Essential R

🚀 Essential tools and libraries for programming in R

About

This document serves as a personal list of

• tools for package development,
• good practices for programming,
• and most frequently used packages

with an emphasis on data science, Bayesian stats and probabilistic ML.

R package development

A minimum R-package stack at least consists of the following packages of tools:

• goodpractice for advice on writing R packages
• devtools for general package development
• testthat for unit testing
• roxygen2 for method documentation
• covr to generate coverage reports
• lintr for static code analysis
• styler to automatically format code
• usethis for automation of repetitive tasks during development
• remotes to install R packages from Git repositories, CRAN and Bioconductor
• pkgdown to generate websites of your package
• rcmdcheck to check your package within R
• profvis to visualize profiling data
• bench to time R expressions
• microbenchmark to also time R expressions
• lobstr to pry open R
• here to find files and folders

Many more can be find on github at r-lib or rstudio.

Programming

• purrr for functional programming
• magrittr for pipeing function calls
• R6 for object-oriented programming with encapsulation
• Rcpp (with RcppArmadillo, RcppEigen and BH) for integration of C++ code
• reticulate for interfacing to Python
• cpp11 as alternative (and complement) to Rcpp
• compiler, doParallel, parallel, foreach for speeding up R
• rlang as low-level API for programming in R

C++

• meson as modern build system and autotools since R is old
• VTune and the Parallel Studio to profile code
• Callgrind as an alternative to the above
• gdb and lldb for debugging C++
• sanitizers to detect memory leaks, etc.
• cppcheck for static code analysis
• cpplint to check C++ style
• clang-format to format C++ files
• doxygen for code documentation
• boost for unit tests, data structures and basically everything you ever need

Furthermore, some good reading:

• Scott Meyers: Effective C++
• Scott Meyers: Effective Modern C++
• Scott Meyers: Effective STL
• Kurt Guntheroth: Optimized C++
• David Vandevoorde: C++ Templates - the compete guide
• Nicolai Josuttis: C++17 - the complete guide

Working with data

• tidyverse (dplyr, tidyr, ...) for working with data in general
• data.table as a fast alternative to R`s native data frame
• datastructures for advanced data structures
• dbplyr to work with data bases
• RSQLite to work with SQLite data bases

Machine learning and Bayesian statistics

• rstan to fit Bayesian models
• cmdstanr as light-weight interface to cmdstan
• rstanarm for applied regression modeling
• brms for multilevel models
• bayesplot to visualize Bayesian inferences
• loo for approximate LOO-CV and PSIS
• projpred for projection predictive variable selection
• rstantools for developing R Packages interfacing with Stan
• posterior for working with output from Bayesian models
• coda for summarizing and working with MCMC output
• MCMCpack provides some utility functions
• LaplacesDemon for even more Bayes utility
• mgcv for generalized additive mixed models
• tensorflow for numerical computation
• tfprobability for statistical computation and probabilistic modeling
• keras to work with neural networks
• sparklyr for big data processing
• kernlab for kernel-based machine learning

Graphical models and causal inference

• bnlearn for BN structure learning
• pcalg for causal inference using graphical models
• ggdag for visualizing DAGs
• dagitty for analysis of structural equation models

Optimization

• nloptr for non-linear optimization
• cvxr for disciplined convec optimization

Visualization

• ggplot2 as base package for visualization
• paletteer for all the palettes
• vapoRwave for even more palettes
• gpubr for publication ready visualization
• ggthemes for additional themes and scales for ggplot2
• hrbrthemes for even more themes
• ggsci for sci-fi themes
• ggthemr for even mroe themes
• colourlovers for access to the COLOURlovers API
• patchwork to easily compose plots
• colorspace for color manipulation
• cowplot for plot annotations
• scales and swatches
• plotly for interactive plots
• ggraph to visualize graphs
• gganimate to animate plots
• DiagrammeR for diagrammes, graphs and networks
• highcharter as alternative to plotly
• tidybayes for geoms for Bayesian models
• scico for more color palettes
• ggnetwork for geoms for networks
• tweenr to interpolate data
• magick to work with images
• imager as complement to magick
• r2d3 to interface to d3

Reporting

• shiny for interactive web applications
• rmarkdown to generate websites, pdf documents, etc from markdown files
• bookdown for authoring books
• tufte for Tufte-style documents
• xaringan for HTML presentations

Other

• igraph to work with graphs
• drake for building pipelines

Author

Simon Dirmeier simon.dirmeier@web.de

License

This work by Simon Dirmeier is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.