Oh what a week.
I finished up designing the experiments and implementing them.
Each of them were writing the number of nodes moved and the Adjusted
Rand Index (ARI) at each step (every time a node was moved) to
a file. But I thought, wait, one run of an experiment isn't going
to tell us much, since various experiments can differ greatly,
even with the same methods and data sets. So I wrote a very simple
method to take the file, read it in and add the new ARI to it.
Then my fileMerge program takes all four input files (one for
each experiment) and gets the accumulated ARIs plus the number
of times each experiment was run. It then averages the data from
each experiment by the number of times the experiment was run.
But just to make it clear, the program isn't averaging the different
experiments together, it's merging the different runs
of the same experiment together.
So let's talk
experiments. We have four of them.
In all of these experiments, the program is going to put the selected
node in the correct cluster.
1) A purely
manual experiment. No interactive layout, no force-direction,
no springs. The program selects a node at random and places it
near the correct cluster. Repeat until all nodes have been moved.
2) Interactive
layout, with force-direction and springs. The program selects
the node farthest from its correct cluster and then places it
in it's correct cluster. Repeat until all nodes have been moved.
3) Interactive
layout, with force-direction, springs and PCKMeans clustering.
This program selects a node at random, finds the coordinates that
are near the correct cluster, and then simulates a MouseEvent
to make the program think that the methods itemPressed, itemDragged
and itemReleased are called. Since itemReleased calls clustering
it then performs as the program usually does, assigning nodes
to clusters with edges and centers.
4) Interactive
layout, with force-direction, springs and PCKMeans clustering.
This program acts exactly like the above one, except that it uses
the farthest-first heuristic like experiment two does.
I've also been
focusing on writing the paper for this program, which is nearly
finished. I just have to figure out Latex's bibliography style,
and am sure that it's fairly easy, I've just had my mind on other
things. It's my first time writing a paper with Latex, and I put
in a table today which was exciting.
In Other News:
We went out with the whole lab on Wednesday for ice-cream to say
goodbye. It was a lot of fun, and it actually was the first time
that I've seen all of these people together in a long time. So
it was nice to see everybody. Except that most of the CORAL lab
is in Scotland. I think that I'll get to see some of them before
I go though. And I'm leaving on Saturday! Yikes!
As you can
probably tell, I'm writing some of this in advance, since my computer
doesn't have internet access at home, and I need to tar this on
UNIX. I could use Knoppix, but my parents get edgy when their
computer boots up into something they've never seen before. So
it's near the end of the week, and the rest of the week should
be documentation and system clean-up. So this is my last entry!
I'll probably update this site sporadically during the year, when
I get news on where the project.
