Home > Bugs, experiments, what i've been doing > Selected Nodes – with fixed HGCE output

Selected Nodes – with fixed HGCE output

November 30th, 2011 Leave a comment Go to comments

We identified a problem with the way HGCE was being used, for all other CFAs the community structure used was identical apart from the global parent used by BubbleH, but HGCE outputs a different structure every run. I adapted my code to make sure that BubbleRAP and BubbleH make use of the SAME community structure. In the results below, selected pairs of nodes are chosen based on their contact in the training period. Also, the overall results are reported for the best value for a given metric (i.e. the best run for delivery-ratio is used to show results for cost, latency, etc.). In this case, all use delivery-ratio as the primary metric.

Studivz

Studivz

Social Sensing

Social Sensing

MIT

MIT

Hypertext2009

Hypertext2009

Enron

Enron

Enron still shows a lose for BubbleH for HGCE. Detailed plot compared to previous results:

Delivery Ratio for Enron, BubbleH vs Bubble - old results vs new results - no great change

Delivery Ratio for Enron, BubbleH vs Bubble - old results vs new results - no great change

Still working on coding up an alternative to unlimited flood which allows us to accurately measure delivery ratio for selected pairs.

  1. No comments yet.
  1. No trackbacks yet.