My previous entry, posted the whole of 24hours ago, gave a lovely list of what I want/need to do before I can compare my results. The first on that list, elegantly labelled '0.5' was to figure out why smaller graphs did not follow the suit of the larger graphs; a problem which vexed me as it should be a simple answer, yet none could be found.
Well it's time to rejoice, and somewhat cry. The bad layout in smaller graphs is an unexpected behaviour of the algorithm which generates a value of 'k' too high for a graph with too few vertices and edges. This high value is far too high for the graph to be drawn and so the only movement the graph can do is expand, with no tugging forces of the springs to pull it back into shape.
It should be noted, the calculation for k is currently based on the Frutcherman and Reingold formula which uses the size of the viewing area. Changing this so it becomes a value proportional to the size of the graph, as Walshaw has done, should resolve this petty issue.
Further to this, I have been constructing the testing framework which will run the finished algorithms, measuring their quality and putting the data into neat little tables to show which runs best. I am still researching how to achieve the best viewpoint and which method would be best for my work.
Once again, whilst joining all algorithms into one big framework, I am also looking to keep them as identical as possible to give a fairer test.
For now, I will continue and hopefully put another update out before the end of the week.
No comments:
Post a Comment