Frequently Asked QuestionsQ: How can I write my own GEP functions?A: Simple: from pygep.chromosome import symbol
@symbol('~')
def foo(x, y, z):
return bar(x, y, z)Now you can use foo in your chromosomes just like any of the provided functions. If your function uses a single character name, you do not need to use the symbol decorator. The arity of non-terminal functions is culled from its func_code.co_argcount, meaning that using *args or **kwds will not do anything. See pygep.functions for examples. Q: Can I provide my own linkers?A: Of course! A PyGEP linker just needs to accept *args and return a single result. See the sum_linker code for an example: def sum_linker(*args):
'''Returns the sum of all sub-ETs'''
return sum(args)If you do not provide a linker for your population, the default linker will simply return the top results of each gene's evaluation. Q: I know that GEP programs will always be syntactically correct provided I use compatible types for my operators and operands, but what do I do about semantic errors such as division by zero?A: Chromosome evaluation doesn't swallow run time errors, so anything like this will be raised to whatever level you evaluate at, typically in the _fitness(self, obj) method. The usual practice is to catch any such errors that occur and dock the responsible individual's fitness. See the fitness method of the regression demo for an example of this: def _fitness(self):
total = 0
for x in DataPoint.SAMPLE:
try:
guess = self(x) # Evaluation of this chromosome
diff = min(1.0, abs((x.y - guess) / x.y))
total += self.REWARD * (1 - diff)
except ZeroDivisionError: # semantic error
pass
return total
|
