Python即是面向过程语言,也是面向对象语言,很多其它情况下充当脚本语言的角色。虽是脚本语言,但相同涉及到代码优化的问题,代码优化可以让程序执行更快,它是在不改变程序执行结果的情况下使程序执行效率更高,依据80/20原则。实现程序的重构、优化、扩展以及文档相关的事情通常须要消耗80%的工作量。

优化通常包括双方面的内容:

1.
减小代码的体积、提高代码的可读性及可维护性

2.
改进算法,减少代码复杂度,提高代码执行效率。

选择合适的数据结构一个良好的算法可以对性能起到关键作用。因此性能改进的首要点是对算法的改进。

在算法的时间复杂度排序上依次是:

O(1) > O(lg n) > O(n lg n) > O(n^2) > O(n^3) > O(n^k) > O(k^n) > O(n!)

比方说字典是哈希结构。遍历字典算法复杂度是O(1),而列表算法复杂度是O(n),因此查找对象字典比列表快。

以下列出一些代码优化的技巧。以概要方式总结。因为时间关系。仅仅总结当中一部分。以后会持续更新。

说明

測试的工具: 包含time模块,timeit模块,profile模块或cProfile模块

验证的方式: 包含Python ShelliPythonPython脚本

測试的环境: 包含Python 2.7.6IPython 2.3.1

NOTE:

1.
一般来说c开头是c语言实现,速度更快些,比方cProfile就比profile快。

cPickle比pickle快。

2. 一般来说Python版本号较高。在速度上都有非常大提升。所以測试环境不同,结果不一样。

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+= 比 +快

从Python2.0開始,添加了增强性数据类型,比方说

X += Y

等价于X = X + Y

1. 就优化来说。左側仅仅需计算一次。在X += Y中,X能够使复杂的对象表达式。在增强形式中,则仅仅须要计算一次。

然而,在完整的X = X + Y中,X出现两次,必须运行两次。因此增强赋值语句通常更快些。

from timeit import Timer   #记得导入timeit模块

In [4]: Timer('S = S + "eggs"','S = "SPAM"').timeit()
Out[4]: 2.8523161220051065 In [5]: Timer('S += "eggs"','S = "SPAM"').timeit()
Out[5]: 2.602857082653941

2. 优化技术会自己主动选择,对于支持原处改动的对象而言,增强形式会自己主动运行原处的改动。

普通复制:

>>> M = [1,2,3]
>>> L = M
>>> M = M + [5]
>>> M;L
[1, 2, 3, 4]
[1, 2, 3]

原处改动:

>>> M  = [1,2,3]
>>> L = M
>>> M += [4]
>>> M;L
[1, 2, 3, 4]
[1, 2, 3, 4] >>> Timer('L = L + [4,5,6]','L = [1,2,3]').timeit(20000)
4.324376213615835
>>> Timer('L += [4,5,6]','L = [1,2,3]').timeit(20000)
0.005897484107492801

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

可变对象内置函数比合并操作快

第一种方法: 普通加入来实现

>>> L = [1,2,3]
>>> L = L + [4]
>>> L
[1, 2, 3, 4]

另外一种方法: 内置函数来实现

>>> L = [1,2,3]
>>> L.append(4)
>>> L
[1, 2, 3, 4]

其所花费的时间,相差数百倍:

>>> Timer('L = L + [4]','L = [1,2,3]').timeit(50000)
8.118179033256638
>>> Timer('L.append(4)','L = [1,2,3]').timeit(50000) #内置函数append()方法
0.01078882192950914
>>> Timer('L.extend([4])','L = [1,2,3]').timeit(50000) #内置函数extend()方法
0.020846637858539907

普通的合并操作尽管没有共享引用带来的副作用,与等效的原处改动相比。但速度非常慢。合并操作必须建立新的对象,复制左側的列表,再复制右側的列表。与之相比的是:在原处的改动法仅仅会在内存块的末尾加入元素。

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

布尔測试比边界測试快

>>> Timer('X < Y and Y < Z','X=1;Y=2;Z=3').timeit(100000000)  #布尔測试
7.142944090197389
>>> Timer('X < Y < Z','X=1;Y=2;Z=3').timeit(100000000) #边界測试,推断Y结余X,Z之间
11.501173499654769

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

短路运算比and运算快

在Python中,if用于条件推断,有以下几种情况

X and Y:  X与Y同一时候为真。方为真

X  or Y:  X或Y任一位真,就为真。 也叫短路运算。即假设前面为真,后面则不推断

not X:    X为假时方为真

In [28]: Timer('2 or 3').timeit(100000000)  #短路运算:前面为真,后面不运算,所以速度快些
Out[28]: 3.780060393088206 In [29]: Timer('2 and 3').timeit(100000000) #and。必须运算为全部的,速度相对慢些
Out[29]: 4.313562268420355 In [30]: Timer('0 or 1').timeit(100000000) #or运算,但前面为假。所以和前面速度相当
Out[30]: 4.251177957004984 In [31]: Timer('not 0').timeit(100000000) #not运算,仅仅须要推断一个条件,速度快些
Out[31]: 3.6270803685183637

在前面三个表达式中,短路运算和not运算无疑速度快些,and运算和or中前面条件为假者速度慢些。

所以在程序中适当使用,能够提高程序效率.

