有时候需要比较大的计算量,这个时候Python的效率就很让人捉急了,此时可以考虑使用numba 进行加速,效果提升明显~
(numba 安装貌似很是繁琐,建议安装Anaconda,里面自带安装好各种常用科学计算库)
1
2
3
4
5
6
7
8
|
from numba import jit
@jit
def t(count = 1000 ):
total = 0
for i in range ( int (count)):
total + = i
return total
|
测试效果:
(关于__wrapped__ 见我的博文: 浅谈解除装饰器作用(python3新增) )
1
2
3
4
5
6
|
In [ 17 ]: % timeit - n 1 t.__wrapped__()
1 loop, best of 3 : 52.9 µs per loop
In [ 18 ]: % timeit - n 1 t()
The slowest run took 13.00 times longer than the fastest. This could mean that an intermediate result is being cached.
1 loop, best of 3 : 395 ns per loop
|
可以看到使用jit 加速后,即使设置测试一次,实际上还是取了三次的最优值,如果取最坏值(因为最优值可能是缓存下来的),则耗时为395ns * 13 大概是5us 还是比不使用的52.9us 快上大概10倍,
增大计算量可以看到使用numba加速后的效果提升更加明显,
1
2
3
4
5
6
|
In [ 19 ]: % timeit - n 10 t.__wrapped__( 1e6 )
10 loops, best of 3 : 76.2 ms per loop
In [ 20 ]: % timeit - n 1 t( 1e6 )
The slowest run took 8.00 times longer than the fastest. This could mean that an intermediate result is being cached.
1 loop, best of 3 : 790 ns per loop
|
如果减少计算量,可以看到当降到明显小值时,使用加速后的效果(以最差计)与不加速效果差距不大,因此如果涉及到较大计算量不妨使用jit 加速下,何况使用起来这么简便。
1
2
3
4
5
6
7
8
9
10
11
12
|
% timeit - n 1 t( 10 )
1 loop, best of 3 : 0 ns per loop
% timeit - n 100 t.__wrapped__( 10 )
100 loops, best of 3 : 1.79 µs per loop
% timeit - n 1 t( 1 )
The slowest run took 17.00 times longer than the fastest. This could mean that an intermediate result is being cached.
1 loop, best of 3 : 395 ns per loop
% timeit - n 100 t.__wrapped__( 1 )
100 loops, best of 3 : 671 ns per loop
|
以上这篇使用numba对Python运算加速的方法就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。
原文链接:https://blog.csdn.net/xiaodongxiexie/article/details/79180158