import numpy as np
from random import choice, sample
[docs]def mutate(offspring, individuals, params, *args):
"""Perform BioGP mutation functions.
Standard mutation:
Randomly select and regrow a subtree of an individual.
Small mutation:
Randomly select a node within a tree and replace it with either a function of the
same arity,
or another value from the terminal set.
Mono parental:
Randomly swap two subtrees within the same individual.
Parameters
----------
offspring : list
List of individuals to mutate.
individuals : list
List of all individuals.
params : dict
Parameters for breeding. If None, use defaults.
"""
prob_mut = params.get("prob_mutation", 0.3)
prob_stand = 1 / 3 * prob_mut
prob_point = 1 / 3 * prob_mut
prob_mono = prob_mut - prob_stand - prob_point
prob_replace = prob_mut
r = np.random.rand()
for ind in offspring:
if r <= prob_stand:
# Standard mutation
#
# This picks a random subtree anywhere within the tree
rand_node = choice(ind.nodes[1:])
tree = ind.grow_tree(method="grow", depth=rand_node.depth, ind=rand_node)
rand_node.value = tree.value
rand_node.roots = tree.roots
# This picks a whole subtree at depth=1 under the linear node
# rand_subtree = np.random.randint(len(ind.roots))
# del ind.roots[rand_subtree]
# ind.grow_tree(method="grow", ind=ind)
ind.nodes = ind.get_sub_nodes()
elif r <= prob_point + prob_stand:
# Small mutation
for node in ind.nodes[1:]:
if np.random.rand() < prob_replace and callable(node.value):
value = choice(node.function_set)
while node.value.__code__.co_argcount != value.__code__.co_argcount:
value = choice(node.function_set)
node.value = value
elif np.random.rand() < prob_replace:
node.value = choice(node.terminal_set)
ind.nodes = ind.get_sub_nodes()
elif r <= prob_mono + prob_point + prob_stand:
# Mono parental
swap_nodes = sample(ind.nodes[1:], 2)
tmp_value = swap_nodes[0].value
tmp_roots = swap_nodes[0].roots
swap_nodes[0].value = swap_nodes[1].value
swap_nodes[0].roots = swap_nodes[1].roots
swap_nodes[1].value = tmp_value
swap_nodes[1].roots = tmp_roots
ind.nodes = ind.get_sub_nodes()
else:
pass
[docs]class BioGP_mutation:
def __init__(self, probability_mutation: float):
self.probability_mutation: float = probability_mutation
self.probability_stand = probability_mutation / 3
self.probability_point = probability_mutation / 3
self.probability_mono = probability_mutation / 3
self.probability_replace = probability_mutation
[docs] def do(self, offspring):
r = np.random.rand()
for ind in offspring[:, 0]:
if r <= self.probability_stand:
# Standard mutation
#
# This picks a random subtree anywhere within the tree
rand_node = choice(ind.nodes[1:])
tree = ind.grow_tree(
method="grow", depth=rand_node.depth, ind=rand_node
)
rand_node.value = tree.value
rand_node.roots = tree.roots
# This picks a whole subtree at depth=1 under the linear node
# rand_subtree = np.random.randint(len(ind.roots))
# del ind.roots[rand_subtree]
# ind.grow_tree(method="grow", ind=ind)
ind.nodes = ind.get_sub_nodes()
elif r <= self.probability_point + self.probability_stand:
# Small mutation
for node in ind.nodes[1:]:
if np.random.rand() < self.probability_replace and callable(
node.value
):
value = choice(node.function_set)
while (
node.value.__code__.co_argcount
!= value.__code__.co_argcount
):
value = choice(node.function_set)
node.value = value
elif np.random.rand() < self.probability_replace:
node.value = choice(node.terminal_set)
ind.nodes = ind.get_sub_nodes()
elif r <= self.probability_mutation:
# Mono parental
swap_nodes = sample(ind.nodes[1:], 2)
tmp_value = swap_nodes[0].value
tmp_roots = swap_nodes[0].roots
swap_nodes[0].value = swap_nodes[1].value
swap_nodes[0].roots = swap_nodes[1].roots
swap_nodes[1].value = tmp_value
swap_nodes[1].roots = tmp_roots
ind.nodes = ind.get_sub_nodes()
else:
pass
return np.asarray(offspring)
