In hadley/ggstat: Statistical Computations for Visualisation

library(microbenchmark)
library(ggstat)
options(digits = 3)
knitr::opts_chunk$set(comment = "#>", collapse = TRUE)

Counting

For logicals, about 50x faster than table(), and twice as fast as tabulate() (but tabulate doesn't count missings)

x <- sample(c(T, F, NA), 1e4, rep = T)
microbenchmark(
  compute_count_vec(x),
  tabulate(as.numeric(x) + 1, 2),
  table(x)
)

A little slower than tabulate for factors, probably because of the extra work to deal with the two types of missing values in factors.

y  <- sample(factor(c(NA, letters)), 1e4, rep = T)
microbenchmark(
  compute_count_vec(y),
  tabulate(y, length(levels(y))),
  table(y)
)

About 50x faster than table when counting floats:

z <- sample(runif(100), 1e4, rep = T)
microbenchmark(
  compute_count_vec(z),
  table(z)
)

About 3x for strings:

a <- sample(replicate(100, paste(sample(letters), collapse = "")), 1e4, rep = T)
microbenchmark(
  compute_count_vec(a),
  table(a)
)

hadley/ggstat documentation built on May 17, 2019, 10:40 a.m.