I started using R about 3 years ago. It was slow going at first. R had tricky and less intuitive syntax than languages I was used to, and it took a while to get accustomed to the nuances. It wasn't immediately clear to me that the power of the language was bound up with the community and the diverse packages available.
R can be more prickly and obscure than other languages like Python or Java. The good news is that there are tons of packages which provide simple and familiar interfaces on top of Base R. This post is about ten packages I love and use everyday and ones I wish I knew about earlier.
One of the steepest parts of the R learning curve is the syntax. It took me a while to get over using
=. I hear people say a lot of times How do I just do a
VLOOKUP?!? R is great for general data munging tasks, but it takes a while to master. I think it's safe to say that
sqldfwas my R "training wheels".
sqldflet's you perform SQL queries on your R data frames. People coming over from SAS will find it very familiar and anyone with basic SQL skills will have no trouble using it--
sqldfuses SQLite syntax.
I don't do time series analysis very often, but when I do
forecastis my library of choice.
forecastmakes it incredibly easy to fit time series models like ARIMA, ARMA, AR, Exponential Smoothing, etc.
My favorite feature is the resulting forecast plot.
When I first started using R, I was using basic control operations for manipulating data (for, if, while, etc.). I quickly learned that this was an amateur move, and that there was a better way to do it.
In R, the apply family of functions is the preferred way to call a function on each element of a list or vector. While Base R has this out of the box, its usage can be tricky to master. I've found the
plyrpackage to be an easy to use substitute for
combinefunctionality in Base R.
plyrgives you several functions (
ldply) following a common blueprint: Split a data structure into groups, apply a function on each group, return the results in a data structure.
ddplysplits a data frame and returns a data frame (hence the dd).
daplysplits a data frame and results an array (hence the da). Hopefully you're getting the idea here.
I find base R's string functionality to be extremely difficult and cumbersome to use. Another package written by Hadley Wickham,
stringr, provides some much needed string operators in R. Many of the functions use data structures that aren't commonly used when doing basic analysis.
stringris remarkably easy to use. Nearly all of the functions (and all of the important ones) are prefixed with "str" so they're very easy to remember.
install.packages("RPostgreSQL") install.packages("RMySQL") install.packages("RMongo") install.packages("RODBC") install.packages("RSQLite")
Everyone does it when they first start (myself included). You've just written an awesome query in your preferred SQL editor. Everything is perfect - the column names are all snake case, the dates have the right datatype, you finally debugged the
"must appear in the GROUP BY clause or be used in an aggregate function"issue. You're ready to do some analysis in R, so you run the query in your SQL editor, copy the results to a csv (or...God forbid... .xlsx) and read into R. You don't have to do this!
R has great drivers for nearly every conceivable database. On the off chance you're using a database which doesn't have a standalone driver (SQL Server), you can always use
Next time you've got that perfect query written, just paste it into R and execute it using
RODBC. In addition to preventing you from having tens of hundreds of CSV files sitting around, running the query in R saves you time both in I/O but also in converting datatypes. Dates, times, and datetimes will be automatically set to their R equivalent. It also makes your R script reproducible, so you or someone else on your team can easily produce the same results.
lubridateis one of those magical libraries that just seems to do exactly what you expect it to. The functions all have obvious names like
Here's a really handy reference card that I found in a paper. It covers just about everything you might conceivably want to do to a date. I've also found this Date Cheat Sheet to be a handy reference.
Another Hadley Wickham pacakge and probably his most widely known one.
ggplot2ranks high on everyone's list of favorite R packages. It's easy to use and it produces some great looking plots. It's a great way to present your work, and there are many resources available to help you get started.
- Elegant Graphics for Data Analysis by Hadley Wickham (Amazon)
- A Rosetta Stone for Excel to
- Hadley Wickham ggplot2 Presentation at Google (youtube)
- R Graphics Cookbook by Winston Chang (Amazon)
qccis a library for statistical quality control. Back in the 1950s, the now defunct Western Electric Company was looking for a better way to detect problems with telephone and electrical lines. They came up with a set of rules to help them identify problematic lines. The rules look at the historical mean of a series of datapoints and based on the standard deviation, the rules help judge whether a new set of points is experiencing a mean shift.
The classic example is monitoring a machine that produces lug nuts. Let's say the machine is supposed to produce 2.5 inch long lug nuts. We measure a series of lug nuts: 2.48, 2.47, 2.51, 2.52, 2.54, 2.42, 2.52, 2.58, 2.51. Is the machine broken? Well it's hard to tell, but the Western Electric Rules can help.
While you might not be monitoring telephone lines,
qcccan help you monitor transaction volumes, visitors or logins on your website, database operations, and lots of other processes.
I always find that the hardest part of any sort of analysis is getting the data into the right format.
reshape2is yet another package by Hadley Wickham that specializes in converting data from wide to long format and vice versa. I use it all the time in conjunction with
It's a great way to quickly take a look at a dataset and get your bearings. You can use the
meltfunction to convert wide data to long data, and
dcastto go from long to wide.
This list wouldn't be complete without including at least one machine learning package you can impress your friends with. Random Forest is a great algorithm to start with. It's easy to use, can do supervised or unsupervised learning, it can be used with many different types of datasets, but most importantly it's effective! Here's how it works in R.
- Learn R Blog
- Learn R via Python
- Tidy Data by Hadley Wickham
- Intro to Plyr on Slideshare
- R Overview
- Example R graphics
- Hadley Wickham's R programming style guide
- Google's R Programming Style Guide
- R Bloggers
- Must have R Packages for Social Scientists by Drew Conway