Hi All,
I have in the past years done a bunch of Genetic Algorithms, and some Genetic Programming. But now I want to do a real Genetic Programming program.
I generate a random set of instruction and parameters, and run the program that they represent.
4 bit for the instruction and 2x4 bit for the memory location
e.g.
move content of memory location 2 to locating 10
copy content of memory location 5 to location 2
add content of memory location 10 to location 2
a.s.o.
So my genes are just an array of bits.
But my problem is that if a program is just a single bit off from, for instants the optimal solution; it can have a fitness that is very pore.
I have googled the web again and again, and read a bunch of books, but nowhere can I find a solution to the problem.
Maybe, there isn’t a solution.
A discreet fitness landscape isn’t useful..
Anyone out there have any ideas on how to make a fitness function that makes sense in this case or how to convert my discrete space to something more useful.
Thanks,
Henning