GeneticEngine
GeneticEngine is a framework for using Genetic Programming, empowered by Domain Knowledge, provided by the user.
In GeneticEngine, the user specifies the intended shape of their program using type annotations. Terminals and non-terminals can be defined using regular classes (although we recommend dataclasses), while abstract classes can be used to specify different options.
class MyExpr(ABC):
def eval(self):
...
@dataclass
class Plus(MyExpr):
left: MyExpr
right: MyExpr
def eval(self):
return self.left.eval() + self.right.eval()
@dataclass
class Literal(MyExpr):
value: int
def eval(self):
return self.value
In this example, we are defining the language that supports the plus operator and integer literals. GeneticEngine will be able to automatically generate all possible expressions, such as Plus(left=Plus(left=Literal(12), right=Literal(12)), right=Literal(15)). You can override __str__ to have a more readable version of the tree.
From the definition of a grammar, and a fitness function like the one below, GeneticEngine can search for an expression that minimizes (or maximizes) the given function.
def fitness_function(e:MyExpr) -> int
return abs(42 - e.eval())
And then you can the program using:
grammar = extract_grammar([Literal, Plus], MyExpr)
alg = SimpleGP(
grammar=grammar,
fitness_function=fitness_function,
minimize=True,
population_size=200,
max_evaluations=10000,
)
(b, bf) = alg.search()
print(bf, b)
To learn more about how to use Genetic Engine, you can follow the tutorial.
If you want to use GeneticEngine as a classifier or a regressor, you can use our sklearn-compatible api.