您好，我对您给出的demo，做了一些改动，把自变量个数设为1。

def schaffer(p):
    x = p
    # return 0.5 + (np.sin(x) - 0.5) / np.square(1 + 0.001 * x)
    return x + 10 * np.sin(5 * x) + 7 * np.cos(4 * x)

 ga = GA(func=schaffer, n_dim=1, size_pop=50, max_iter=800, lb=[-1], ub=[1], precision=1e-7)

但是提示报错：

Traceback (most recent call last): File "...", line 25, in best_x, best_y = ga.run() File "...", line 80, in run self.crossover() File "...\venv\lib\site-packages\sko\operators\crossover.py", line 44, in crossover_2point_bit Chrom1 ^= mask2 ValueError: non-broadcastable output operand with shape (25,1,25) doesn't match the broadcast shape (25,25,25)

我理解的schaffer是用来定义自己的目标函数的这样对吗？ 那么为什么dim不能设为1呢？ 另外您给的例子中

    x1, x2 = p
    x = np.square(x1) + np.square(x2)
    return 0.5 + (np.sin(x) - 0.5) / np.square(1 + 0.001 * x)

如果x1,x2为自变量的话，为什么要取一个x为中间量？ 仅仅是为了方便书写吗？ 感谢！

非常感谢您的及时回答。 您好，我使用demo中约束关系列表为差分算法施加了约束关系，但是计算出来的结果好像约束关系没有发挥作用。 constraint_ueq = [] for i in range(16): for j in range(15): constraint_ueq.append(lambda x: x[0 + j+i16] - x[1 + j+i16]) for i in range(16): for j in range(15): constraint_ueq.append(lambda x:x[0+j16+i] - x[16+j16+i]) for i in range(6): constraint_ueq.append(lambda x:x[256+i] - x[257+i]) for i in range(11): constraint_ueq.append(lambda x:x[264+i] - x[263+i]) 我在做一个汽车系统的自动标定，涉及到一个256的曲面和一个7的曲线，12的曲线。 曲面自身和曲线自身有平顺性的要求（沿x轴和y轴递增），所以我用以上关系施加了不等式约束，但是优化出来的结果，好像没有反应出这种约束关系。我查看了差分算法的源代码，使用的是罚函数做约束施加？代码如下？ if not self.has_constraint: self.Y = self.Y_raw else: # constraint penalty_eq = np.array([np.sum(np.abs([c_i(x) for c_i in self.constraint_eq])) for x in self.X]) penalty_ueq = np.array([np.sum(np.abs([max(0, c_i(x)) for c_i in self.constraint_ueq])) for x in self.X]) self.Y = self.Y_raw + 1e5 * penalty_eq + 1e5 * penalty_ueq 请问如何让我的约束关系在算法的种群生成，变异、交叉、选择中发挥作用？ 非常感谢您的及时回答。

在我的程序中，一共有15个变量，整数规划取值范围是从0~1363，因此我设置lb与ub未0、1363，但是我发现程序会生成大于1363的数字，例如生成了 IndexError: index 1901 is out of bounds for axis 0 with size 1363 但是我发现只要把ub边界设置小于1000，就没有问题

from sko.GA import GA
K=15
ga = GA(func=cost_func, n_dim=K, max_iter=500, lb=[0 for i in range(K)], ub=[1363 for i in range(K)], precision=[1 for i in range(K)])
rs = ga.run()
print(rs)

Hi, thanks for the package with multiple metaheuristics which can be implemented with ease. I had a doubt whether we can define our own Vehicle routing problems with constraints like time windows and use this package. Or we have to define our own custom functions? Thanks!

Hi,

I would like to use GA for statsmodels sarimax function. I was wondering, how would it be possible to parallelize the fitting process of each individual in GA using torch. Crossovers, selections are parallelized however the function part is left out for such functionality. I would be glad if you can help me.

