Biomedical Engineering Reference
In-Depth Information
TABLE 12.7 Migration Policy Analysis (freq-rate)
LAN
n
=
2
n
=
4
n
=
8
freq-rate
b
f
b
f
b
f
5-1
99,904
76,447
75,317
62,908
52,127
35,281
5-10
61,910
37,738
68,703
55,071
56,987
52,128
5-20
92,927
72,445
72,029
66,368
59,473
54,312
20-1
98,133
86,490
75,824
66,854
66,021
56,776
20-10
82,619
45,375
70,497
57,898
53,941
48,968
20-20
89,211
74,236
72,170
65,916
59,324
53,352
50-1
95,670
70,728
77,024
65,257
64,612
55,786
50-10
92,678
41,465
66,046
51,321
59,013
51,842
50-20
95,374
76,627
72,540
62,371
59,923
52,650
For the first problem instance, the parallel GAs sampled less points in the search
space than the serial one, whereas for the second instance, the panmictic algorithm is
mostly similar in the required effort with respect to the parallel ones.
Increasing of the number of islands (and CPUs) results in a reduction in search
time, but it does not lead to a better fitness value. For the second problem instance,
the average fitness was improved by a larger number of islands. However, for the first
problem instance, we observed a reduction in the fitness value as we increased the
number of CPUs. This counterintuitive result clearly states that each instance has a
different number of optimum number of islands from the accuracy point of view.
The best tradeoff is for two islands
(n
2
)
for the two instances, as this value
yields a high fitness at an affordable cost and time.
Table 12.10 gives the speed-up results. As it can be seen in the table, we always
obtain an almost linear speedup for the first problem instance. For the second instance,
we also have a good speedup with a low number of islands (two and four islands);
eight islands make the efficiency decrease to a moderate speedup (6.42).
=
TABLE 12.8 Parameters When Heading and
Optimum Solution of the Problem
Independent runs
30
Popsize
512
Fitness function
F1
Crossover
OR (1.0)
Mutation
Swap (0.3)
Cutoff
30
Migration frequency
20
Migration rate
1
Search WWH ::
Custom Search