Information Technology Reference
In-Depth Information
To compare the results of experiments 2 and 3 with a random search, a random
population was generated consisting of 1,000, 4,000, 10,000 and 20,000 trackers.
Given the longest trend (T7) has four price values, and can be found by the
TEA with no data redundancy, each random tracker had a randomly determined
length between one and four. The randomly generated population would then be
mapped to antigen A, and the memory trackers compared to those of experiments
2 and 3 to see whether the TEA can outperform a purely random search. Results
are shown in Table 3, ticks indicate the trend was found, crosses indicate the
trend was not detected.
Table 3.
Trends detected using a random search
Analysis of Trends Detected
Pop Size T1 T2 T3 T4 T5 T6 T7 T8 Total
1,000
✓
X
✓
X
✓
X X n/a
3
4,000
✓
✓
✓
✓
✓
X X
✓
6
✓
✓
✓
✓
✓
✓
✓
10,000
X
7
20,000
✓
✓
✓
✓
✓
✓
X
✓
7
With a randomly generated population of size 1,000 only 3 of the 7 trends T1
to T7 were detected. The random search failed to find trends T2, T4, T6 and
T7. In comparison, during experiment 2 the TEA found 6 trends consistently,
missing only T7 70% of the time. The detection rate of the TEA is twice that
of the random search with just 1,000 trackers.
With 4,000 random trackers 6 of the 8 trends are found, trends T6 and T7
were undetected by the random search. Increasing the random population size to
10,000 trackers, 7 of the 8 trends are detected as T6 is now found. The random
search fails to find T7, even if we increase the tracker population to 20,000. This
contrasts to experiment 3 where the TEA, with only 4,000 trackers, can generate
a memory pool that detects T1, T2, T3, T4, T6 and T8 every time across all 10
runs, and T5 and T7 9 and 8 times out of 10 respectively. The TEA therefore
outperforms a random search.
7
Discussion
From experiment 1 it is seen that the TEA can evolve a population of trackers
that generate a memory pool able to successfully map to trends in a simple data
set (such as A1) with 100% accuracy and eciency. Increasing the number and
complexity of the trends to be found, as was achieved through the presentation
of A2, causes the algorithm to struggle to identify these potential trends.
Without knowledge of the trends from A1 being carried forward in the system,
detection rates of the TEA to the more complex trends falls significantly. This
can be corrected in the TEA by increasing the degree of proliferation to raise
the detection rate in the system. But what is of interest to us in this paper
