我的每一个遗传算法模型都以同样的方式变异

Question

我一直在研究一种遗传算法，我试图采用我的种群中最好的模型(基于得分)，并将整个种群纳入该模型。我已经成功地做到了这一点，但当我尝试分别变异每个模型时，所有模型都会变异为相同的参数。我知道这是因为我使用了一个对象，并将其克隆到一个列表中，但我不知道要更改什么，所以它不能以这种方式工作。下面我制作了一个在Python3.9中运行的可重现的例子。我知道代码不是特别小，但这是我能做的最小的代码。提前感谢，任何帮助都将不胜感激。

import torch
import torch.nn as nn

torch.manual_seed(0) #Reproducibility

population_size = 3 #Defining the size of my population
population = [nn.Linear(1,1) for _ in range(population_size)] #Initializing the population
input = torch.rand(1)# Creating dummy input
def mutate(m): #Function to mutate a model
    if type(m) == nn.Linear:
        m.weight = nn.Parameter(m.weight+torch.randn(m.weight.shape))
        m.bias = nn.Parameter(m.bias+torch.randn(m.bias.shape))

for i in population:
    print (i(input))

population = [x.apply(mutate) for x in population]

print ('\n')

for i in population:
    print (i(input))
#The above works as expected

#I want to fill my entire population with that model.
#I've been filling the population with the best model by doing the following:

best_model = population[0] #Say the first model in the list was the best performing one

population = [best_model for _ in range(population_size)] #This is the line I think needs to change, I just don't know what to change it to. 

#This does fill the population with my best model, but when I try to mutate it, every model is mutated to the same parameters

population = [x.apply(mutate) for x in population] #I know this is because I am using best_model while replacing the population, but I don't know how else to copy the model

for i in population:
    print (i(input)) #Just to show that the population all gives the same result

Answer 1

您可以制作模型的深层副本。确保为import copy，然后更改

population = [best_model for _ in range(population_size)]

至

population = [copy.deepcopy(best_model) for _ in range(population_size)]