Information Technology Reference
In-Depth Information
To reflect the instability and high death rate prevalent in the short term
memory pool clones are eliminated five generations after their creation if they
do not improve on their anity to an antigen trend. This ensures excessive
population growth is carefully regulated and a return to a stable population
level soon after antigen presentation ceases.
Reviewing the mechanisms within the TEA one can see a close similarity
exists to algorithms such as CLONALG [5], however a number of notable dif-
ferences exist. Compared to CLONALG apoptosis occurs across all population
members in the TEA, not just the lowest anity members. In addition, due to
it's specialised nature, mutation in the TEA is not directly related to anity fit.
TEA also proliferates all bound trackers to form the short term memory pool,
encouraging diversity in the search space. The process in CLONALG is more eli-
tist, as only the 'n' fittest population members are proliferated and mutated, and
from these only the best fitting clone becomes a memory candidate. All remain-
ing clones are eliminated. In essence CLONALG skips the short term memory
pool stage as it looks to find the best fitting candidate using the minimum of
resources. In comparison the TEA maintains the population of clones in order
to match and anticipate patterns arising in the data fed live to the system.
5
Testing Methodology
5.1
Methodology
In order to test the ability of the TEA to identify trends in a data series, a simple
antigen 'A' was constructed. 'A' contains 20 fictitious price movements, and 8
trends, T1 to T8. These represent the complete set of trends in A in accordance
with the definition described in Section 2. The antigen and trends T1 to T8 are
listed in Table 1.
To assess the ability of the TEA to associate new novel antigen with those
experienced during past presentations we split antigen A at the mid point into
Table 1. Antigen data sets with observed trends
Antigen Price Movements
A
[ 1, 2, 1, -0.5, 1, 2, 1, 0.5, -0.5, 0.5, 2, 1, 2, -0.5, 2, 1, 2, -0.5, 1, 1.5 ]
A1
[ 1, 2, 1, -0.5, 1, 2, 1, 0.5, -0.5, 0.5 ]
A2
[ 2, 1, 2, -0.5, 2, 1, 2, -0.5, 1, 1.5 ]
Trends
T1
[1,2]-seeninA,A1andA2
T2
[1,2,1]-seeninAandA1
T3
[2,1]-seeninA,A1andA2
T4
[1,2,-0.5]-seeninAandA2
T5
[2,-0.5]-seeninAandA2
T6
[2,1,2]-seeninAandA2
T7
[2,1,2,-0.5]-seeninAandA2
T8
[ -0.5, 1 ] - seen in A
 
Search WWH ::




Custom Search