Information Technology Reference
In-Depth Information
For example, in Fig. 3 it is assumed a window of size 7. The last reference
network will be the network with 16 antibodies. The networks with 17, 18, 19,
20, 21, 22 and 23 antibodies are evaluated for comparison. Since the classification
performance of these networks are not superior to that produced by the reference
network, the convergence criterion was reached and the resultant topology will
be the network with 16 antibodies. The maximum number of antibodies to be
inserted into the network is given by the quarter part of the dataset size. This
limit of the number of antibodies in the network is a empirical value and was
chosenbasedonprevioustests.Inthisimplementation, the performance criterion
used is the TCR (Total Cost Ratio), that will be better explained in Section 5.
2.5
SRABNET PseudoCode
The steps described in theprevious subsections are presentedhere in a pseudocode
format.
Algorithm 1.
Pseudo-code of the SRABNET algorithm
1:
Begin
2: Initialization;
3: Initialize the network with one antibody per class, using the training dataset. The
weight vector of each antibody corresponds to the mean of the samples belonging
to the class to which the antibody was assigned;
4:
while
the convergence criterion is not met
do
5:
for
each input pattern (antigen)
do
6: Present a randomly chosen antigen to the network;
7: Calculate the Euclidean distance between the antigen presented and the anti-
bodies in the network;
8: Find the winner antibody according to Eq. (1);
9: Increase the concentration level of the winner;
10: Update the weights of the winner antibody according to Eq. (2);
11:
end for
12: Choose the antibody to be cloned. The antibody to be cloned will be the one
that recognizes the antigen with the lowest anity (highest Euclidean distance);
13: The weight vector of the new antibody is the midpoint between the parent
antibody and the antigen with the lowest anity;
14: The new antibody will belong to the class with the maximum number of elements
among the ones recognized by the new antibody;
15:
if
the concentration level of a given antibody is zero and it is not the unique of
its class
then
16: prune it from the network
17:
end if
18:
end while
19:
End
Supported by the dynamic behavior illustrated in Fig. 1, the pseudo-code
describes the whole algorithm including the growing and pruning processes.
