Limbo

Last time I described my plans for tackling the problem os squashed graphs in my implementation of the Barnes Hut quad/octree approximation for a general graph drawing algorithm. I have since implemented such plans and found something interesting: nothing.

The results are inconclusive, the animated drawing of the graphs shows that convergence is met quickly but is not achieving a good layout, the refinement of forces led to no real improvement and the output each time a force is made gave me a long list of numbers, none which are significant (none doing what it shouldn't be doing).

In retaliation to the animated drawing, I theorised that the cooling schedule could be affecting the algorithm reaching a local minimum, however, changing this led to no improvements. I am/will be researching Yifan Hu's cooling schedule to see if this differs but I don't recall any large changes.

As the edges are both too long and too short, it suggests the problem still resides in the force calculations, however, testing and improvement led to no significant or "better" changes to the layout.

There is a ray of hope though, the extensive stretching of edges in the Z direction suggests an error within the code or a miscalculation somewhere, which could be the cause of other issues, predominately the compression. Finding the cause of this stretching is my one of my next tasks.

That's all the news I have for now, so as a parting gift, here is the graph 3elt (though it looks closer to data so I may have labelled it wrong) drawn in 3D.

P.S. I have now moved all images regarding my research into a timeline-folder type system which allows viewers to see my work from beginning to end. It's quite impressive to see progression from simple 3 dimensional cubes to massive graphs. Hopefully I will be make it available soon!