Information Technology Reference
In-Depth Information
The best solution created with the GEP-RNC algorithm was found in gen-
eration 4987 of run 26. Its genes and their corresponding arrays of random
numerical constants are shown below (the sub-ETs are linked by addition):
Gene 1: **?***?aa??aa3508773
C
1
:
{-1.68454, 0.798279, 1.39566, 1.081695, -0.497864,
1.834442, 0.68457, -1.634003, -1.548554, 1.413025}
Gene 2: *aECC*?a?aaa?7277993
C
2
:
{0.504761, -0.312164, 0.37912, 1.399841, 0.078949,
1.280243, -0.284759, -0.672516, -0.983948, 0.898652}
Gene 3: E--SE-??????a3265159
C
3
:
{-1.369293, 0.075317, 0.783661, 0.558258, -0.239593,
-1.260437, 0.398193, -1.696289, 0.372406, 1.407165}
Gene 4: -EE-?a?aaaaaa7193625
C
4
:
{-1.263519, -1.874268, 1.334442, 1.417694, 1.470489,
1.417694, -0.117675, -1.993897, 0.801728, 0.080383}
Gene 5: -L-*EEaaa?a?a4511699
C
5
:
{0.903077, 1.595642, -0.763794, 1.066101, -0.158966,
1.764313, -1.357117, 0.955719, -0.027954, -0.122955} (5.21a)
It has a fitness of 999.957 and an R-square of 0.999999672 evaluated over
the training set of 20 fitness cases and an R-square of 0.999999204 evalu-
ated against the same testing set used in the previous two approaches, and
thus is better than the models (5.19) evolved with the GEA-B algorithm and
better than the model (5.20) created with the GEP-NC approach. More for-
mally, the model (5.21a) can be expressed by the following C++ function:
double apsModel(double d[])
{
double dblTemp = 0.0;
dblTemp = ((((-1.68454*(-1.548544))*1.834442)*
(d[0]*d[0]))*1.081695);
dblTemp += (d[0]*exp(cos(cos((-0.672516*d[0])))));
dblTemp += exp(((exp(0.783661)-(0.398193-
(-1.260437)))-sin(0.558258)));
dblTemp += (exp((d[0]-(-1.874268)))-exp(-1.993897));
dblTemp += (log((d[0]*d[0]))-(exp(d[0])-exp(-0.158966)));
return dblTemp;
}
(5.21b)