` import torch import numpy as np import pandas as pd from scipy.stats import norm import statsmodels.api as sm import matplotlib.pyplot as plt from datetime import datetime import requests from io import BytesIO

# Dataset
wpi1 = requests.get('https://www.stata-press.com/data/r12/wpi1.dta').content
data = pd.read_stata(BytesIO(wpi1))
data.index = data.t
# Set the frequency
data.index.freq="QS-OCT"

def arima(X):
      X = [int(x) for x in X]
      order = [X[0],X[1],X[2]]

      mod = sm.tsa.statespace.SARIMAX(data['wpi'], 
                                      order=order, 
                                      seasonal_order = [X[3],X[4],X[5],6],
                                      enforce_stationarity=False,
                                      enforce_invertibility=False
                                      )
      res = mod.fit(disp=False)
      #print(X)
      #print(res.aic)
      return res.aic



ga = GA(func=arima, n_dim=6, size_pop=50, max_iter=50, lb=[0,0,0,0,0,0], ub=[6,4,6,4,3,4], precision=1)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

ga.to(device=device)

ga.run()

`

demo_func=lambda x: x[0]**2 + x[1]**2 + x[2]**2
ga = GA(func=demo_func,n_dim=3, max_iter=500, lb=[-1, -10, -5], ub=[2, 10, 2])
best_x, best_y = ga.fit()

如果 x[0] 只能取整数, 代码应该怎么写? 谢谢

我外层搜索的目标函数是这样写的，我需要根据ant.step4（）得到的结果（存储在ant的属性当中）再进行一次搜索，请问这个应该如何去写呢？ def obj_func(x): jobs=read_data(Const.filename)#读取作业信息 ant = Antibody(jobs, x)#创建抗体实例 ant.step1()#按照随机键进行分组 ant.step2()#根据分组计算发车时间 ant.step3() ant.step4() #ant_a=Antibody_a(ant) ant.step5() ant.cal_obj()#计算该抗体目标函数值 return ant.obj

我希望把目标函数放在对象里面，但是参数部分写入self不行，我需要传入一些成员变量到目标函数里面的对象进行计算。请问应该怎么写。

本人的目标函数是这样写的

    def cal_score(self,p):
        print(p)
        week_or_month, frequency_week, start_week, start_month_day, target_profit, sell_share = p
        player = Player("bot", self.money, self.funds_objects)
        fund_ids = self.funds_objects.keys()
        for fund_id in fund_ids:
            player.simulation(fund_id, self.input_start_time, self.input_end_time, self.week_or_month, self.frequency_week, self.target_profit,self.sell_share, self.start_week, self.start_month_day)
        player.caculate_score(self.target_profit)
        score = player.score
        average_yield = player.average_yield
        print("score", score)
        print("average_yield", average_yield)
        return -score

希望可以兼容self参数，这样可以更容易放到类里面使用

官方的示例是这样的， def demo_func(x): x1, x2, x3 = x return x1**2+x2*1+x3 如果我的目标函数是类似于这样的： def demo_func(x,model,y): x1,x2,x3=x model,y 是我要传递的其他参数，请问该怎么写 PSO(func=demo_func, dim=3, pop=40, max_iter=150, lb=[0, -1, 0.5], ub=[1, 1, 1], w=0.8, c1=0.5, c2=0.5) 这个里面的demo_func.

使用scikit-opt的遗传算法模型解决TSP问题时，效果好像不理想，比如stplib数据库里面的att48.tsp时，跑出来的结果跟随机的顺序差不多，att48的链接如下：http://elib.zib.de/pub/mp-testdata/tsp/tsplib/tsp/att48.tsp 我的参数设置为： ga_tsp = GA_TSP(func=cal_total_distance, n_dim=num_points, size_pop=300, max_iter=3000,prob_mut=0.1) 谢谢

