Genetic Programming

Construct expression trees and evolve symbolic models using TreeSpace, PrimitiveSet, and GP.
Example Code

import random
import numpy as np

from opytimizer import Opytimizer
from opytimizer.core.function import Function
from opytimizer.core.graph.primitive_set import PrimitiveSet
from opytimizer.core.stopping import MaxIterations
from opytimizer.optimizers.single_objective.evolutionary.gp import GP
from opytimizer.spaces.tree import TreeSpace
from opytimizer.visualization import plot_graph


class ArrayType:
    pass


SEED = 100
np.random.seed(SEED)
random.seed(SEED)


def protected_exp(a):
    return np.exp(np.clip(a, -10, 10))


def protected_log(a):
    return np.log(np.abs(a) + 1e-5)


def custom_if_gt0(cond, val_if_pos, val_if_neg):
    return np.where(cond > 0, val_if_pos, val_if_neg)


X_data = np.linspace(-3, 3, 100)
Y_target = np.cos(X_data) + protected_exp(-(X_data**2))

pset = PrimitiveSet(name="TrigRegression", root_type=ArrayType)

pset.add_primitive(np.sin, (ArrayType,), ArrayType, name="sin")
pset.add_primitive(protected_exp, (ArrayType,), ArrayType, name="exp")
pset.add_primitive(protected_log, (ArrayType,), ArrayType, name="log")

pset.add_primitive(np.add, (ArrayType, ArrayType), ArrayType, name="add")
pset.add_primitive(np.multiply, (ArrayType, ArrayType), ArrayType, name="mul")

pset.add_primitive(
    custom_if_gt0, (ArrayType, ArrayType, ArrayType), ArrayType, name="if_gt0"
)

pset.add_terminal(value=X_data, output_type=ArrayType, name="X")
pset.add_terminal(value=np.ones_like(X_data), output_type=ArrayType, name="1.0")

pset.validate()


def evaluate_trig(tree):
    try:
        y_pred = tree.evaluate()

        mae = np.mean(np.abs(y_pred - Y_target))
        penalty = tree.depth * 0.01
        return mae + penalty
    except Exception:
        return np.inf


function = Function(evaluate_trig)

space = TreeSpace(
    n_agents=10,
    n_objectives=1,
    pset=pset,
    min_depth=2,
    max_depth=4,
    method="half_and_half",
)

optimizer = GP()

opt = Opytimizer(space=space, function=function, optimizer=optimizer)

opt.start(MaxIterations(1000))

print(opt.space.best_agent.position)
Example Code​

Example Code
