Skip to content

R

R is an easy-to-learn high-level programming language especially tailored to data science applications.

Pros:

  • Easy to learn
  • High-level
  • Dynamically typed
  • Dynamic evaluation, enabling powerful functions and more readable specification of formulas and models
  • Very reliable ecosystem, compared to Python. Packages need to meet many requirements and demonstrate compatibility with dependencies.
  • High-quality data science packages
  • Large community and help sources
  • Rapid prototyping
  • Excellent glue language
  • Many packages interface with C or other fast languages to achieve performant code
  • Interoperable with many other languages
  • Reliable and standardized tooling
  • Fluent inferface: supports chaining functions, yielding much more readable analysis scripts.
  • Excellent notebook support
  • Native vector/matrix support in the language
  • Good documention: functions are generally well-documented and include code examples.

Cons:

  • Lack of proper namespaces: you are guaranteed to stumble into import order conflicts or naming conflicts.
  • Cryptic error messages: R's standard error messages are very confusing. The inconsistent/unreadable stack trace does not help either.
  • Confusing object-oriented systems: there are at least three systems for implementing object-oriented programming to some degree (S3, S4, R6). Enough said.
  • Inconsistent standard library: Quite a learning hurdle to get over.
  • Dot operator: increases the flexibility of functions, but documentation of options is often severely lacking.
  • Very slow: This is not a deal-breaker though for a glue/scripting language, and for prototyping purposes.
  • Dynamic evaluation: A powerful feature, but is very confusing to new users, and difficult to implement for advanced users.
  • Environments: generally confusing for users, and can lead to fun surprising (serializing a simple linear formula resulting in a 10 GB file).
  • Dynamically typed: great for rapid prototyping, not for large frameworks.

Sourcing

Running an R script

Action Code Details
Run R script file 
source('file.R')
Get sourced file from within script 
as.character(sys.call(1))[2]
Does not work during start-up!
Get sourced directory from within script 
dirname(as.character(sys.call(1))[2])

Options

Action Code Details
Enable debug (browse) on error 
options(error=recover)
Treat warnings as errors 
options(warn=2)
More concise (readable) traceback 
options(error=function() traceback(max.lines=3))
Disable scientific notation 
options(scipen=999)

Version checks

Action Code Details
Run code conditional on R version being later or equal than the given version number 
if (compareVersion(paste(version$major, version$minor, sep='.'), '3.6.0') >= 0)

Random number generation

Action Code Details
Use legacy RNG (R<3.6) for newer R versions 
if (compareVersion(
    paste(version$major, version$minor, sep='.'),
    '3.6.0') >= 0
) {
    RNGkind(sample.kind='Rounding')
}
A fix to reproduce RNG of R 3.5.0, for reproducing old results

Output

Action Code Details
Output object 
cat(x)
Output string 
cat('hello')
Print object 
print(x)
Print string 
print('hello')
Show message 
message('msg')
Show warning 
warning('warning')
Show warning right now 
warning('warning', .immediate=TRUE)
Trigger and show error 
stop('this is serious')
Redirect output to file 
sink('log.txt')
Use sink() to restore
Capture any output as string 
txt = capture.output({...})
Suppress all output 
capture.output({...})
Suppress automatic printing of object in interactive mode 
invisible(x)
Suppress all output 
sink('nul:')
'/dev/null' on linux?
Suppress messages 
suppressMessages({...})
Suppress package messages on load 
suppressPackageStartupMessages({...})
Suppress warnings 
suppressWarnings({...})