参考官网:Scipy.
全局优化的目的是在可能存在许多局部最小值的情况下,在给定的范围内找到一个函数的全局最小值。通常情况下,全局优化器会有效地搜索参数空间,同时也会使用局部优化器(如最小化)。SciPy包含了许多优秀的全局优化器。在这里,我们将在同一目标函数上使用这些优化器,即
eggholder
函数。
eggholder函数
def eggholder(x):
return (-(x[1] + 47) * np.sin(np.sqrt(abs(x[0]/2 + (x[1] + 47))))
-x[0] * np.sin(np.sqrt(abs(x[0] - (x[1] + 47)))))
bounds = [(-512, 512), (-512, 512)]
查看其形状:
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
x = np.arange(-512, 513)
y = np.arange(-512, 513)
xgrid, ygrid = np.meshgrid(x, y)
xy = np.stack([xgrid, ygrid])
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.view_init(45, -45)
ax.plot_surface(xgrid, ygrid, eggholder(xy), cmap='terrain')
ax.set_xlabel('x')
ax.set_ylabel('y')
ax.set_zlabel('eggholder(x, y)')
plt.show()
使用全局优化
我们现在使用全局优化器来获得最小值和最小值的函数值。我们将把结果存储在一个字典里,这样我们以后就可以比较不同的优化结果。
from scipy import optimize
results = dict()
results['shgo'] = optimize.shgo(eggholder, bounds)
print(results['shgo'])
其结果为:
fun: -935.3379515604948
funl: array([-935.33795156])
message: 'Optimization terminated successfully.'
nfev: 47
nit: 2
nlfev: 42
nlhev: 0
nljev: 10
success: True
x: array([439.48097554, 453.97741501])
xl: array([[439.48097554, 453.97741501]])
另一种优化器:
results['DA'] = optimize.dual_annealing(eggholder, bounds)
print(results['DA'])
结果为:
fun: -959.6406627208503
message: ['Maximum number of iteration reached']
nfev: 4079
nhev: 0
nit: 1000
njev: 26
status: 0
success: True
x: array([512. , 404.23180442])
所有的优化器都会返回一个OptimizeResult
,除了解决方案之外,它还包含了函数评估的数量、优化是否成功等信息。
还有一种优化器为:
results['DE'] = optimize.differential_evolution(eggholder, bounds)
print(results['DE'])
其结果为:
fun: -894.578900390448
jac: array([1.13686748e-05, 1.13686748e-05])
message: 'Optimization terminated successfully.'
nfev: 588
nit: 18
success: True
x: array([-465.69411618, 385.71669541])
另外shgo
有第二个方法,它返回所有的局部最小值,而不是只返回它认为是全局最小值的值。
results['shgo_sobol'] = optimize.shgo(eggholder, bounds, n=200, iters=5,
sampling_method='sobol')
全部代码
import numpy as np
def eggholder(x):
return (-(x[1] + 47) * np.sin(np.sqrt(abs(x[0]/2 + (x[1] + 47))))
-x[0] * np.sin(np.sqrt(abs(x[0] - (x[1] + 47)))))
bounds = [(-512, 512), (-512, 512)]
from scipy import optimize
results = dict()
#results['shgo'] = optimize.shgo(eggholder, bounds)
#print(results['shgo'])
#results['DA'] = optimize.dual_annealing(eggholder, bounds)
#print(results['DA'])
#results['DE'] = optimize.differential_evolution(eggholder, bounds)
#print(results['DE'])
results['shgo_sobol'] = optimize.shgo(eggholder, bounds, n=200, iters=5,
sampling_method='sobol')
print(results['shgo_sobol'])