📦 [archived] Installing and loading R packages

One reason of R popularity is an ocean of packages. Even though it is pretty straightforward to manage packages, there are a couple of tricks, do’s and don’ts, and other things which require a care.

Disclaimer: This post is outdated and was archived for back compatibility: please use with care! This post does not reflect the author’s current point of view and might deviate from the current best practices.

Installing

The standard built-in function works perfectly:

install.packages("ggplot2")

With several packages as well:

install.packages(c("ggplot2", "plyr"))

Sometimes if one needs to install particular version of a package, devtools package provides the following function:

# install.packages("devtools")
devtools::install_version(package = "ggplot2",
                          version = "2.1.0",
                          repos = "http://cran.us.r-project.org")

Finally, in order to see all installed packages I use more frequently library() instead of installed.packages(), which is an overkill in a way.

Loading and using

Two functions could be used for loading packages, namely require() and library(). The latter is a little bit more preferable, because it throws an error if the package does not exist. require() only returns FALSE and generates a warning.

The function argument package can be both either a character vector (must be of length 1) or a name. In other words, both expression will load package ggplot2:

library(ggplot2) # or
library("ggplot2")

The latter is possible thanks (or not thanks) to non-standard evaluation (NSE). I strongly recommend avoid NSE by using the one with quotes, which would save you from situations as follows (only God knows what kind of pervert would write this):

ggplo2 <- "pryr"
library(ggplot2)

There is also a nice trick how to load a several packages at once.

x <- c("ggplot2", "copula", "rvest")
lapply(x, library, character.only = TRUE)

To see all packages loaded, we typically use search().

One very important thing: do not use neither library() nor require() in packages’ scripts explicitly. Instead, specify all dependencies in DESCRIPTION file. Then, it is also best practice explicitly specify namespace of the desired package using the following syntax name_of_package::name_of_function(), e.g.:

copula::Bernoulli(2)

This one can be also done when the package installed but not loaded in non-package scripts.

Typically nobody gives a care about these small details. I also suggests to stick to one particular way of doing things, but also be aware of possible drawbacks.