import numpy as np
from random import shuffle
#from numba import njit
[docs]class SBX_xover:
"""Simulated binary crossover.
Parameters
----------
ProC : float, optional
[description], by default 1
DisC : float, optional
[description], by default 30
"""
def __init__(self, ProC: float = 1, DisC: float = 30):
"""[summary]
"""
self.ProC = ProC
self.DisC = DisC
[docs] def do(self, pop: np.ndarray, mating_pop_ids: list = None) -> np.ndarray:
"""Consecutive members of mating_pop_ids are crossed over
in pairs. Example: if mating_pop_ids = [0, 2, 3, 6, 5] then the individuals
are crossover as: [0, 2], [3, 6], [5, 0]. Note: if the number of elements
is odd, the last individual is crossed over with the first one.
Parameters
----------
pop : np.ndarray
Array of all individuals
mating_pop_ids : list, optional
Indices of population members to mate, by default None, which shuffles and
mates whole population
Returns
-------
np.ndarray
The offspring produced as a result of crossover.
"""
pop_size, num_var = pop.shape
if mating_pop_ids is None:
shuffled_ids = list(range(pop_size))
shuffle(shuffled_ids)
else:
shuffled_ids = mating_pop_ids
mating_pop = pop[shuffled_ids]
mate_size = len(shuffled_ids)
if len(shuffled_ids) % 2 == 1:
# Maybe it should be pop_size-1?
mating_pop = np.vstack((mating_pop, mating_pop[0]))
mate_size = mate_size + 1
# The rest closely follows the matlab code.
offspring = np.zeros_like(mating_pop) # empty_like() more efficient?
for i in range(0, mate_size, 2):
beta = np.zeros(num_var)
miu = np.random.rand(num_var)
beta[miu <= 0.5] = (2 * miu[miu <= 0.5]) ** (1 / (self.DisC + 1))
beta[miu > 0.5] = (2 - 2 * miu[miu > 0.5]) ** (-1 / (self.DisC + 1))
beta = beta * ((-1) ** np.random.randint(0, high=2, size=num_var))
beta[np.random.rand(num_var) > self.ProC] = 1 # It was in matlab code
avg = (mating_pop[i] + mating_pop[i + 1]) / 2
diff = (mating_pop[i] - mating_pop[i + 1]) / 2
offspring[i] = avg + beta * diff
offspring[i + 1] = avg - beta * diff
return offspring
# TODO: Make sure the following works correctly, then replace the code above.
"""@njit
def create_offsprings(pop:np.ndarray, mate_size:int,ProC: float=1, DisC: float=30):
pop_size, num_var = pop.shape
beta = np.zeros(num_var)
offspring = np.zeros_like(pop)
for i in range(0, mate_size, 2):
beta[:] = 0
miu = np.random.rand(num_var)
beta[miu <= 0.5] = (2 * miu[miu <= 0.5]) ** (1 / (DisC + 1))
beta[miu > 0.5] = (2 - 2 * miu[miu > 0.5]) ** (-1 / (DisC + 1))
beta = beta * ((-1) ** np.random.randint(0, high=2, size=num_var))
beta[np.random.rand(num_var) > ProC] = 1 # It was in matlab code
avg = (pop[i] + pop[i + 1]) / 2
diff = (pop[i] - pop[i + 1]) / 2
offspring[i] = avg + beta * diff
offspring[i + 1] = avg - beta * diff
return offspring"""
