The package you want may not be in your R installation. One of the big advantages of open source software is that people love to share. People all over the world have written their own special-purpose R packages, placing them in the CRAN repository and elsewhere.
User contributions to CRAN go through a vetting process and are generally of high quality. They are, however, not tested as throughly as R itself.
One way to install a package is to use the install_packages() function. For example, suppose you wish to use the mvtnorm package, which computes multivariate normal cumulative distribution functions and other quantities. First, choose a directory in which you wish to install the package (and maybe others in the future), say /a/b/c. Then at the R prompt, type this:
> install.packages("mvtnorm","/a/b/c/")This will cause R to automatically go to CRAN, download the package, compile it, and load it into a new directory: /a/b/c/mvtnorm.
You do need to tell R where to find that package once it’s installed, which you can do via the .libPaths() function:
> .libPaths("/a/b/c/")This will add that new directory to the ones R was already using. If you use that directory often enough, you may wish to add that call to .libPaths() in your .Rprofile startup file in your home directory.
A call to .libPaths() without an argument will show you a list of all the places R will currently look for loading a package when requested.
Sometimes you need to install “by hand” to make modifications required to make a particular R package work on your system. The following example demonstrates how I did so in one particular instance, and it will serve as a case study on handling situations in which ordinary methods don’t work.
Situations in which you need to install packages by hand typically are operating system dependent and require more computer expertise than is generally assumed in this book. For help in very specific cases, the r-help mailing list is invaluable. To access it, go to the R home page (http://www.r-project.org/), click the FAQs link and then the R FAQ link, and scroll down to Section 2.9, “What mailing lists exist for R?”
I wanted to install the Rmpi package on our department’s instructional machines in the directory /home/matloff/R. I tried using install.packages() first but found that the automated process could not find the MPI library on our machines. The problem was that R was looking for those files in /usr/local/lam, whereas I knew they were in /usr/local/LAM. Since these were public machines, not my own, I did not have the authority to change the name. So, I downloaded the Rmpi files in the packed form Rmpi_0.5-3.tar.gz. I unpacked that file in my directory ~/tmp, producing a directory named ~/tmp/Rmpi.
If I had not experienced this problem, at this point, I could have just typed the following in a terminal window from within the ~/tmp directory:
R CMD INSTALL -l /home/matloff/R Rmpi
That command would install the package contained in ~/tmp/Rmpi, placing it in /home/matloff/R. This would have been an alternative to calling install.packages().
But as noted, I had to deal with a problem. Within the ~/tmp/Rmpi directory, there was a configure file, so I ran this command on my Linux command line:
configure --help
It told me that I could specify the location of my MPI files to configure, as follows:
configure --with-mpi=/usr/local/LAM
This applies if you run configure directly, but I ran it via R:
R CMD INSTALL -l /home/matloff/R Rmpi --configure-args=--with-mpi=/usr/local/LAM
Well, that seemed to work, in the sense that R did install the package, but R also noted that it had a problem with the threads library on our machines. Sure enough, when I tried to load Rmpi, I got a runtime error, saying that a certain threads function wasn’t there.
I knew that our threads library was fine, so I went into the configure file and commented out two lines:
# if test $ac_cv_lib_pthread_main = yes; then
MPI_LIBS="$MPI_LIBS -lpthread"
# fiIn other words, I forced it to use what I knew (or was fairly sure) would work. I then reran R CMD INSTALL, and the package loaded without any problems.