In nutterb/pixiedust: Tables so Beautifully Fine-Tuned You Will Believe It's Magic

The process for comparing the speed of xtable and pixiedust will be a random sample of r format(1e4, big.mark = ",") rows from the mtcars dataset, with replacement. A table with this many rows is certainly at the fringe of the size of tables people my try to produce using either package and is large enough to give us some idea of how the two packages differ in terms of speed.

library(dplyr)
library(ggplot2)
library(microbenchmark)
library(stargazer)
library(xtable)
set.seed(100)
LargeTable <- mtcars[sample(1:nrow(mtcars), 1000, replace = TRUE), ]

The xtable times are calculated as follows:

Xtable <- microbenchmark(xtable = print.xtable(xtable(LargeTable, type = "html"), type = "html",
                                      print.results = FALSE),
                         times = 10, unit = "ms")

The stargazer times are calculated in a similar manner.

Stargazer <- microbenchmark(stargazer = {x <- capture.output(stargazer(LargeTable, 
                                                          type = "html", summary = FALSE))},
                            times = 10, unit = "ms")

The pixiedust times are calculated below. We apply the default background pattern just to add a little more complexity to the table. This should elaborate if adding more sprinkles adds to the processing time. Further investigation will be needed to determine if the time is added in the sprinkle function, or in the printing.

libs <- list.files("F:/pixiedust_library", full.names = TRUE)
lib_versions <- gsub("(pixiedust-|[.]tar[.]gz)", "", basename(libs))

for (i in seq_along(libs)){
  library(pixiedust, 
          lib.loc=file.path("F:/pixiedust_library", lib_versions[i]))

  Pixie <- microbenchmark({dust(LargeTable) %>% 
                             sprinkle_print_method("html") %>%
                             sprinkle(bg_pattern_by = "rows")},
                            times = 10, unit = "ms")
  Pixie$expr <- paste0("pixiedust ", lib_versions[i])


  assign(paste0("Pixie_", lib_versions[i]), Pixie)

  detach("package:pixiedust", unload=TRUE)
}
rm(list = c("Pixie", "lib_versions", "libs", "LargeTable", "i", "x"))

Compare <- bind_rows(mget(x = ls()))

ggplot(Compare,
       aes(x = expr, y = time)) + 
  geom_boxplot() + 
  theme(axis.text.x = element_text(angle = 90, hjust = 1))

Median <- Compare %>%
  group_by(expr) %>%
  summarise(median_time = median(time)) %>%
  mutate(relative_time = median_time / min(median_time)) %>%
  print()