Introduction to R

• Software for statistical computing
• Open source
• Official webpage CRAN
• Popularity in data analysis, machine learning, statistics, bioinformatics, …
• More than 5000 packages.

• In May 1976 at the Bell Labs, the first discussion about a software for statistical computing began.
• Called S, implemented mainly by Rick Becker and John Chambers, not free.
• In 1996 in New Zealand, Ross Ihaka and Robert Gentleman developed a system compatible with S, called R.

• Free - Big open source community
• Designed for data analysis/statistics/machine learning, econometrics, bioinformatics (Bioconductor), …
  • Has all basics in those fields
  • Has the majority of new developments.
• Wide range of applications, Task View. Among others: finance, social science, clinical trials…
• Efficient workflow (in contrast to Excel, SPSS, …).
• Flexible (in contrast to C++, Java, …).
• In constant development.

• R has a steep learning curve.
• Sometimes tricky:
  • When possible, use functions from the base package (they are compiled in C).
  • Use vectorization
  • Natural ways of programming are sometimes very inefficicent: avoid using for (use the apply family instead).
• Computational time. Interpreted language.
  

Overall

• Provide code examples.
• Show a global overview: what can be done with R.

This document

• Can be used as a quick reference.
• Provide external references: what to look for, where to look it.
• Done with RStudio.
• * means for people with experience programming.

Formal references

• John Chambers, “Software for Data Analysis: Programming with R” (classic book, basic R language).
• W.N. Venables and, B.D. Ripley “Modern Applied Statistics with S” (classic book, statistics with R).
• The R journal.
• The Journal of Satistical Software.

Blogs

• R-Bloggers.

Tutorials

• Quick R (R basics webpage).
• R Cookbook (R basics webpage).

Local R Communities

• Comunidad R Hispano
• RugBcn, Barcelona R Users Group.

Help

• R Inferno: If you are using R and you think you're in hell, this is a map for you.
• StackOverflow
• R-help, Q&A, developments
• R-help.es, Q&A in spanish, from Comunidad R Hispano.
• RugBcn googlegroups: rugbcn@googlegroups.com, from RugBcn

• RStudio (recommended)
• Eclipse
• ESS - Emacs Speaks Statistics
• Revolution Analytics commercial - intended for commercial applications and big data.
• “Windows type”: R Commander, DeduceR.

• Official packages can be found on CRAN.
• Very simple installation and loading: install.packages("ggplot2"), require(ggplot2).
• Wide range of areas: Task View.
• Making packages is very easy due to the recent developments of RStudio.
• Common documentation usually hard to read. If you are lucky you will find a vignette (how to use the package and some examples). For insatnce, data.table package.
• Recomended basic packages: ggplot2, reshape, plyr, lubridate, stringr (from Hadley Wickham)

• Main type in R: like a survey table.
• Columns may have different data type (characters, numerics, logical, …)
• When all columns have the same type we can use a matrix.

• simple manipulation example: LmExample.R
• base plots example: Plots.R

• ggplot2 package: nice plots in R, by Hadley Wickham. Example: GGPlot2.R
• Hadley Wickham “ggplot2: Elegant Graphics for Data Analysis” (book).
• ggmap = ggplot2 + RGoogleMaps. Example: GGMap.R

Facebook Formation of Love

• Packages for managing spatial data (shapefile format, projections, …): sp, maptools, rgdal.
• Google motion charts with package googleVis. Example: GoolgeVis.R.
• Interactive charts rCharts (using D3.js*). Examples.

More on visualization (including R)

• spatial.ly
• flowingdata

• Area between statistics, optimization and computer science.
• Main objective: learn from a historic database to make predictions. Example: MachineLearning.R
• Succesfully applied to (examples from a huge list):
  • Electricity market demand forecasting.
  • Image classification (i.e. number plate).
  • Spam filtering.
  • Recomender Systems (i.e. Amazon's books suggestions).
  • Natural language processing.
  • Fraud detection.
  • Medical diagnosis.

• Base functions are implemented in C or Fortran.
• Integration with Cpp through package RCpp (currently very active and well integrated in RStudio).
• Integration with Java through package rJava.
• R can be called from python using RPy or PypeR. The other way round with rPython.

• Parallel computing: two types in the base package
  • multicore (works in anything but Windows).
  • snow : opens several hidden R sessions and initializes each one with the required data. Wrapper package snowfall.
• Indexed structures with data.table(instead than dataframe).
• Byte code compilation with compiler.
• Use Rcpp.

• Integration with Hadoop.
• Dealing with files that do not fit in memory: ff package, bigmemory project.
• Direct access to data bases: RODBC, RMySQL, ROracle, RPostgreSQL, … packages (among others).
• Revolution Analytics
• SAP HANA integrates R.

• Latex was traditionally integrated with Sweave. Currently with knitr. Example: Latex.R
• RMarkdown (from knitr): html reporting (like this document).

• Web applications: RStudio's Shiny. Example.
• Web scraping: RCurl, XML i Relenium. Example.