approx_fprime - ValueError: setting an array element with a sequence

Question

我在执行 approx_fprime

时收到错误（标题）

import numpy as np
import scipy.optimize as op
def J(x):
    return x*x

xk = np.arange(1,10).reshape(3,3).astype('float32')
print(op.approx_fprime(xk.ravel(), J, 0.01))

我做错了什么？

我必须更改函数，它应该 return 标量而不是向量。

def J(x): 
    return np.sum(x**2)

Answer 1

您正在推翻 approx_fprime 的假设：

Finite-difference approximation of the gradient of a scalar function.

Parameters:

...

f : callable

The function of which to determine the gradient (partial derivatives). Should take xk as first argument, other arguments to f can be supplied in *args. Should return a scalar, the value of the function at xk.

在本例中，您 return 是一个大小为 9 的向量。标量函数不应该 return。原因很简单：

x = np.array([1,2,3])
print((x*x).shape)
# (3,)

approx_fprime - ValueError: setting an array element with a sequence

approx_fprime - ValueError: setting an array element with a sequence

gradient

machine-learning

scipy

python-2.7

conv-neural-network