Home > Datasets, experiments, what i've been doing > Selected Pairs evaluation

Selected Pairs evaluation

November 23rd, 2011 Leave a comment Go to comments

Pádraig suggested that we use only those nodes that have communicated in the training phase, as the message pairs  for the testing phase. To this end, I implemented this as follows:

Given the entire edge list for the training period, take each edge, and create a pair from the source and destination node. Do not create a pair of the pairing exists (either direction). At simulation startup for the test period, send messages between only those pairs, in both directions. So that messages are sent:  A->B and B->A. This reduces the number of messages sent, and it can be argued that only nodes that know each other already, will want to communicate in future. This will give us a much better picture.

Enron-Split-A

Enron Split - A results

Enron Split - A results

Prophet and Hold&Wait

Hypertext2009

Hypertext2009-split results

Hypertext2009-split results

Prophet and Hold&Wait

Studivz-split

Studivz-split results

Studivz-split results

Social Sensing Split

Social Sensing Split results

Social Sensing Split results

Very strange for Undelivered hops, but these charts will change when they use ONLY the best run for a given metric, not the best of any.

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