import torch
from evox.utils import maximum, minimum
[docs]
def polynomial_mutation(
x: torch.Tensor,
lb: torch.Tensor,
ub: torch.Tensor,
pro_m: float = 1,
dis_m: float = 20,
) -> torch.Tensor:
"""Polynomial mutation.
Inspired by PlatEMO.
:param x: The input population (size: n x d).
:param lb: The lower bounds for the decision variables.
:param ub: The upper bounds for the decision variables.
:param pro_m: Probability of mutation.
:param dis_m: The distribution index for polynomial mutation.
:return: The mutated population. (size: n x d)
"""
n, d = x.size()
# Random numbers for mutation
site = torch.rand(n, d, device=x.device) < pro_m / d
mu = torch.rand(n, d, device=x.device)
# Apply mutation for the first part where mu <= 0.5
temp = site & (mu <= 0.5)
lower = lb
upper = ub
pop_dec = maximum(minimum(x, upper), lower)
norm = torch.where(temp, (pop_dec - lower) / (upper - lower), 0.0)
pop_dec = torch.where(
temp,
pop_dec
+ (upper - lower)
* (
torch.pow(
2 * mu + (1 - 2 * mu) * torch.pow(1 - norm, dis_m + 1),
1 / (dis_m + 1),
)
- 1
),
pop_dec,
)
# Apply mutation for the second part where mu > 0.5
temp = site & (mu > 0.5)
norm = torch.where(temp, (upper - pop_dec) / (upper - lower), torch.zeros_like(pop_dec))
pop_dec = torch.where(
temp,
pop_dec
+ (upper - lower)
* (
1
- torch.pow(
2 * (1 - mu) + 2 * (mu - 0.5) * torch.pow(1 - norm, dis_m + 1),
1 / (dis_m + 1),
)
),
pop_dec,
)
return pop_dec
EvoX