问题
I am currently trying to optimize my Python program and got started with Cython in order to reduce the function calling overhead and perhaps later on include optimized C-libraries functions.
So I ran into the first problem:
I am using composition in my code to create a larger class. So far I have gotten one of my Python classes converted to Cython (which was difficult enough). Here's the code:
import numpy as np
cimport numpy as np
ctypedef np.float64_t dtype_t
ctypedef np.complex128_t cplxtype_t
ctypedef Py_ssize_t index_t
cdef class bendingForcesClass(object):
cdef dtype_t bendingRigidity
cdef np.ndarray matrixPrefactor
cdef np.ndarray bendingForces
def __init__(self, dtype_t bendingRigidity, np.ndarray[dtype_t, ndim=2] waveNumbersNorm):
self.bendingRigidity = bendingRigidity
self.matrixPrefactor = -self.bendingRigidity * waveNumbersNorm ** 2
cpdef np.ndarray calculate(self, np.ndarray membraneHeight):
cdef np.ndarray bendingForces
bendingForces = self.matrixPrefactor * membraneHeight
return bendingForces
From my composed Python/Cython class I am calling the class-method calculate, so that in my composed class I have the following (reduced) code:
from bendingForcesClass import bendingForcesClass
cdef class membraneClass(object):
def __init__(self, systemSideLength, lowerCutoffLength, bendingRigidity):
self.bendingForces = bendingForcesClass(bendingRigidity, self.waveNumbers.norm)
def calculateForces(self, heightR):
return self.bendingForces.calculate(heightR)
I have found out that cpdef makes the method/functions callable from Python and Cython, which is great and works, as long as I don't try to define the type of self.bendingForces beforehand - which according to the documentation (Early Binding For Speed) is necessary in order to remove the function-calling overhead. I have tried the following, which does not work:
from bendingForcesClass import bendingForcesClass
from bendingForcesClass cimport bendingForcesClass
cdef class membraneClass(object):
cdef bendingForcesClass bendingForces
def __init__(self, systemSideLength, lowerCutoffLength, bendingRigidity):
self.bendingForces = bendingForcesClass(bendingRigidity, self.waveNumbers.norm)
def calculateForces(self, heightR):
return self.bendingForces.calculate(heightR)
With this I get this error, when trying to build membraneClass.pyx with Cython:
membraneClass.pyx:18:6: 'bendingForcesClass' is not a type identifier
building 'membraneClass' extension
Note that the declarations are in two separate files, which makes this more difficult.
So I how do I get this done? I would be very thankful if someone could give me a pointer, as I can't find any information about this, besides the link given above.
Thanks and best regards!
回答1:
Disclaimer: This question is very old and I am not sure the current solution would work for 2011 Cython code.
In order to cimport an extension class (cdef class) from another file you need to provide a .pxd file (also known as a definitions file) declaring all C classes, attributes and methods. See Sharing Extension Types in the documentation for reference.
For your example, you would need a file bendingForcesClass.pxd, which declares the class you want to share, as well as all cimports, module level variables, typedefs, etc.:
# cimports
cimport numpy as np
# typedefy you want to share
ctypedef np.float64_t dtype_t
ctypedef np.complex128_t cplxtype_t
ctypedef Py_ssize_t index_t
cdef class bendingForcesClass:
# declare C attributes
cdef dtype_t bendingRigidity
cdef np.ndarray matrixPrefactor
cdef np.ndarray bendingForces
# declare C functions
cpdef np.ndarray calculate(self, np.ndarray membraneHeight)
# note that __init__ is missing, it is not a C (cdef) function
All imports, variables, and attributes that now are declared in the .pxd file can (and have to be) removed from the .pyx file:
import numpy as np
cdef class bendingForcesClass(object):
def __init__(self, dtype_t bendingRigidity, np.ndarray[dtype_t, ndim=2] waveNumbersNorm):
self.bendingRigidity = bendingRigidity
self.matrixPrefactor = -self.bendingRigidity * waveNumbersNorm ** 2
cpdef np.ndarray calculate(self, np.ndarray membraneHeight):
cdef np.ndarray bendingForces
bendingForces = self.matrixPrefactor * membraneHeight
return bendingForces
Now your cdef class bendingForcesClass can be cimported from other Cython modules, making it a valid type identifier, which should solve your problem.
回答2:
You need to use a declaration ".pxd" file and cimport. (Essentially, cimport happens at compile time, while import happens at run time so Cython can't make use of anything important).
Create "utils.pxd":
cdef class MyClass:
cdef readonly int field
cdef void go(self, int i)
"utils.pyx" now reads
cdef class MyClass:
def __init__(self, field):
self.field = field
cdef void go(self, int i):
self.field = i
all declarations which have been in the pyx file go into the .pxd file.
Then in mymodule.pyx
from utils import MyClass
from utils cimport MyClass
# other code follows...
// Extended answer from here: Cython: using imported class in a type declaration
回答3:
These are probably not the source of the error, but just to narrow down the problem, you might try to change the following:
Could it be that you are using bendingForces as the name of the variable here:
cpdef np.ndarray calculate( self, np.ndarray membraneHeight ) :
cdef np.ndarray bendingForces
bendingForces = self.matrixPrefactor * membraneHeight
return bendingForces
and also the name of the member object here:
cdef class membraneClass( object ):
cdef bendingForcesClass bendingForces
Also, bendingForcesClass is the name of the module as well as the class. Finally, how about making a ctypedef from the class bendingForcesClass?
来源：https://stackoverflow.com/questions/5331016/using-self-defined-cython-code-from-other-cython-code