Misc improvements

[ugurdogrusoz]

• 1. We need to calculate the structural properties of the graphs in our dataset (those that are converted from the original Rome dataset). Initially, let's calculate the average node degree.
• 2. Let's also calculate the mean, median, and standard deviation for each metric of each ratio. So for phi=0.4 for example, we need the mean, median, and standard deviation of the number of edge crossings, number of node overlaps, etc.
• 3. Currently, we're comparing the performance of our algorithm for the directed part with Dagre. Let's do the same thing with CoSE for the undirected part. For this, we basically need to take only the undirected part of each graph in our test suite and run an CoSE (non-incremental, randomize: true) on it. Then, we need the same metrics for running CoSE on this undirected part:
  • no of edge crossings
  • no of node overlaps
  • average edge length
  • total area
  • run time
• 4. Rename the sample "paper" in the demo as "class" and add a new one ("knowledge graph") from the paper
• 5. We need 1 more HySE layout example in the supplement which is twice as big as the biggest so far if possible.
• 6. We need 2 examples (one with only a dozen nodes or so, the other with 50-60 nodes or so) laid out with HySE. For this we also need to layout the hierarchical parts of the two graphs with Dagre and undirected parts of the two graphs with CoSE. We will present these side by side in the paper.

[ugurdogrusoz]

For item 3, we need to run one more experiment where phi = 4 (i.e. there are 4 times as many undirected nodes as we have directed nodes). This is when undirected nodes dominate and we hope that execution time will be similar to that of CoSE.