dplyr_filter_out

NoteAdded in 0.5.0

What it does

Checks dplyr::filter() calls with complex conditions and suggests using dplyr::filter_out() instead.

Why is this bad?

Using filter() with negated conditions can be hard to read, especially when we also want to retain missing values. filter(!condition) drops rows where condition is TRUE and rows where it is NA, meaning that if we want to retain those then we have to complement the condition with is.na():

# We want to drop rows whose value for `col` is larger than the average
# of `col`:
larger_than_average <- function(x) x > mean(x, na.rm = TRUE)
x |> filter(!larger_than_average(col) | is.na(larger_than_average(col)))

dplyr 1.2.0 introduced filter_out() as a complement to filter(). filter_out() drops rows that match the condition, meaning that rows where the condition is NA are retained. We can then rewrite the code above like this:

x |> filter_out(larger_than_average(col))

This rule suggests an automatic fix to rewrite them with filter_out(). It is only valid for dplyr >= 1.2.0, and only works on filter() calls where all conditions are made of one expression + is.na() on the same column.

Example

library(dplyr)
x <- tibble(a = c(1, 2, 2, NA), b = c(1, 1, 2, 3))

x |> filter(a > 1 | is.na(a))

x |> filter(a > 1 | is.na(a), is.na(b) | b <= 2)

Use instead:

library(dplyr)
x <- tibble(a = c(1, 2, 2, NA), b = c(1, 1, 2, 3))

x |> filter_out(a <= 1)

x |> filter_out(a <= 1 | b > 2)

References