An Introduction to Genetic Programming

Introduction


Genetic programming falls into the category of
evolutionary algorithms.


Genetic algorithms vs. genetic programming.


Concept developed by John Koza.


Types of research


Determining which problem domains genetic
programming can successfully be applied to and
compare GP with other AI techniques.


Applying genetic programming to derive solutions to
problems not previously solved.

Genetic Programming
Algorithm


Create the initial population


Repeat


Evaluate the population


Select parents for the next generation


Apply genetic operators

Until the termination criteria are met

Symbolic Regression Problem


Given


x:

11 , 4 , 1 , 8 , 5 , 9 , 3 , 7 , 2 , 6


y:

1742400.0 , 3600.0 , 0.0 , 254016.0 , 14400.0 ,

518400.0 , 576.0 , 112896.0 ,

36.0, 44100.0



What function does y represent?



What else do we know?


x forms part of the function


One or more of +,
-
, /, *, sqrt forms part of the function




Fitness

Cases

Terminal set

Function Set

The Artificial Ant Problem


What are the moves that must be made by the
ant to pick up all the food pellets? (Note: the
shaded squares represent food pellets)

Legal Moves:

Move left

•
Move right

•
Move up

•
Move down

•
Pick up food

Function

Set

Terminals and Functions


The terminal set contains a variable
representing each input to a problem.


The terminal set may also include constants and
operators that do not take any arguments.


The function set consists of operators or legal
moves, e.g. +, if, for, move
-
left.


Terminals and functions are collectively referred
to as primitives.


Closure and sufficiency.


Extraneous primitives degrade system
performance.

Representing an Individual


Parse trees are commonly used to represent
each member of the population.


Each tree is constructed by randomly choosing
elements from the function and terminal sets

F = { + ,
-

, * , / }

T = { x , y }

F = { + ,
-

,
*

, / }

F = {
+

,
-

, * , / }

F = { + ,
-

, * , / }

T = {
x

, y }

T = { x ,
y

}

T = {
x

, y }

T = { x , y }

A maximum tree depth must

be specified as a GP parameter