Page to collect and organise different ideas. Expect random musings and code snippets.

© 2017. All rights reserved.

A sensible .Rprofile

20 Dec 2016

Here are some settings that I found useful for working with R. Note that I’m mainly using basic R scripts in an HPC environment and not Rstudio most of the time.

TL;DR: have a look at the whole file.

Custom library path

R is not hard to install, but there is a couple of compile-time configuration steps to take care of:

For these reasons, it will make sense to have a central install on an HPC cluster to get the details right. However, that comes with some challenges.

In particular, the separation of user-readable system packages and user-writable personal packages does not always work.

This is why I found it easiest to specify an explicit library path.

.libPaths("~/.R")

This way, R will ignore the system-provided packages and install all required packages in the above directory where you have got full write access.

Custom options

Here are some tweaks about converting strings to factors, selecting a default CRAN mirror, and warn about partial matching of subsetting and function arguments.

options(
    stringsAsFactors = FALSE,

    menu.graphics = FALSE,
    repos = c(CRAN="http://mirrors.ebi.ac.uk/CRAN/"),

    warnPartialMatchAttr = TRUE,
    warnPartialMatchDollar = TRUE
)

There is some controversy about how R is supposed to handle strings. Convert characters to factors by default or not? I’ve been a strong proponent to disabling converting, for which I’ve been accused to not understand how they work. True, code is less portable by setting a non-default option, but factors mess up enough that I don’t want to deal with it.

Another tweak is to not display the graphical menu to select a mirror when downloading from CRAN (this just delays what I actually wanted to do) and select a default mirror.

A third is to dispaly a warning when I either try to pass a named function argument that does not exactly match what the function is expecting (helps not mixing up arguments with confusing errors after) or when subsetting a list or data.frame using the $ sign.

Traceback on non-interactive scripts

When scripts run non-interactively and they fail, it is often challenging to find the source of the error or the function call in which R failed, as only the error message is reported and no additional context by default.

Using the below lines, you can get the traceback on errors when scripts fail that will not impact your interactive sessions.

if (!interactive())
    options(error = function() {
        traceback(2, max.lines=10)
        quit(save="no", status=1)
    })

The guard to exit R after printing the traceback is essential, because otherwise it will happily try to execute the rest of the script with an undefined state.

Related Posts