GEP Book

  Home
  News
  Author
  Q&A
  Tutorials
  Downloads
  GEP Biblio
  Contacts

  Visit Gepsoft

 

C. FERREIRA In N. Nedjah, L. de M. Mourelle, A. Abraham, eds., Genetic Systems Programming: Theory and Experiences, Studies in Computational Intelligence, Vol. 13, pp. 21-56, Springer-Verlag, 2006.

Automatically Defined Functions in Gene Expression Programming

General Settings
 
The sextic polynomial of this section x6-2x4+x2 was chosen by Koza (Koza 1994) because of its potentially exploitable regularity and modularity, easily guessed by its factorization:

x6-2x4+x2 = x2(x-1)2(x+1)2

(14)

For this problem, the basic parameters common to both GEP and GP, irrespective of the presence or absence of ADFs, were kept exactly the same as those used by Koza. Thus, a set of 50 random fitness cases chosen from the interval [-1.0, +1.0] was used; and a very simple function set, composed only of the four arithmetic operators F = {+, -, *, /} was used. As for random numerical constants, we will see that their use in this problem is not crucial for the evolution of perfect solutions. Nonetheless, evolution still goes smoothly if integer constants are used and, therefore, one can illustrate the role random constants play and how they are integrated in Automatically Defined Functions by choosing integer random constants. So, when used, integer random constants are chosen from the interval [0, 3].

The fitness function used to evaluate the performance of each evolved program is based on the relative error and explores the idea of a selection range and a precision. The selection range is used as a limit for selection to operate, above which the performance of a program on a particular fitness case contributes nothing to its fitness. And the precision is the limit for improvement, as it allows the fine-tuning of the evolved solutions as accurately as necessary.

Mathematically, the fitness fi of an individual program i is expressed by the equation:

(15)

where R is the selection range, P(ij) the value predicted by the individual program i for fitness case j (out of n fitness cases) and Tj is the target value for fitness case j. Note that the absolute value term corresponds to the relative error. This term is what is called the precision and if the error is lower than or equal to the precision then the error becomes zero. Thus, for a perfect fit, the absolute value term is zero and fi = fmax = nR. In all the experiments of this work we are going to use a selection range of 100 and a precision of 0.01, thus giving for the set of 50 fitness cases fmax = 5,000. It is worth pointing out that these conditions, especially the precision of 0.01, guarantee that all perfect solutions are indeed perfect and match exactly the target function (14). This is important to keep in mind since the performance of the different systems will be compared in terms of success rate.

 

Home | Contents | Previous | Next