In r-hub/pkgsearch: Search and Query CRAN R Packages

knitr::opts_chunk$set(
  comment = "#>",
  tidy = FALSE,
  error = FALSE,
  fig.width = 8,
  fig.height = 8)
options(width = 90)
options(max.print = 200)

Installation

Install the latest pkgsearch release from CRAN:

install.packages("pkgsearch")

The development version is on GitHub:

pak::pak("r-hub/pkgsearch")

Usage

Search relevant packages

Do you need to find packages solving a particular problem, e.g. "permutation test"?

library("pkgsearch")
library("pillar") # nicer data frame printing
pkg_search("permutation test")

pkgsearch uses an R-hub web service and a careful ranking that puts popular packages before less frequently used ones.

Do it all clicking

For the search mentioned above, and other points of entry to CRAN metadata, you can use pkgsearch RStudio add-in!

Select the "CRAN package search" addin from the menu, or start it with pkg_search_addin().

Get package metadata

Do you want to find the dependencies the first versions of testthat had and when each of these versions was released?

cran_package_history("testthat")

Discover packages

Do you want to know what packages are trending on CRAN these days? pkgsearch can help!

cran_trending()
cran_top_downloaded()

Keep up with CRAN

Are you curious about the latest releases or archivals?

cran_events()

Search features

More details

By default it returns a short summary of the ten best search hits. Their details can be printed by using the format = "long" option of pkg_search(), or just calling pkg_search() again, without any arguments, after a search:

library(pkgsearch)
pkg_search("C++")

pkg_search()

Pagination

The more() function can be used to display the next batch of search hits, batches contain ten packages by default. ps() is a shorter alias to pkg_search():

ps("google")

more()

Stemming

The search server uses the stems of the words in the indexed metadata, and the search phrase. This means that "colour" and "colours" deliver the exact same result. So do "coloring", "colored", etc. (Unless one is happen to be an exact package name or match another non-stemmed field.)

ps("colour", size = 3)
ps("colours", size = 3)

Ranking

The most important feature of a search engine is the ranking of the results. The best results should be listed first. pkgsearch uses weighted scoring, where a match in the package title gets a higher score than a match in the package description. It also uses the number of reverse dependencies and the number of downloads to weight the scores:

ps("colour")[, c("score", "package", "revdeps", "downloads_last_month")]

Preferring Phrases

The search engine prefers matching whole phrases over single words. E.g. the search phrase "permutation test" will rank coin higher than testthat, even though testthat is a much better result for the single word "test". (In fact, at the time of writing testthat is not even on the first page of results.)

ps("permutation test")

If the whole phrase does not match, pkgsearch falls back to individual matching words. For example, a match from either words is enough here, to get on the first page of results:

ps("test http")

British vs American English

The search engine uses a dictionary to make sure that package metadata and queries given in British and American English yield the same results. E.g. note the spelling of colour/color in the results:

ps("colour")
ps("color")

Ascii Folding

Especially when searching for package maintainer names, it is convenient to use the corresponding ASCII letters for non-ASCII characters in search phrases. E.g. the following two queries yield the same results. Note that case is also ignored.

ps("gabor", size = 5)
ps("Gábor", size = 5)

Configuration

Options

• timeout: pkgsearch follows the timeout options for HTTP requests (i.e. for pkg_search() and advanced_search(). timeout is the limit for the total time of the HTTP request, and it is in seconds. See ?options for details.

More info

See the complete documentation.

License

MIT @ Gábor Csárdi, RStudio, R Consortium.

r-hub/pkgsearch documentation built on Sept. 28, 2024, 10:55 p.m.