使用ctypes封装C代码后调用Scipy direct优化遇类型错误求助

我已用ctypes将C实现的styblinski_tang函数封装到Python中，单独调用（如输入1,1）可得到正确结果-10。但用Scipy的direct方法优化时，先出现TypeError: this function takes at least 2 arguments (1 given)错误；添加Python包装函数后，又触发ctypes参数类型错误。

初始Python代码（ctype_tutorial.py） #

import ctypes
from ctypes import *
import numpy as np
from numpy import ctypeslib
from scipy.optimize import minimize
from scipy.optimize import direct, Bounds, basinhopping

clibrary = ctypes.CDLL(r"C:\Users\mhaja\OneDrive\Desktop\Github\Honours\ctypes\clibrary.so")

styblinski_tang = clibrary.styblinski_tang

styblinski_tang.argtypes = [ctypes.c_int, ctypes.c_int]
styblinski_tang.restype = ctypes.c_int


print(styblinski_tang(1, 1))

result = direct(styblinski_tang, [[-4,4], [-4,4]])
print(result.x, result.fun, result.nfev)

C代码（clibrary.c） #

#include <stdio.h>
#include <math.h>

int styblinski_tang(int x, int y){
    return 0.5 * (pow(x, 4) - 16*pow(x, 2) + 5*x + pow(y, 4) - 16*pow(y, 2) + 5*y);
}

初始报错信息 #

-10
Traceback (most recent call last):
  File "c:\Users\mhaja\OneDrive\Desktop\Github\Honours\ctypes\ctypes_tutorial.py", line 44, in <module>
    result = direct(styblinski_tang, [[-4,4], [-4,4]])
  File "C:\Users\mhaja\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\scipy\optimize\_direct_py.py", line 256, in direct
    x, fun, ret_code, nfev, nit = _direct(
  File "C:\Users\mhaja\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\scipy\optimize\_direct_py.py", line 249, in _func_wrap
    f = func(x)
TypeError: this function takes at least 2 arguments (1 given)

添加包装函数后的Python代码 #

import ctypes
from ctypes import *
import numpy as np
from numpy import ctypeslib
from scipy.optimize import minimize
from scipy.optimize import direct, Bounds, basinhopping

clibrary = ctypes.CDLL(r"C:\Users\mhaja\OneDrive\Desktop\Github\Honours\ctypes\clibrary.so")

styblinski_tang = clibrary.styblinski_tang

styblinski_tang.argtypes = [ctypes.c_int, ctypes.c_int]
styblinski_tang.restype = ctypes.c_int


print(styblinski_tang(1, 1))

# 包装函数
def foo(xy):
    x,y = xy
    return styblinski_tang(x, y)

print(foo([1,1]))

result = direct(foo, [[-4,4], [-4,4]])
print(result.x, result.fun, result.nfev)

更新后报错信息 #

-10
-10
Traceback (most recent call last):
  File "c:\Users\mhaja\OneDrive\Desktop\Github\Honours\ctypes\ctypes_tutorial.py", line 51, in <module>
    result = direct(foo, [[-4,4], [-4,4]])
  File "C:\Users\mhaja\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\scipy\optimize\_direct_py.py", line 256, in direct
    x, fun, ret_code, nfev, nit = _direct(
  File "C:\Users\mhaja\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\scipy\optimize\_direct_py.py", line 249, in _func_wrap
    f = func(x)
  File "c:\Users\mhaja\OneDrive\Desktop\Github\Honours\ctypes\ctypes_tutorial.py", line 47, in foo
    return styblinski_tang(x, y)
ctypes.ArgumentError: argument 1: TypeError: wrong type

问题根源 #

1. Scipy的direct函数会将优化变量打包为单个numpy数组传入目标函数，但你的C封装函数需要两个独立的int参数，导致参数数量不匹配。
2. 添加包装函数后，direct传入的是numpy浮点类型元素，而ctypes要求严格匹配argtypes的c_int类型，直接传递numpy数值会触发类型错误。
3. C函数返回值用int类型，但计算结果是浮点数，会被截断，导致精度丢失。

修正步骤 #

1. 修改C代码，改用浮点类型

将参数和返回值改为double，避免精度损失：

#include <stdio.h>
#include <math.h>

double styblinski_tang(double x, double y){
    return 0.5 * (pow(x, 4) - 16*pow(x, 2) + 5*x + pow(y, 4) - 16*pow(y, 2) + 5*y);
}

2. 修正Python封装与包装函数

更新ctypes的参数/返回类型，同时在包装函数中将numpy数组元素转换为ctypes兼容类型：

import ctypes
import numpy as np
from scipy.optimize import direct

clibrary = ctypes.CDLL(r"C:\Users\mhaja\OneDrive\Desktop\Github\Honours\ctypes\clibrary.so")

styblinski_tang = clibrary.styblinski_tang
# 更新为double类型
styblinski_tang.argtypes = [ctypes.c_double, ctypes.c_double]
styblinski_tang.restype = ctypes.c_double

print(styblinski_tang(1.0, 1.0))  # 输出-10.0，保留浮点精度

def foo(xy):
    # 显式转换为ctypes兼容类型
    x = ctypes.c_double(xy[0])
    y = ctypes.c_double(xy[1])
    return styblinski_tang(x, y)

print(foo([1.0,1.0]))

result = direct(foo, [[-4,4], [-4,4]])
print(result.x, result.fun, result.nfev)

补充说明 #

• 优化算法通常需要浮点变量，C函数改用double更合理，避免整数截断导致的精度问题。
• 包装函数必须显式转换numpy数值，因为direct传入的是numpyfloat64，ctypes不会自动隐式转换numpy数组元素。
• 若坚持使用整数参数，只需将包装函数中的转换改为ctypes.c_int(int(xy[0]))，同时保持C函数的int类型，但优化结果会受限（仅能取整数点）。

内容的提问来源于stack exchange，提问作者mehrad hajati