使用R中的dplyr / magrittr删除/删除列

Question

How can I remove a column with dplyr/magrittr in R?

Here I want to delete columns which have more than 50% NAs (this does not work of course):

delNAcols <- function(x){ ifelse( mean(is.na(x))>0.5, NULL, x ) }
d <- data.frame(x=c(1,2,NA),y=c(NA,NA,4))
d %>% mutate_each(funs(delNAcols))

Solution

Both answers (from user3949008 and akrun) are good.

If the processing is at the beginning of the low, one could use a combination of both answers, due to the best balance of shortness and magrittr-style, in other words, best readability:

d %>%
  sapply(function(x) mean(is.na(x)) < 0.5) %>% 
  extract(d,. )

Because of the reuse of d this does not work if its later in the flow. Then user3949008's answer can be used after small change (and a small readability improvement):

d %>% select_(.dots = names(.)[which(sapply(., function(x) mean(is.na(x)) < 0.5))])

If one wants to have the whole thing even more concise, one can write

select_each <- function(df, fun) { df %>% sapply(fun) %>% extract(df,.) }
select_each <- function(df, fun) { df %>% select_(.dots = names(.)[which(sapply(., fun))]) }

d %>%
  select_each( function(x) mean(is.na(x)) < 0.5 )

with both select_each function being equal in functionality. However, I benchmarked them and the first one is three times as fast.

Answer 1

We can use base R

Filter(function(x) mean(is.na(x)) <= 0.5, d)

Or if we need to use the flow

library(dplyr)
library(magrittr)
d %>%
   summarise_each(funs(mean(is.na(.)) <= 0.5)) %>% 
   unlist %>% 
   extract(d,. )

Answer 2

This is one way to do it - using select_ (because we will be supplying names to select as character vector):

library(dplyr)
d <- data.frame(x = c(1,2,NA), y = c(NA,NA,4), z = c(1,2,3), a = c(NA,NA,2), b = c(1,NA,2))
select_(d, .dots = names(d)[which(sapply(d, function(x) mean(is.na(x)) < 0.5))])