is_na {cheapr}

Efficient functions for dealing with missing values.

Description

is_na() is a parallelised alternative to is.na().
num_na(x) is a faster and more efficient sum(is.na(x)).
which_na(x) is a more efficient which(is.na(x))
which_not_na(x) is a more efficient which(!is.na(x))
row_na_counts(x) is a more efficient rowSums(is.na(x))
row_all_na() returns a logical vector indicating which rows are empty and have only NA values.
row_any_na() returns a logical vector indicating which rows have at least 1 NA value.
The col_ variants are the same, but operate by-column.

Usage

is_na(x)

## Default S3 method:
is_na(x)

## S3 method for class 'POSIXlt'
is_na(x)

## S3 method for class 'vctrs_rcrd'
is_na(x)

## S3 method for class 'data.frame'
is_na(x)

num_na(x, recursive = TRUE)

which_na(x)

which_not_na(x)

any_na(x, recursive = TRUE)

all_na(x, recursive = TRUE)

row_na_counts(x, names = FALSE)

col_na_counts(x, names = FALSE)

row_all_na(x, names = FALSE)

col_all_na(x, names = FALSE)

row_any_na(x, names = FALSE)

col_any_na(x, names = FALSE)

Arguments

x

A vector, list, data frame or matrix.

recursive

Should the function be applied recursively to lists? The default is TRUE. Setting this to TRUE is actually much cheaper because when FALSE, the other NA functions rely on calling is_na(), therefore allocating a vector. This is so that alternative objects with is.na methods can be supported.

names

Should row/col names be added?

Details

These functions are designed primarily for programmers, to increase the speed and memory-efficiency of NA handling.
Most of these functions can be parallelised through options(cheapr.cores).

Common use-cases

To replicate complete.cases(x), use !row_any_na(x).
To find rows with any empty values, use which_(row_any_na(df)).
To find empty rows use which_(row_all_na(df)) or which_na(df). To drop empty rows use na_rm(df) or sset(df, which_(row_all_na(df), TRUE)).

is_na

is_na Is an S3 generic function. It will internally fall back on using is.na if it can't find a suitable method. Alternatively you can write your own is_na method. For example there is a method for vctrs_rcrd objects that simply converts it to a data frame and then calls row_all_na(). There is also a POSIXlt method for is_na that is much faster than is.na.

Lists

When x is a list, num_na, any_na and all_na will recursively search the list for NA values. If recursive = F then is_na() is used to find NA values.
is_na differs to is.na in 2 ways:

• List elements are counted as NA if either that value is NA, or if it's a list, then all values of that list are NA.

• When called on a data frame, it returns TRUE for empty rows that contain only NA values.

Value

Number or location of NA values.

Examples

library(cheapr)
library(bench)

x <- 1:10
x[c(1, 5, 10)] <- NA
num_na(x)
which_na(x)
which_not_na(x)

row_nas <- row_na_counts(airquality, names = TRUE)
col_nas <- col_na_counts(airquality, names = TRUE)
row_nas
col_nas

df <- sset(airquality, j = 1:2)

# Number of NAs in data
num_na(df)
# Which rows are empty?
row_na <- row_all_na(df)
sset(df, row_na)

# Removing the empty rows
sset(df, which_(row_na, invert = TRUE))
# Or
na_rm(df)
# Or
sset(df, row_na_counts(df) < ncol(df))