Information Technology Reference
In-Depth Information
the trends prevalent in an antigen. Experiment 4 generates a random population
of trackers, whose size is approximately equivalent to the population levels gen-
erated during experiments 2 and 3 to examine whether the TEA performs better
than a random search in terms of trend detection rates and memory eciency.
5.2
Performance Evaluation
The results of the TEA are evaluated as an average across 10 runs. The perfor-
mance of the algorithm is assessed using two measures i) the number of trends
identified against the maximum available for detection and ii) the eciency of
the trackers in the long term memory pool to map to the trends. Eciency
can be measured as the number of price change values included in the memory
tracker that are not contained within the match sequence 'MS'. For example if
the trend to be found was [2.0, 2.5] and the best fitting tracker was [2.0, 2.5,
3.0] the price value 3.0 within the tracker is redundant given the MS of [2.0,
2.5]. The degree of eciency, or to be more precise ineciency, would therefore
be calculated as 1 over 3, or 33%. The TEA was written in C++ and run on a
windows machine with a Pentium M 1.7 Ghz processor with 1.0 Gb of RAM.
6R su s
The results of experiments 1 to 3 are discussed in the following sections and
are listed in Table 2 whilst those of experiment 4 are found in Table 3. TEA
execution times varied from approximately 40 to 50 seconds for experiments 1
and 2, and 7 to 8 minutes for experiment 3.
Table 2.
Detection rate and memory eciency results
Trend Detection Frequency
Experiment T1 T2 T3 T4 T5 T6 T7 T8 Total Detection Rate
1
10 10 10 6
2
1
0 n/a
39
55.7%
2
9
9 10 9
9
7
3 n/a
56
80.0%
3
10 10 10 10 9 10 8
10
77
96.3%
Redundant memory values
Experiment T1 T2 T3 T4 T5 T6 T7 T8 Total Ineciency Rate
1
0
2
0
0
0
1
0 n/a
3
3.2%
2
0
0
0
0
0
2
1 n/a
3
2.1%
3
0
0
0
0
0
0
1
3
4
2.0%
6.1
Experiment 1. No Long Term Memory Pool Interaction
In accordance with Section 5 A1 was presented to the TEA from generations 1
to 10. The tracker population at generation 30 was replaced by the randomly
