So here's the
general gist of the program:
1) We read
in an xml file using the fileReader provide by Prefuse
2) We gather
the information about the nodes, normalize the data and create
instance(s) with the doubles
3) A user moves
two nodes. A constraint of "same" or "not same"
is placed on those two nodes depending on the distance between
them.
4) PCKMeans
is called using an ArrayList of said constraints and the Instances
comprised of the instance(s) and it returns the instances in the
requested number of clusters.
5) We create
nodes for the cluster centers, and then attach edges from the
centers to the associated nodes.
6) Remove cluster
centers and edges the next time the user moves a node.
7) Repeat steps
3 through 6 until satisfied.
The idea is
to incorporate user feedback into the clustering. The clustering
algorithm can go against the user-created constraints, but only
if it really, really wants to, and even then it invokes a penalty.
So if the user screws up entirely, the program will know. By using
user feedback, it should take less time to arrive at the correct
cluster by manually moving the nodes.
We'll be running
some experiments, and I'll post the results of those up here,
as well a our final paper and my final report.
