Information Technology Reference
In-Depth Information
Problem with Beliefs
especially for the b-agents results.
5.4.2
Evidence of Incentive Compatibility
From the experiments in the previous section, the DNCF mechanism
is shown to be both stable and scalable, particularly if the agents are
employing b-strategy. So there is one more task remaining: to show
that it is to the benefit of the agent to follow the b-strategy.
Again, a series of test cases with number of agents ranging from
4 to 10 are generated. This time, in each case, half of the agents are
designated as experimenters. Each case is run twice with identical set-
tings, except that each experimenter employs the b-strategy in the first
run, and the nb-strategy in the second run.
Each experimenter agent then compares the outcome he obtains
by using the two strategies according to his preference order for the
bids: each time that he obtains a more preferable result by using the
b-strategy, a 'win' is rewarded to the b-strategy, a 'lose' if the result
by using the b-strategy is worse, and a 'tie' if the results by the b-
strategy and nb-strategy are the same. Similarly, the number of 'wins',
'ties' and 'loses' are also recorded for nb-strategy. The experiment is
repeated 1,000 times and we count the number of 'wins', 'ties', and
'loses' obtained by each strategies.
The result is shown in Figure 5.5, where we see that the results
obtained by using the two strategies are actually very close, with the
far majorities of the results being in ties, despite the fact the number of
proposals sent by nb-agents using nb-strategy greatly outnumber those
sent by b-agents (see Figure 5.4), which indicates that the b-strategy
is actually superior in terms of e
ciency.
Finally, we also test the b-strategy against a third strategy which
we call the passive strategy. In this experiment, certain agents are
designated to employ the passive strategy, which means that the agent
will simply propose nothing even when it is his chance to propose,
Search WWH ::
Custom Search