Information Technology Reference
In-Depth Information
dedicated to only that task so all the experiments were ran multiple times and
the results depicted are averages of the results thus gained.
ρ
-index Creation Time
10.5.1
First of all we present how much time the creation of the
ρ
-index consumes for
certain sizes of the testing graphs and particular parameter setting. Figure 10.10
represents the experiments performed on our four testing graphs. The
ρ
-index
created for each graph had 4 levels. The maximal size of a segment on the lowest
level is represented by the values on the x-axis. The other parameters were chosen
to be 10 at the second level, 5 on the third and 2 at the top level. Just to remind
the parameters are the maximal sizes of the segment at the particular level.
The results of this evaluation showed that the
ρ
-index is sensitive to underfill
of the structure. This can be observed for the case of the smallest graph when
even the ideal parameter setting which is around the value 8 for the max segment
size for the lowest level lead into a creation time which was greater than the best
time of a graph twice as large. We assume that this is caused by a inadequate
ρ
-
index setting. The
ρ
-index for this testing graph should have been created using
only three levels or smaller maximal segment sizes at the second and third level.
The creation times of the remaining three graphs indicate that the
ρ
-index
is highly dependent on the parameter setting. We can observe that the creation
times form a curve of a parabolic shape for all graphs and the size of the testing
graph determine the shift of the values on the y-axis. This implies that the
optimum parameter setting can be easily predicted for graphs at this particular
graph size category upon these experimental results. As a category we consider
35
G5000
G10000
G20000
G30000
30
25
20
15
10
5
0
0
10
20
30
40
50
60
70
lowest level cluster size
Fig. 10.10.
A
ρ
-index creation time
