及@DavidArenburg 在上述评论中列出的案例的基准:
method time
1: which := NULL 1.434
2: for set NULL 3.432
3: lapply(.SD) 16.041
4: ..idx 10.343
5: with FALSE 4.896
Run Code Online (Sandbox Code Playgroud)
代码:
library(data.table)
NN = 1e7
kk = 50
n_na = 5
set.seed(021349)
DT = setDT(replicate(kk, rnorm(NN), simplify = FALSE))
DT[ , (sample(kk, n_na)) := NA_real_]
DT2 = copy(DT)
t1 = system.time(
DT2[ , which(sapply(DT2, function(x) all(is.na(x)))) := NULL]
)
rm(DT2)
DT2 = copy(DT)
t2 = system.time({
for (col in copy(names(DT2)))
if (all(is.na(DT2[[col]]))) set(DT2, , col, NULL)
})
rm(DT2)
DT2 = copy(DT)
t3 = system.time({
DT3 = DT2[ , lapply(.SD, function(x) if (all(is.na(x))) NULL else x)]
})
rm(DT3)
t4 = system.time({
idx = sapply(DT2, function(x) !all(is.na(x)))
DT3 = DT2[ , ..idx]
})
rm(DT3)
t5 = system.time({
idx = sapply(DT2, function(x) !all(is.na(x)))
DT3 = DT2[ , idx, with = FALSE]
})
data.table(
method = c('which := NULL', 'for set NULL',
'lapply(.SD)', '..idx', 'with FALSE'),
time = sapply(list(t1, t2, t3, t4, t5), `[`, 'elapsed')
)
Run Code Online (Sandbox Code Playgroud)