extRemes web tutorial: Details

Trouble Shooting
Is it Really Necessary to Give a Path to the library Command Every Time?
Software Requirements
The Underlying Functions
Miscellaneous

Details

Trouble Shooting

If the main toolkit dialog does not appear on startup or it appears, but many of the functions do not work, then check the following possible causes.

It may be that R does not know where the extRemes library is located. R assumes that all libraries (packages) are in the same place. Often, however, a user may wish to have a package somewhere else; for example, a unix user who does not have root privileges cannot install packages in the location where R checks for them. In this case, it is necessary to tell R where the package is located. If, for example, the package is installed in the directory, /home/[user]/src/library, then the toolkit must be loaded into R using the following command.

library( extRemes, lib.loc="/home/[user]/library/")
Another possible cause for the dialog to not appear is that this toolkit depends on the R package tcltk, which interfaces with the Tcl/Tk programming language. The Tcl/Tk programming language must also be installed on your system and R must know where to find it in order for the toolkit to work. Please see Software Requirements for more information on obtaining, installing and pointing R to Tcl/Tk.
If you receive an error message that says,

Error in eval(expr, envir, enclos) : Object "gev.diag" not found

then the package ismev is not loaded. In order to load this package from the R prompt, simply type:

library( ismev)

Or, if it is installed in a library where R does not know to look, such as /home/[user]/src/library, then type:

library( ismev, lib.loc="/home/[user]/src/library")
Of course, this package will not load if it has not been installed. If it is not installed, then it can be installed from the R prompt by the command:

install.packages( "ismev")

If this does not work, then it is likely that you do not have permission to write to the file where R wants to install packages. If this is the case, then it is possible to tell R to put it someplace else. For example, to install ismev in the directory /home/[user]/src/library, use the command:

install.packages( "ismev", lib="/home/[user]/src/library")

Once ismev is installed on your system, it needs to be loaded into R (see above).

Is it Really Necessary to Give a Path to the `library` Command Every Time?

On the unix and linux platforms, if you do not have root privileges, then you will have had to type:

library( extRemes, lib.loc="[path to extRemes library]")

every time you want to load the extRemes library. Similarly for any other R package, like ismev, that you install onto your own space. It is possible to set up a file called .Rprofile that will be called by R every time you start an R session. Inside this file, you can tell it where to look for packages that you install. To make this file available to any R session it is necessary to put .Rprofile in your home directory. Assuming that your packages are in /home/[user]/src/library, the .Rprofile file should look something like:
.First function() { cat("Hello! You can put any R function that you want run upon start-up in here.") } # Ok, this next command points R to where your packages exist. # Note that R will still look in the default path as well. .libPaths("/home/[user]/src/library") # Now you will no longer need to use the lib.loc argument to library # when calling a package located in /home/[user]/src/library.

.Last function() { cat("Good-bye! You can put any R function that you want run while exiting R here.") } Note that many linux networks now use a different method for supplying R libraries, and the above .libPaths may interfere and cause problems with this new paradigm.

Back to Top

Software Requirements

The following directions were current at the time this tutorial was first written and apply to R 1.7.0, so please consult the Windows FAQ on the R project web site for more up-to-date directions. First, Tcl/Tk libraries and R must be installed on the system. R comes with a large amount of documentation detailing installation. To install R, go to:

CRAN

To obtain the necessary Tcl/Tk software, go to:
http://dev.scriptics.com

Important! The Tcl/Tk interface package, tcltk, for R versions 1.7.0 only work with Tcl version 8.3.x and for R version 1.7.0, it only works with the newer Tcl/Tk version 8.4.x. For Windows users, R version 1.7.0 now installs Tcl/Tk for you by default. If you are on Windows and using R version 1.7.0, please see the Windows FAQ on the R project site (http://www.R-project.org) for more information. If you do not know which version of R you have, type (from the R prompt):

R.version.string

For instructions on installing the Tcl/Tk software go to:

http://www.tcl.tk/doc/howto/compile.html

Notes:

To install Tcl/Tk on unix, you may want to ask your systems administrator to do it for you as it is a rather onerous affair. Generally, Tcl/Tk will already be installed on unix/linux.
In unix, you may have to set an environment variable to let R know where to find Tcl/Tk. Something like:
setenv TCL_LIBRARY /opt/local/tcl8.3.2/lib/tcl8.3 setenv TK_LIBRARY /opt/local/tk8.3.2/lib/tk8.3
Again, check with your system administrator about specifics to your system. For instance, the path to tcl8.3 will probably be different from the one given above. Ask your systems administrator where it is, or try the following unix/linux command.

find /[base directory] -name init.tcl -print

Note that [base directory] should be replaced with the directory where you suspect tcl might be. Something like /opt (above example), /usr or something of the kind.

Once you have set the correct TCL_LIBRARY and TK_LIBRARY paths it is recommended that you enter these commands in your .login or .cshrc or other appropriate file so that these variables are set automatically in the future.
In Windows, if you are using an R version 1.7.0 you will also need to tell R where to find Tcl/Tk. It may behoove you to simply upgrade to version 1.7.0 (or greater). Otherwise, you will need to set an environment variable and possibly a path. This can be done from within your R session with the following type of commands (see the R-CRAN Windows FAQ for more information):
Sys.putenv("TCL_LIBRARY"="C:/Program Files/Tcl/lib/tcl8.3") Sys.putenv( PATH=paste( Sys.getenv( "PATH"), "C:/Tcl/bin", sep = "))
Note that if you set the environment variable from within R, it will not remember this for the the next session. Better to upgrade to R version 1.7.0 (or greater) and not have to worry about it ever again.

The Underlying Functions

The underlying functions that actually perform the extreme value analyses were written by Stuart Coles for S-Plus and were ported into R by Alec Stephenson. For information on these functions please see Coles (2001) (b) and more specifically the accompaniment to this book Coles (2001) (a). For information on the R port see the web page: http://www.maths.lancs.ac.uk/~stephena/software.html. The primary difference between the original S-PLUS version and the R port is that the R port uses the optim function for finding MLEs instead of the S-PLUS nlm function. The following notes are nearly verbatum from Alec Stephenson's notes on the differences.

As mentioned above, the R port uses the general purpose optimization function optim. If R cannot find this function make sure you have the latest version of R.
Both R and S may give warning messages of the form 'NaNs produced in: log(x)'. This is a result of evaluating log at a negative number and may occur when the likelihood is evaluated outside of the valid parameter space. These warnings can generally be ignored.
In the S version, the $conv element of the returned fit list is either true or false (T or F). When true, a local minimum has theoretically been found. In the R port, the $conv element is the return code provided by the optim function. See the help file for optim for the details. A local minimum has theoretically been found when this is zero.
The optim function in R allows the user to select which optimization method to use. These may be selected from the extreme toolkit dialogs as well. The default method is Nelder-Mead. Another useful method is BFGS. Generally, if one method seems to fail try the other.

Miscellaneous

Whenever a GEV, GPD or PP model is fit to a data object, the entire fitted object is stored within the original data object. Because this toolkit uses Stuart Coles' routines for the fits, the original data is duplicated in the fitted object. For larger datasets this can quickly increase the size of the .RData file and suck up memory. If you are using a relatively large dataset and are performing many fits on it, then it would be a good idea to remove fits that you no longer need. For example, if you want to remove the first GPD fit performed on the "ev.data" object foo, do the following from the R prompt: