|
|
|
|
|
|
by Rusky
5373 days ago
|
|
|
GP would just turn all the programmers to digital shepherds :P It would be a really interesting challenge to get broad enough coverage in the fitness function that things like the previously-mentioned "400 NOPS" don't cause problems. Besides unit (or otherwise) tests, it should also be possible to use static analysis to measure fitness and/or use a static type system to restrict the possibilities. I'd think this would have some of the same problems, though- anybody know of any work on that? |
|
|
All the branches of evolutionary computing agree that evolved systems are really good at finding flaws in your problem statements. Most of the evolved systems you see in production are not the product of the first attempt to evolve the system.
> Besides unit (or otherwise) tests, it should also be possible to use static analysis to measure fitness and/or use a static type system to restrict the possibilities. I'd think this would have some of the same problems, though- anybody know of any work on that?
Type consistency for GPs is a whole subfield in itself. See the textbook I referred to, it has a chapter on it from memory.