Custom Genes
Although radiate provides a variety of built-in gene types, you can create your own custom genes by subclassing the Gene class. In general, its going to be best to model you're problem domain as closely as possible with the built-in genes, but sometimes you may need a specialized gene type. In rust, this is done by implementing the Gene trait for your custom struct, while in python you can subclass the AnyGene class.
AnyGene follows how polars handles custom data types by using a generic Any type. This means that you can create genes that hold any type of data you want, as long as its a valid python object.
Simply inherit from rd.AnyGene and implement the necessary methods such as __init__, __repr__, and any other methods you may need for your specific use case.
import radiate as rd
from datetime import datetime, timezone
rd.random.seed(42)
class ObjectGene(rd.AnyGene):
def __init__(self):
self.number = rd.random.int(min=0, max=10)
self.date = datetime(2020, 1, 1, tzinfo=timezone.utc)
def __repr__(self):
return f"ObjectGene(number={self.number}, date={self.date})"
def fitness_function(phenotypes: list[list[ObjectGene]]) -> list[float]:
return [sum(gene.number for gene in individual) for individual in phenotypes]
engine = rd.GeneticEngine(
rd.AnyCodec(ObjectGene() for _ in range(10)),
fitness_func=rd.BatchFitness(fitness_function),
objective="min",
)
result = engine.run(rd.ScoreLimit(0), ui=True)
for obj_gene in result.value():
print(obj_gene)
AnyGene is compatible with the rest of the radiate system, meaning alters and other components can work with it seamlessly.
Valid alters for AnyGene include:
MultiPointCrossoverUniformCrossoverShuffleCrossoverMeanCrossoverSwapMutatorScrambleMutatorInversionMutatorUniformMutatorArithmeticMutator
Converting between rust and python custom genes behind the scenes is a bit more involved that other genes, so reach for built-in genes when possible for best performance.