----------------------------------------------------------------
Some helpful notes from Sam Lamphier in Spring 2008
----------------------------------------------------------------

I installed the Windows version, which provides a GUI.  I did get the
latest dev version, 3.5.7, as their later versions supported Java
1.5... and I figured if we work with that, Java 1.5 has alot that 1.4
doesn't.  I didn't notice any reason to not use the latest version
vs. the book version, so please let me know if you come across
something.

The easiest way to figure out how to use the UI is through their
tutorial (ppt)... look for the following on the Documentation page,
under "Tutorials":

    "A presentation demonstrating all graphical user interfaces (GUI)
    in Weka. (Warning: this is a large Powerpoint file.) [WekaDoc]"

To run the project J48, it's as simple as loading the arff file, then
choosing the J48 classifier.  I think the results might be slightly
different, in that there's a little more model info printed at the top
than if run from the command line.  By default, I think it does 10 way
cross validation.  Since the weka.jar is also at the root of the
install dir, in theory you can run java from a command prompt with the
same args as described for Linux by Dr. Resnik, but I didn't try it.
I believe I did try their UI SimpleCLI and that does allow you to
enter the args and did produce the same output as using the UI (maybe
some cosmetic differences, I can't remember).

To train for the homework, I had to edit the RunWeka.ini file to set
the -Xmx field (max memory for the JVM) to be -Xmx512m (the default
was something like -Xmx128m).  Otherwise the JVM ran out of memory.
The file that runs weka is RunWeka.bat -- the ini file is in the same
directory as the bat file (c:\Program Files\Weka 3.5 depending on
where you installed it).

Training classifiers is just choosing the "classify" tab from their
Explorer app (under Applications) and choosing the classifier of
interest (J48 is under the "tree" hierarchy).  Then the Start button
kicks it off... the tutorial (referenced above) is a nice place to see
the steps for this sort of thing.

When you choose a classifier, the command line args are displayed in
the GUI -- helpful if you want to compare the command line to GUI.

The source code for weka is in weka-src.jar, also at the root of the
install.

All in all, they've done some very nice work making this product, as
far as I can tell so far.

I did notice that the arff format allows Strings as a data type, but
not sure if you can use a string instead of the enumerated attributes
that we used for the project.  If you can use strings, then we might
not need to convert U.S. to UXSX.  However, there would have to be
some delimiter specified, so I'm not sure what limitations would be
there.  Possibly using their xml format (xrff) would solve any
delimiter issue.

    http://weka.sourceforge.net/wekadoc/index.php/en:XRFF_%283.5.6%29

[Additional note from me: I tried using String rather than enumerated
values, since that would certainly be simpler to implement, but it
complained.  I then tried to use one or two of the Filters that seem
as if they do the conversion for you, and couldn't get those to work.]