So why do I assume you haven’t heard of them?

Well, they’re hardly ever spoken about or searched for! Hadley Wickham is a huge promoter of the benefits of visualisation in data science² but has only ever mentioned htmlwidgets on Twitter explicitly 3 times prior to February 2018. And just compare the Google Search Trends for tidyverse and htmlwidgets, there’s desperately little love for htmlwidgets.

The crux of the problem really is that htmlwidgets is a developer tool. The actual packages that folks build using it are really popular! htmlwidgets spiritual home is htmlwidgets.org which is maintained by the RStudio folks, where they keep a list of featured htmlwidgets libraries available oin CRAN. Here’s a quick overview of these libraries (as of February 2018), with the following notes:

• so.tag contains the tag(s) used on StackOverflow for questions relating to this package
• mention.htmlwidgets is TRUE if the website for the package clearly mentions in simple language that it is a htmlwidgets library (or the outputs of the package are htmlwidgets)
• rglwidget was dropped from the list because it’s changed to rgl but this has not been reflected in the htmlwidgets.org website.

It’s really easy to get download figures for CRAN using the cranlogs library. In the table below I’ve collected the daily download figures for the featured_htmlwidget_libraries from the first date htmlwidgets was published until 2020-05-15 and then calculated the median daily downloads.

library("versions")
## Get first release date of htmlwidgets
## This takes much longer than I'd like to run, so I've cheated
# vhistory_htmlwidgets <- versions::available.versions("htmlwidgets")
# release_date_htlmlwidgets <- vhistory_htmlwidgets$htmlwidgets %>%
#   filter(date == min(date)) %>%
#   select(date) %>%
#   .[[1]]
release_date_htlmlwidgets <- lubridate::ymd("2014-12-09")

library("cranlogs")
cran_downloads <- featured_htmlwidget_libraries %>%
  select(package) %>%
  pmap(function(package, ...) {
    cran_downloads(packages = package, from = "2014-12-09", to = Sys.Date())
  }) %>%
  bind_rows() %>%
  as_tibble()

median_daily_downloads <- cran_downloads %>%
  filter(count > 0) %>%
  filter(package != "htmlwidgets") %>%
  group_by(package) %>%
  summarise(median.downloads = median(count)) %>%
  arrange(desc(median.downloads))

dt_median_daily_downloads <- median_daily_downloads %>%
  datatable(
    options = list(dom = "t", pageLength = 13),
    colnames = c("htmlwidget library", "Median daily downloads (excluding zero download days)")
  )
frameWidget(dt_median_daily_downloads)

The top 6 libraries all have an median of more than 100 daily downloads, is that impressive? Well, I don’t have any feel for that unfortunately. The cranlogs tells us about top 100 downloaded packages, but these download figures are going to suffer from a highly positively skewed distribution. It would be fun to investigate this distribution and come back to this question in the future. All I want to show here is that the htmlwidgets libraries are really popular, but that there’s little exposure for the htmlwidgets branding.

I can demonstrate the growing popularity of the libraries with a timeseries chart using dygraphs, I’ve chosen to perform a simple 28-day moving average on the data as there is a lot of inter-day noise:

library("dygraphs")
## CSS hackery to place the legend nicely and account for the yellow colour in the "Paired"
## colour scheme
dygraph_css <- tempfile(fileext = ".css")
dygraph_css_file <- file(dygraph_css)
writeLines(".dygraph-title {color: navy;font-weight: bold;}
.dygraph-axis-label {font-weight: bold;}
.dygraph-legend {background: #ebebeb !important;left: 100px !important;top: 100px !important;}
.dygraph {background: #ebebeb !important;}", dygraph_css_file)
close(dygraph_css_file)
 
cran_downloads_28ma <- cran_downloads %>%
  filter(package != "htmlwidgets") %>%
  group_by(package) %>%
  mutate(mean.count = rollmean(
    count, k = 28, na.pad = TRUE,
    align = "right"
  )) %>%
  ungroup() %>%
  replace_na(list(mean.count = 0)) %>%
  select(-count)

cran_downloads_28ma %>%
  mutate(package = fct_relevel(package, median_daily_downloads$package)) %>%
  spread(package, mean.count) %>%
  # filter(date > ymd("2017-08-01")) %>%
  timetk::tk_xts(date_var = date) %>%
  dygraph(main = "Featured htmlwidgets library downloads from CRAN<br> <span style='font-size:18px;'>(28 day moving average)</span>",
          width = "100%") %>%
  dyCSS(dygraph_css) %>%
  dyOptions(colors = RColorBrewer::brewer.pal(12, "Paired")) %>%
  dyAxis("y", label = "CRAN Downloads") %>%
  dyLegend(labelsSeparateLines = TRUE)

I hoped to get a good idea on Twitter about how many folks were using htmlwidgets libraries without knowing that’s what they were. Unfortuntely, the reach of my poll was very small with the tweet only appearing in a timeline 307 times. I’d really like to collect more data about this and return to this topic in the future.

#rstats #dataviz folks! Do you use any of these libraries:

leaflet, dygraph, plotly, rbokeh, highcharter, visNetwork, or DT?

If so, did you know these are examples of htmlwidgets?

(See https://t.co/DtVNBSFJZA if you didn't know!)

— Charlie Joey Hadley (@charliejhadley) February 5, 2018