Question

I have a program in python using numpy and scipy. adding cython in it will be a time taking process as there are many changes in data types

The cprofile of it looks like this: https://i.stack.imgur.com/O2J9O.jpg

Most of the time (73% ) is used up by <scipy.integrate_odepack.odeint> My question basically is, does using cython speed up this function(maybe by faster calling) and the rest. This function is called ~10^6 times in this example. And if so, How much speed up can I expect? I consider pursuing this if speed up is atleast 4x to 5x

注:

If the information provided is insufficient, please comment below and I d be glad to provide

谢谢。

Answer 1

<代码><scipy.integrate_ode Pack.odeint> 正在<代码>dx_dv和dx_du内使用。这些观念像他们重新发挥 functions的作用一样。这些职能,特别是如果复杂的话,可以针对。

One thing to bear in mind is that if the slow down is just because of the calling overhead * number of calls then I wouldn t expect things to improve much. That overhead won t disappear, in fact it might be made more complex. Instead of SciPy C -> Python, you ll be doing SciPy C -> Python -> your C.

Answer 2

Chances are you re not going to see much (if any) improvement. SciPy (and even more specific the code you re most interested in, odeint) is already a CPython module(written in Fortran), not just python.

If this were a python-only function call, then writing it in Cython (which can be easier than writing a full blown CPython module, though personally I d write a c or c++ dll and call it using ctypes...) would increase your performance. But probably not in this situation.

友情链接