我在算法内部又添加一层搜索之后就出现了这个报错，但是报错的是外层的搜索而不是内层的。问题是在selection这里，似乎是说对于np.where函数，条件矩阵和后面的矩阵维数不同。但是我将条件矩阵打印出来发现它就是一个（20，1）的矩阵，不知道为什么程序认定它为（400，1） [[ True] [ True] [ True] [ True] [ True] [False] [False] [ True] [ True] [False] [ True] [False] [ True] [ True] [False] [False] [ True] [ True] [False] [ True]] 迭代次数 0 最优值 [27.6] Traceback (most recent call last): File "D:/Desktop/半集成式VRPSDP/代码/main.py", line 121, in best_x, best_y = de.run() File "D:\841370\anaconda\lib\site-packages\sko\DE.py", line 86, in run self.selection() File "D:\841370\anaconda\lib\site-packages\sko\DE.py", line 76, in selection self.X = np.where((f_X < f_U).reshape(-1, 1), X, U) File "<array_function internals>", line 5, in where ValueError: operands could not be broadcast together with shapes (400,1) (20,30) (20,30)

此外还有一个报错VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray return np.array([func_cached(tuple(x)) for x in X])没有找到原因。

Would be nice to see the users be able to control the number of processors/threads used when defining 'multiprocessing' or 'multithreading'.

I would think you could pass in a worker argument (default can be None) to set_run_mode(), something like:

set_run_mode(obj_func, 'multiprocessing', n_workers=3)

..and then further down in the func_transformer() function:

    elif mode == 'multiprocessing':
        from multiprocessing import Pool
        
        pool = Pool()
        if n_workers != None:
            pool = Pool(n_workers)

        def func_transformed(X):
            return np.array(pool.map(func, X))

I'm not sure if this would be the appropriate or working syntax, but just an idea I had. Sometimes we don't want to just default to using ALL CPU cores when using multiprocessing.

蚁群算法似乎只实现了TSP问题的接口，没有实现连续域优化的方法，关于连续域的蚁群优化算法论文见参考文献

reference： Socha K, Dorigo M. Ant colony optimization for continuous domains[J]. European journal of operational research, 2008, 185(3): 1155-1173.

Using code from the SA_TSP example at: https://github.com/guofei9987/scikit-opt/blob/master/examples/demo_sa_tsp.py I am receiving significantly sub-optimal results.

import numpy as np
from sko.SA import SA_TSP
from scipy import spatial
import matplotlib.pyplot as plt
from matplotlib.ticker import FormatStrFormatter

num_points = 20
points_coordinate = np.random.rand(num_points, 2)

# using SA_TSP example code
num_points = points_coordinate.shape[0]
distance_matrix = spatial.distance.cdist(points_coordinate, points_coordinate, metric='euclidean')

def cal_total_distance(routine):
    '''The objective function. input routine, return total distance.
    cal_total_distance(np.arange(num_points))
    '''
    num_points, = routine.shape
    return sum([distance_matrix[routine[i % num_points], routine[(i + 1) % num_points]] for i in range(num_points)])

sa_tsp = SA_TSP(func=cal_total_distance, x0=range(num_points), T_max=100, T_min=1, L=10 * num_points)
best_points, best_distance = sa_tsp.run()

# Plot both solutions
fig, ax = plt.subplots(1, 2)

best_points_ = np.concatenate([best_points, [best_points[0]]])
best_points_coordinate = points_coordinate[best_points_, :]

ax[0].plot(sa_tsp.best_y_history)
ax[0].set_xlabel("Iteration")
ax[0].set_ylabel("Distance")
ax[0].set_title('Optimization')
ax[1].plot(best_points_coordinate[:, 0], best_points_coordinate[:, 1],
           marker='o', markerfacecolor='b', color='c', linestyle='-')
ax[1].xaxis.set_major_formatter(FormatStrFormatter('%.3f'))
ax[1].yaxis.set_major_formatter(FormatStrFormatter('%.3f'))
ax[1].set_xlabel("X")
ax[1].set_ylabel("Y")
ax[1].set_title('Coordinates')

plt.show()