Information Technology Reference
In-Depth Information
double ADF2(double d[])
{
double dblTemp = 0.0;
dblTemp = (((d[0]-d[4])-(d[1]+d[0]))/((d[4]*d[3])*d[0]));
return dblTemp;
}
double apsModel(double d[])
{
double ROUNDING_THRESHOLD = 0.5;
double dblTemp = 0.0;
dblTemp = (((ADF0(d)+ADF2(d))-ADF2(d))*
((ADF2(d)-ADF2(d))+(ADF1(d)+ADF1(d))));
return (dblTemp >= ROUNDING_THRESHOLD ? 1:0);
}
(6.8b)
Note that, despite being called from four different places in the main pro-
gram, the calls to ADF
2
cancel themselves out and, therefore, only ADF
0
and
ADF
1
have an impact on the decision. As you can see, ADF
0
is called once
from the main program, whereas ADF
1
is called twice. It is also interesting to
see that this extremely accurate model uses only five of the nine available
analyses to make the diagnosis, namely, clump thickness, marginal adhe-
sion, bare nuclei, bland chromatin, and mitoses.
One of the best solutions created with the ADF-RNC algorithm was dis-
covered in generation 759 of run 4. Its homeotic gene plus all the normal
genes and their respective arrays of random numerical constants are shown
below (the neutral cells are not shown):
Gene 0: -.*.?./.+.d8.+.d1.d5.d6.d0.d7.d1.d2.d6.3.9.5.5.2.6.3.6
C
0
:
{1.777649, -1.266601, -0.487305, -0.586578, -0.147583,
0.402771, -1.077301, -0.016326, 1.421356, -1.279846}
Gene 1: *.+.d8.*.d2./.d3.d6.d4.?.d7.d8.d0.d3.?.5.2.3.7.1.0.6.1
C
1
:
{-0.384216, 0.10672, 0.475586, -0.800049, 0.122528,
-1.120483, 1.075531, 1.535095, 0.461121, 0.364654}
Gene 2: /.*.?.+.+.d2.+.d1.d5.d6.d0.d7.d1.d2.d0.5.2.7.6.0.8.1.0
C
2
:
{1.777649, -1.266601, -0.487305, -0.586578, -0.147583,
0.316253, -1.077301, -0.016326, 1.421356, -1.279846}
(6.9a)
Homeotic Gene: *.+.*.2.*./.+.2.2.1.2.2.0.1.0
This model classifies correctly 340 out of 350 fitness cases in the training set
and 171 out of 174 fitness cases in the testing set. This corresponds to a
