|
Sequence induction with 5j4+4j3+3j2+2j+1
Table 1
Parameters for the symbolic regression problem
|
| Function set |
+,-,*,/ |
| Terminal set |
a (j) |
| Number of fitness cases |
10 |
| Number of runs |
100 |
| Number of generations |
100 |
| Population size |
60 |
| Success rate |
77% |
Table 2
Comparison of GP and GEP on the sequence induction problem
|
| |
Exp |
Gen |
Pop |
Fit. cases |
Succ |
Rz |
Fz |
GP |
Koza [1] |
51 |
500 |
20 |
0.15 |
29 |
14,790,000 |
GEP |
Table 1 |
100 |
60 |
10 |
0.77 |
4 |
240,000 |
Conclusion:
In this case, GEP outperforms GP in 62 times (see How to evaluate the performance for details)
. It's worth noticing that GEP does not use the so called ephemeral random constant (R), whereas Koza [1] uses R = {0,1,2,3}, i.e., the terminal set is in this case {j,0,1,2,3}.
Download the executable
Bibliography:
1. Koza, J. R. (1992). Genetic Programming: On the Programming of Computers by Means of Natural Selection. Cambridge, MA: MIT Press.
Back
***
|
