Next: 3 Symbolic Regression
Up: 2 Even Parity
Previous: 2 Parity Fitness Assessment
The Parity problem has also been investigated in
detail in [Langdon and Poli, 2002].
The most characteristic aspects of this problem are the few
test cases that define fitness and the requirement that all
terminals are processed by a solution to achieve the ideal fitness.
Additionally, because of the
nature of the problem, a random guess on all fitness test cases
will lead to a score of 50%. O'Reilly and Oppacher (1994,1995)
performed a comparison of several hill-climbing
methods using similar genetic programming operators to show that hill-climbing
generally performed better than genetic programming on the 6 and 11 Multiplexer
problem. Similar results were also obtained in [Juels and Wattenberg, 1995].
While the two problems, Multiplexer and Parity, are not
identical, they share several similar features (e.g. boolean functions
and terminals and very few fitness function values).