This is
not meant to be a thorough guide
to the software, I just wanted to make it available to those
that are interested. Tamayo et al's
PNAS article
describes the algorithms and their applications in looking at gene
expression levels. Please
Email me
if you have questions or comments.
This is what it looks like. Basically, you place points
in the window, define your simulation and display
properties, then click "Start Animation." The agents will
move around and eventually settle into a formation that
depends on the initial placement of the data, and the
parameters you define.
This software was designed
for data of many dimensions, but the graphical display
(the fun part) can only represent two of those
dimensions. You can choose which dimensions you want to
see from the Display Properties menu, but the visual
behavior won't be what you might expect, as the
interactions in the higher dimensions govern the behavior
you see.
There are a few ways to get
data into the software: Clicking on the screen places a
point that can be animated (tick); right-clicking
places a static node (stump). From the Agents->New Agents... file
menu you can import from a tab-delimited file, distribute
points at random (make sure "dimensions" is set to at
least 2!), or place named points at specific
locations.
|
 |
|
This is the heart of the algorithm. The modification of
these paramenters will determine the way your data
clusters. I won't describe in detail the algorithms
used, because they're not relevant unless you're
really using the software. If, however, you
truly want to know how the parameters affect the
clustering, and it's not clear from the Simulation
Settings menu, email me
and I'd be more than happy to explain everything.
|
 |
|
The display options are fairly straightforward. The
"coordinate text" shows the names and coordinates of
points in the screen, and is only useful if the data
is imported. The "Monitor this coordinate as..." setting
allows you to choose which coordinates are represented in
the visual box.
|
 |