Python作为动态语言其功能虽然强大,但是在数值计算方面有一个最大的缺点:速度不够快。在Python级别的循环和计算的速度只有C语言程序的百分之一。因此才有了NumPy, SciPy这样的函数库,将高度优化的C、Fortran的函数库进行包装(封装),以供Python程序调用。如果这些高度优化的函数库无法实现我们的算法,必须从头开始写循环、计算的话,那么用Python来做显然是不合适的。因此SciPy提供了快速调用C++语言程序的方法– Weave。下面是对NumPy的数组求和的例子:
import scipy.weave as weave
import numpy as np
import time
def my_sum(a):
n=len(a)
code="""
int i;
double counter;
counter =0;
for(i=0;i<n;i++){
counter=counter+a(i);
}
return_val=counter;
"""
err=weave.inline(
code,['a','n'],
type_converters=weave.converters.blitz,
compiler="gcc"
)
return err
a = np.arange(0, 10000000, 1.0)
# 先调用一次my_sum,weave会自动对C语言进行编译,此后直接运行编译之后的代码
my_sum(a)
start = time.clock()
for i in xrange(100):
my_sum(a) # 直接运行编译之后的代码
print "my_sum:", (time.clock() - start) / 100.0
start = time.clock()
for i in xrange(100):
np.sum( a ) # numpy中的sum,其实现也是C语言级别
print "np.sum:", (time.clock() - start) / 100.0
start = time.clock()
print sum(a) # Python内部函数sum通过数组a的迭代接口访问其每个元素,因此速度很慢
print "sum:", time.clock() - start
my_sum: 0.0294527349146
np.sum: 0.0527649547638
sum: 9.11022322669
注意:以上是一个老版本的scipy.weave的用法,如果使用最新版本的scipy的,将会提示如下错误:
ImportError: No module named 'scipy.weave'
查阅文档,我们发现:
The scipy.weave module is deprecated. It was the only module never ported to Python 3.x, and is not recommended to be used for new code - use Cython instead. In order to support existing code, scipy.weave has been packaged separately: https://github.com/scipy/weave. It is a pure Python package, and can easily be installed with pip install weave
.
References
[1] 用Weave嵌入C语言