在Cython中包装GSL时出错:无法将Python对象转换为'double(*)(double,void *)nogil'

时间:2015-08-22 10:53:49

标签: python c numpy cython numerical-recipes

首先,请原谅我对编程术语的不良运用以及糟糕(甚至完全错误)的编程习惯;我仍然想找到我的脚: - )

总之,我正在尝试在Numerical Recipes,3rd Edition,pgs中编写多维集成例程的一个版本。 198-199。在Cython中,使用GSL numerical integration library。我已经复制了以下NR书中的相关片段:

struct NRf3 {
double xsav, ysav;
double (*func3d)(const double, const double, const double);
double operator()(const double z)
{
    return func3d(xsav, ysav, z);
}
    }

struct NRf2 {
    NRf3 f3;
    double (*z1)(double, double);
    double (*z2)(double, double);
    NRf2(double zz1(double, double), double zz2(double, double)):    z1(zz1), z2(zz2)} {}
    double operator()(const double y)
    {
        f3.ysav = y;
        return qgaus(f3, z1(f3.xsav, y), z2(f3.xsav, y));
    }
}

struct NRf1 {
    double (*y1)(double);
    double (*y2)(double);
    NRf2 f2;
    NRf1(double yy1(double), yy2(double), z1(double, double), z2(double,  double)) : y1(yy1), y2(yy2), f2(z1, z2) {}
    double operator()(const double x){
        f2.f3.xsav = x;
        return qgaus(f2, y1(x), y2(x));
    }
}

template <class T>
double quad3d(T &func, const double x1, const double x2, double     y1(double), double y2(double), double z1(double, double), double z2(double, double))
{
    NRf1 f1(y1, y2, z1, z2);
    f1.f2.f3.func3d = func;
    return qgaus(f1, x1, x2)
}

此脚本确实应用了先前编写的集成例程qgaus(稍后我将用GSL中的例程替换),该例程采用双重函数和两个表示集成限制的双精度作为输入参数。

在数学上,代码执行公式中显示的集成。这个link的(4.6.2)。积分限制y [i]和z [i]分别具有(x)和(x,y)依赖性,因为积分在(x,y)平面中的预定义区域上运行(参见图4.6.1)上面相同的链接)。

我自己在Cython中的尝试如下(注意我的问题,我认为,与GSL的接口无关,但仍然处于正确的Cythoning级别):

from cython_gsl cimport *
from numpy import *

ctypedef double * double_ptr
ctypedef void * void_ptr

cdef double sigma = 0.045
cdef double mu = 0.37

## Test : a test function in x, y, z
    cdef double gauss(double x, double y, double z, void * params) nogil:
    cdef double mu = (<double_ptr> params)[0]
    cdef double sigma = (<double_ptr> params)[1]
    return mu * x + sigma * y + z

cdef class NRf3:
    cdef double xsav, ysav
    cdef double (*func3d)(double x, double y, double z, void * params)

    def __init__(self, xsav, ysav):
        self.xsav = xsav
        self.ysav = ysav

    def func(self, double z):
        xsav = self.xsav
        ysav = self.ysav
        return func3d(xsav, ysav, z, void * params)

## Calling the gsl integrator qags
## Function: int gsl_integration_qags (const gsl_function * f, double a,     double b, double epsabs, double epsrel, size_t limit,     gsl_integration_workspace * workspace, double * result, double * abserr)

cdef class NRf2:
    cdef NRf3 f3
    cdef double z1
    cdef double z2
    cdef double paramsf2[1]
    cdef size_t nevalf2

    def __init__(self, double z1, double z2):
        self.z1 = z1
        self.z2 = z2

    def __call__(self, double y):
        cdef double resultf2, errorf2
        cdef gsl_integration_workspace * W
        W = gsl_integration_workspace_alloc(1000)
        cdef gsl_function F2
        z1 = self.z1
        z2 = self.z2

        f3.ysav = y
        f3_temp = f3(f3.xsav, y)
        F2.function = f3_temp
        F2.params = self.paramsf2
        gsl_integration_qags(&F2, z1, z2, 1e-2, 1e-2, 1000, W, &resultf2, &errorf2) # this line calls the GSL routine
        gsl_integration_workspace_free(W) 
        return resultf2

cdef class NRf1:
    cdef NRf2 f2
    cdef double y1
    cdef double y2
    cdef double paramsf1[1]
    cdef size_t nevalf1

    def __init__(self, double yy1, double yy2, double z1, double z2):
        self.y1 = yy1
        self.y2 = yy2
        self.z1 = z1
        self.z2 = z2
        self.f2 = f2(self.z1, self.z2)

    def __call__(self, func, double x):
        cdef double result, error
        cdef gsl_integration_workspace * W
        W = gsl_integration_workspace_alloc(1000)
        cdef gsl_function F1
        f2 = self.f2

        f2.f3.func3d = func
        f2.f3.xsav = x
        y1 = self.y1
        y2 = self.y2
        F1.function = f2
        F1.params = self.paramsf1
        gsl_integration_qags(&F1, y1, y2, 1e-2, 1e-2, 1000, W, &result, &error)
        gsl_integration_workspace_free(W)
        return result

使用通常的

编译此脚本
python min_example.py build_ext --inplace
在Cython网站上找到的

会出现以下错误:

Error compiling Cython file:
------------------------------------------------------------ ...

        z1 = self.z1
        z2 = self.z2

        f3.ysav = y
        f3_temp = f3(f3.xsav, y)
        F2.function = f3_temp
                            ^
------------------------------------------------------------

min_example.pyx:61:29: Cannot convert Python object to 'double
(*)(double, void *) nogil'

Error compiling Cython file:
------------------------------------------------------------ ...

        f2.f3.func3d = func
        f2.f3.xsav = x
        y1 = self.y1
        y2 = self.y2
        F1.function = f2
                       ^
------------------------------------------------------------

min_example.pyx:92:24: Cannot convert Python object to 'double
(*)(double, void *) nogil'

错误分别指代行F2.function = f3_tempF1.function = f2。在这个例子中,gsl_integration_qags是来自gsl的库,但是出于这个问题的目的,接口的正确性并不重要,因为(我认为)问题在于我使用的Cython语法不正确。

我的直觉告诉我,这是一个非常琐碎的新手错误,但我一直无法找到原因。任何输入(关于问题,我的代码一般)都会非常受欢迎。

编辑1:反映实际的错误消息

编辑2:编辑标题以反映错误消息的性质

编辑3 :我应该说我明白错误来自于我尝试分配Python实例对象,即行F2.function = f3_tempF1.function = f2,到struct成员,这是一个带有两个参数的函数。因为这似乎是从C到Cython的相当简单的代码移植,并假设一个人紧紧抓住原始的C实现,我只是想知道如何正确地写这些行。

编辑4 :有点搜索会出现this个帖子。我认为我的困惑可以用同样的方式提炼:如何将Python类成员(如果我已经正确实例化)传递给C函数?

0 个答案:

没有答案