Genetic Programming Operators

Figure 8.8 Genetic operators used in the genetic programming experiments.

A potentially serious drawback of GP programs, unless they are very simple, is that they can rarely combine accuracy with lucidity. If it is important that the user appreciate the logic by which the program has reached its conclusion, it may be necessary to accept diminished performance. There are parallels here with the operation of neural networks, which are effective for many types of problem, but, as black boxes, are usually unable to tell us how they reached their conclusions. This lack of transparency may be a serious drawback if we would like to use genetic programs not only to predict or analyse a phenomenon, but also to help us to understand it. 

MOGP is based on the more traditional optimisation method genetic programming (GP), which is a type of GA [53,54]. The main difference between GP and a GA is in the chromosome representation in a GA an individual is usually represented by a fixed-length linear string, whereas in GP individuals are represented by treelike structures hence, they can vary in shape and size as the population undergoes evolution. The internal nodes of the tree, typically represent mathematical operators, and the terminal nodes, typically represent variables and constant values thus, the chromosome can represent a mathematical expression as shown in Fig. 4. 

Fig. 4 In genetic programming (GP), a chromosome is a tree structure and can be used to represent a mathematical expression where the internal nodes are mathematical operators and the terminal nodes are variable or constant values...

W. M. Spears and V. Anand, A study of crossover operators in genetic programming Springer, 1991, vol. 542. 

Lohl, T. C. Schulz and S. Engell. Sequencing of Batch Operations for Highly Coupled Production Process Genetic Algorithms Versus Mathematical Programming. Comput Chem Eng 22 S579-585 (1998). 

No matter what the effect of the genetic operators may be, the evolving programs have a tendency to grow indefinitely unless limits are placed on their size. These limits exist to reduce the quantity of redundant code (known as junk), which otherwise quickly become excessive. 

The standard GA procedures can be expanded to include various additional operators, such as encapsulation. This is a particularly valuable step, which collapses a section of code into a single node, thereby protecting it from the disruption which may be brought about by the action of the genetic operators. Encapsulation can be used to create automatically defined functions which are subprograms which may evolve in parallel with the main program. These consist of subroutines or subfunctions which perform specific tasks, and are typically called with one or more variable parameters passed across to them. 

---