Mary's Cascade code
(c) Mary McGlohon, 2010  mmcgloho@cs.cmu.edu

***********************************************

This is a small package for generating and processing cascades, from a
list of edges.  It is a work in progress.  It is also research code
that I originally wrote without intending to share with anyone. Before
posting I've cleaned up and documented a few things, and removed some
of the really horrible artifacts from deadline crunch time. But it's still got some rough edges.

CLASSES

--------

Cascade.java

This is the baseline Cascade class.  A cascade is a subgraph of nodes
and edges.  It is typically a tree, but does not have to be (e.g. our
blogs work in SDM 2007).  This keeps track of nodes, along with
authors and timestamps, and can test for isomorphism (either just
structures or structures with authors).

---------

BuildCascadesEfficient.java

This will take a list of edges in the format
src,dest,time,srcauthor

and make file where each line is one cascade (connected component).
If it's important to you to have the authors and timestamps for root
nodes, you will want to have lines for root nodes in format

rootid,-1,roottime,rootauthor

The -1 will not be stored as a node.

-------

CascadeIterator.java

This just takes a file and makes it so you can fetch Cascades line by line.

-------

CascadeUtils.java and CascadeViz.java

These are just utilities files (currently pretty small) that are used by the other classes.


--------

CascadeMarkovModel.java

"Yo dawg, I herd you liek graphs, so I made you a graph of graphs."
This is a tree of cascades.  It begins with a cascade of a single
node, which then has one child of a cascade of two nodes.  That then
has two possible children, depending on where the next node attaches.
In the case of an authored version, these will also have all possible
combinations of different authors.

From there, you can read in cascades one by one, and it will keep
track of the paths these cascades took-- that is, the order in which
nodes were added.

It may be more explanatory if you just go on and look at
uk.sci.weather-cascades.pdf or uk.sci.weather-authored-cascades.pdf

----------

LearnCascadeMarkovModel.java

This is what you'll actually run.  It takes in some arguments and
spits out an appropriate CascadeMarkovModel in graphviz format.

**************************************

TESTS

I have a small test of the cascade isomorphisms.  It does not
currently test on cascades where multiple parents are allowed.

To run the test, you need JUnit installed.  Then compile and run

java org.junit.runner.JUnitCore CascadeTest   

***************************************

INPUT/OUTPUT FILES

The input file you will need is

uk.sci.weather-all
It is a list of all the edges (including root posts) in the sample of this Usenet group.

The output files that will be generated are:

uk.sci.weather-temp
Intermediate file in BuildCascadesEfficient

uk.sci.weather-authored-cascades
uk.sci.weather-cascades
Output files from BuildCascadesEfficient, depending on whether authors used

uk.sci.weather-cascades.dot
uk.sci.weather-authored-cascades.dot
Graphviz files from LearnCascadeMarkovModel

uk.sci.weather-cascades.pdf
uk.sci.weather-authored-cascades.pdf
Dot output from building the .dot files