Drop rows with missing values in all columns [duplicate]

Question

It looks like tidyr's drop_na will drop rows if any of the specified columns contain missing values.

Example:

> library(tidyverse)
> df <- data.frame(a=c(1,NA,2,NA), b=c(3,4,NA,NA))
> df
   a  b
1  1  3
2 NA  4
3  2 NA
4 NA NA
> df %>% drop_na()
  a b
1 1 3

Is there a straightforward way to drop rows where all columns are missing?

pandas' dropna has a how argument that specify whether to look for any missing value or all missing value.

Answer 1

You can use filter

library(dplyr)
df %>% filter(!if_all(everything(), is.na)) 

or in base R:

df[rowSums(is.na(df)) != ncol(df), ]

Answer 2

I usually recommend collapse::na_omit for that. The oddly-specific prop argument will remove rows with a proportion of missing values >= prop. It's the fastest and most flexible option:

collapse::na_omit(df, prop = 1)

#    a  b
# 1  1  3
# 2 NA  4
# 3  2 NA

Answer 3

drop_na does not have that functionality but we can try to add it

library(dplyr)

df %>% drop_na(method = "all")
   a  b
1  1  3
2 NA  4
3  2 NA

The original function uses vctrs::vec_detect_complete. Adding vctrs::vec_detect_missing to get the extra functionality yields something like

drop_na <- function(data, ..., method = "any") {
  stopifnot(method == "any" | method == "all")
  dots <- enquos(...)

  if (rlang::is_empty(dots)) {
    # Use all columns if no `...` are supplied
    cols <- data
  } else {
    vars <- tidyselect::eval_select(expr(c(!!!dots)), data, allow_rename = FALSE)
    cols <- data[vars]
  }

  if (method == "any") {
    loc <- vctrs::vec_detect_complete(cols)
    out <- vctrs::vec_slice(data, loc)
  }
  if (method == "all") {
    loc <- !vctrs::vec_detect_missing(cols)
    out <- vctrs::vec_slice(data, loc)
  }
  out
}

A short version without modifying drop_na

library(dplyr)
library(vctrs)

df %>% filter(!vec_detect_missing(.))