In tdsmith/HomebrewGA: Analytics reports for Homebrew

knitr::opts_chunk$set(echo=FALSE, message=FALSE, warning=FALSE)
library(dplyr)
library(ggplot2)
library(magrittr)
library(reshape2)
library(viridis)

library(HomebrewGA)
invocations = fetch_install_invocations()
popularity = popularity_by_package(invocations)
opt_df = install_invocations_by_package(invocations)

Top 100 packages

Total install invocations for top 100 packages, with and without formula options given; bar length indicates number of installs:

x = function() {
  sort_order = popularity %>% arrange(desc(n)) %>% head(100)
  popular = opt_df %>% filter(package %in% sort_order$package)
  popular %<>% inner_join(popularity) %>%
    filter(option == "none") %>%
    rename(without_option=count) %>%
    mutate(with_option=n-without_option) %>%
    melt(measure.vars=c("with_option", "without_option"),
         variable.name="option_given",
         value.name="count") %>%
    mutate(option_given=relevel(option_given, "without_option"),
           package = factor(package, levels=rev(sort_order$package))) %>%
    arrange(option_given)
  g = ggplot(popular, aes(package, count, fill=option_given)) +
    geom_bar(stat="identity") +
    coord_flip() +
    theme_bw()
  print(g)
}
x()

There are a few packages that are often installed with options; many of them offer --universal or multiple graphics backends.

Top 100 packages, relative proportion

Relative proportion of install invocations with formula options given, for the top 100 packages; bar length indicates percentage of installs of each package:

x = function() {
  sort_order = popularity %>% arrange(desc(n)) %>% head(100)
  popular = opt_df %>% filter(package %in% sort_order$package)
  popular %<>%
    inner_join(popularity) %>%
    filter(option == "none") %>%
    rename(without_option=count) %>%
    mutate(without_option=without_option/n, with_option=1-without_option) %>%
    melt(measure.vars=c("with_option", "without_option"),
         variable.name="option_given",
         value.name="count") %>%
    mutate(option_given=relevel(option_given, "without_option"),
           package = factor(package, levels=rev(sort_order$package))) %>%
    arrange(option_given)

  g = ggplot(popular, aes(package, count, fill=option_given)) +
    geom_bar(stat="identity") +
    scale_y_continuous(labels=scales::percent) +
    theme_bw() +
    coord_flip()
  print(g)
}
x()

Fraction of installs with options vs. popularity

More popular packages are less likely to have many popular options. I would guess this is because more popular packages are less likely to offer options, but I haven't checked.

x = function() {
  options = opt_df %>%
    inner_join(popularity) %>%
    transform(fraction=count/n) %>%
    filter(option == "none") %>%
    mutate(fraction=1-fraction)
  g = ggplot(options, aes(n, fraction)) +
    geom_point(alpha=0.25) +
    geom_density2d(h=c(1,0.1)) +
    scale_x_log10() +
    labs(x="Total package installs", y="Fraction of installs per package with options") +
    theme_bw()
  print(g)
}
x()

The blue density contours show that nearly all packages are relatively unpopular and have very few installs with options, which is otherwise hard to see because of overplotting.

Fraction of installs without options, by package

This shows that most packages are never installed with options. A handful of packages are very commonly installed with options.

x = function() {
  opt_df %<>%
    inner_join(popularity) %>%
    filter(option == "none") %>%
    mutate(default_fraction=count/n)
  g = ggplot(opt_df, aes(default_fraction)) +
    geom_histogram(binwidth=0.05) +
    scale_y_log10(breaks=c(1, 3, 10, 30, 100, 300, 1000, 3000, 10000)) +
    labs(x="Fraction of installs without options", y="Number of packages") +
    theme_bw()
  print(g)
}
x()

Similarly, packages often have a small number of absolute installations with options.

x = function() {
  g = ggplot(opt_df %>% filter(option != "none"), aes(count, ..count..)) +
    geom_freqpoly() +
    scale_y_log10(breaks=c(1, 3, 10, 30, 100, 300, 1000, 3000, 10000)) +
    labs(x="Number of installs with options, per package", y="Number of packages") +
    theme_bw()
  print(g)
}
x()

Packages with popular options (top 100)

The number on the left hand side indicates total popularity.

x = function() {
  overall_popularity = popularity %>% rename(overall=n)
  merged = opt_df %>%
    filter(option == "none") %>%
    inner_join(overall_popularity) %>%
    filter(overall >= 100) %>%
    mutate(options_popularity=1-count/overall) %>%
    arrange(desc(options_popularity)) %>%
    head(100)

  merged$package %<>% factor(levels=rev(merged$package))
  g = ggplot(merged, aes(options_popularity, package)) +
    geom_point(aes(color=overall)) +
    xlim(0, 1) +
    geom_text(aes(size=overall, label=overall), x=0.1) +
    scale_color_viridis(trans="log10", guide="none") +
    scale_size_continuous(trans="log10", guide="none") +
    theme_bw()
  print(g)
}
x()