Geoscience Reference
In-Depth Information
Key
Two parent chromosomes, resulting from tournament selection
modified
chromosome
segment
Chromosome A
Chromosome B
+
*
Source material
/
-
*
Z
A
B
D
E
X
Y
Crossover
*
+
Two parent
offspring
/
-
*
Z
X
Y
D
E
A
B
Mutation
+
*
*
A
B
G
H
Note that when mutation occurs at a 'head' it can be mutated to a
variable or function, but if mutation occurs at the tail then
only a variable can be palced at that point.
Inversion
+
/
*
One parent
offspring
B
A
D
E
Note that in some cases inversion may make no difference to the
solution, i.e. Chromosome A
Transposition
+
/
*
/
A
D
E
D
E
Note that transposition may involve a simple repositioning of parts of
the solution or as in this example duplication and insertion
FIGURE 8.4
Illustrated examples of crossover, mutation, inversion and transposition.
instruction to the GP software to stop the evolution process, usually based on the generation number
or fitness as determined by the raw fitness function. The final solution can then be evaluated by the
user using any number of additional goodness-of-fit metrics or other appropriate tools.
GP is a stochastic process meaning that each GP run will be different to the next, even when
input parameters remain constant - GP can be expected to identify and utilise important input
variables but it will do so differently each time. Thus, 10 runs will probably provide 10 different
solutions. Multiple runs, for example, 10-20, should be completed because even if evolved models
 
Search WWH ::




Custom Search