Question

我正在尝试找到函数Minimum of a function with BFGS method的最小值（PDF文档的第29页）

而且我没有得到与链接中报告的结果相同的结果，我已经尝试过没有jacobian而没有运气。任何帮助，将不胜感激。

到目前为止的代码：

import numpy as np
from scipy.optimize import minimize
def objective(x):
   x1=x[0]
   x2=x[1]
   print ("x1: ",x1,"    ","x2: ",x2)
   return pow(x1,4.0)-2*x2*pow(x1,2.0)+pow(x2,2.0)+pow(x1,2.0)-2.0*x1+5.0

def jacobiano(x):
  x1=x[0]
  x2=x[1]
  jaco=np.zeros(2)
  jaco[0]=4.0*x1-4.0*x2*x1+2.0*x1-2.0
  jaco[1]=-2.0*pow(x1,2.0)+2.0*x2
  print ("dx1: ",jaco[0],"    ","dx2: ",jaco[1])
  return jaco
x0=np.array([1.0,2.0], dtype=np.double)
print(objective(x0))
sol=minimize(objective,x0,method='BFGS',jac=jacobiano, options={'disp': True})
print(sol)

Answer 1

问题出现是因为您错误地计算了雅可比行列式，在您的情况下df/dx1不正确。

如果f = x1**4 -2*x2*x1**2 +x2**2+ x1**2 -2.0*x1+5.0

然后df/dx1 = 4.0*x1**3 -4.0*x2*x1 + 2.0*x1-2.0

import numpy as np
from scipy.optimize import minimize

def objective(x):
   x1, x2 = x
   print ("x1: ",x1,"    ","x2: ",x2)

   return x1**4 -2*x2*x1**2 +x2**2+ x1**2 -2.0*x1+5.0


def jacobiano(x):
  x1, x2 = x
  jaco=np.zeros(2)
  jaco[0]=4.0*x1**3 -4.0*x2*x1 + 2.0*x1-2.0
  jaco[1]=-2.0*x1**2.+2.0*x2

  print("dx1: ",jaco[0],"    ","dx2: ",jaco[1])
  return jaco

x0=np.array([1.0,2.0], dtype=np.double)

sol=minimize(objective,
 x0,method='BFGS',jac=jacobiano, options={'disp': True})
print(sol)

输出：

Optimization terminated successfully.
         Current function value: 4.000000
         Iterations: 7
         Function evaluations: 9
         Gradient evaluations: 9
      fun: 4.000000000002963
 hess_inv: array([[ 0.50324351,  1.0154575 ],
       [ 1.0154575 ,  2.55695728]])
      jac: array([  7.65547714e-06,  -2.90129716e-06])
  message: 'Optimization terminated successfully.'
     nfev: 9
      nit: 7
     njev: 9
   status: 0
  success: True
        x: array([ 1.00000093,  1.0000004 ])

Matlab：

x1=1.00863, x2=1.01932, f=4.00008

<强>的Python：

x1=1.00000093, x2=1.0000004, f=4.000000000002963

最佳解决方案

x1=1.0, x2=1.0, f=4.0

为什么我没有得到与Python中的BFGS最小化相同的结果与MATLAB中的结果相同？

1 个答案: