Unsure why, but I am beginning to like playing with eigenvectors. It seems like forever since I last played with any 'scary' maths, but thanks to a fellow PhD student, I am finally getting my head around this whole eigenvector/eigenvalue subject.
I am aware I have played eigenvectors before, but previously it was to understand how other more spectral graph drawing algorithms worked (and had no intention of implementing them). I am now needing them to use inertia bisection to help find the principal axis of a graph (normally used for graph partitioning) and use that to calculate the best viewpoint (perpendicular to it).
Meanwhile, I remain in waiting for my results of the 2D implementations. They were started yesterday (after more issues developed), and are still calculating away (worryingly only on their second graph). Checking output to the files, it seems they are still generating results and the slowness due to the number of parameters and size of the graphs. I was concerned for a point when I noticed output had stopped half way through a result (suggesting a failure) but thankfully Java is clever with writing to files; sending chunks at a time as opposed to writing directly to it.
Previous results and partial results from single parameters show that my Contraction Approximation is still faster in most cases, and offers competitive results for edge crossings with the less powerful repulsive forces, than the Quad Tree approximation. Future results will provide more solid evidence with which to make conclusions.
For now, I will continue with my work on finding the best viewpoint (sticking to the inertia bisection and principal axes instead of Webber's work on viewpoints).
I will most likely update again next week, for now, it is almost time for tutoring, so ta ta for now.
No comments:
Post a Comment