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

append比insert速度快

列表的append方法要比insert方法快的多,由于兴许的引用必须被移动以便使新元素腾地方.

复杂度append末尾加入,复杂度O(1)。而insert复杂度是O(n)

>>> Timer('L.append(4)','L=[1,2,3,5,6]').timeit(200000)
0.03233202260122425
>>> Timer('L.insert(3,4)','L=[1,2,3,5,6]').timeit(200000)
18.31223843438289

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

成员变量測试:字典和集合快于列表和元祖

能够用in来做成员变量推断,比方'a' in 'abcd'

推断列表和元祖中是否含有某个值的操作要比字典和集合慢的多。

由于Python会对列表中的值进行线性扫描。而另外两个基于哈希表,能够瞬间完毕推断。

数据越大,越明显。

In [44]: Timer('4 in L','L=(1,2,3,4,5,6,7,8,9)').timeit(100000000)
Out[44]: 12.941504527043435 #列表成员推断 In [45]: Timer('4 in T','T=[1,2,3,4,5,6,7,8,9]').timeit(100000000)
Out[45]: 12.883945908790338 #元祖成员推断,和列表几乎相同 In [46]: Timer('4 in S','S=set([1,2,3,4,5,6,7,8,9])').timeit(100000000)
Out[46]: 6.254324848690885 #集合成员推断。和字典几乎相同 In [47]: Timer('4 in D','D={1:"a",2:"b",3:"c",4:"d",5:"e",6:"f",7:"g",8:"h",9:"i"}').timeit(100000000)
Out[47]: 6.3508488422085065 #字典成员推断

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

列表合并extend比+速度快

列表合并(+)是一种相当费资源的操作,由于必须创建一个新列表并将全部对象复制进去。

而extend将元素附加到现有列表中,因此会快非常多,尤其是创建一个大列表时尤其如此.

+操作运行结果:

import profile            #用cProfile会快些

def func_add():           #測试列表合并操作
lst = []
for i in range(5000):
for item in [[0],[1],[2],[3],[4],[5],[6],[7],[8],[9],[10]]:
lst = lst + item if __name__=='__main__':
profile.run('func_add()')
#####測试结果:#####
>>>
5 function calls in 9.243 seconds Ordered by: standard name ncalls tottime percall cumtime percall filename:lineno(function)
1 0.000 0.000 0.000 0.000 :0(range)
1 0.006 0.006 0.006 0.006 :0(setprofile)
1 0.000 0.000 9.237 9.237 <string>:1(<module>)
1 9.236 9.236 9.236 9.236 Learn.py:3(func_add)
1 0.000 0.000 9.243 9.243 profile:0(func_add())
0 0.000 0.000 profile:0(profiler)

extend运行结果:

import profile

def func_extend():
lst = []
for i in range(5000):
for item in [[0],[1],[2],[3],[4],[5],[6],[7],[8],[9],[10]]:
lst.extend(item) if __name__=='__main__':
profile.run('func_extend()') #####输出结果:#####
>>>
55005 function calls in 0.279 seconds Ordered by: standard name ncalls tottime percall cumtime percall filename:lineno(function)
55000 0.124 0.000 0.124 0.000 :0(extend)
1 0.000 0.000 0.000 0.000 :0(range)
1 0.005 0.005 0.005 0.005 :0(setprofile)
1 0.000 0.000 0.274 0.274 <string>:1(<module>)
1 0.149 0.149 0.273 0.273 Learn.py:3(func_extend)
1 0.000 0.000 0.279 0.279 profile:0(func_extend())
0 0.000 0.000 profile:0(profiler)

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

xrange比range快

In [9]: Timer('for i in range(1000): pass').timeit()
Out[9]: 30.839959527228757 In [10]: Timer('for i in xrange(1000): pass').timeit()
Out[10]: 19.644791055468943

xrange是range的C语言实现。更高效的内存管理。

xrange:每次仅仅迭代一个对象

range:一次生成全部数据,须要一个个扫描

NOTE: 在Python3.0中取消了xrange函数,仅仅留range。无论这个range事实上就是xrange,仅仅只是名字变了。

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

内置函数>列表推导>for循环>while循环

http://blog.csdn.net/jerry_1126/article/details/41773277

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

局部变量>全局变量

import profile

A = 5

def param_test():
B = 5
res = 0
for i in range(100000000):
res = B + i
return res if __name__=='__main__':
profile.run('param_test()')
>>> ===================================== RESTART =====================================
>>> 
         5 function calls in 37.012 seconds #全局变量測试结果:37 s    Ordered by: standard name    ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1   19.586   19.586   19.586   19.586 :0(range)
        1    1.358    1.358    1.358    1.358 :0(setprofile)
        1    0.004    0.004   35.448   35.448 <string>:1(<module>)
        1   15.857   15.857   35.443   35.443 Learn.py:5(param_test)
        1    0.206    0.206   37.012   37.012 profile:0(param_test())
        0    0.000             0.000          profile:0(profiler) >>> ===================================== RESTART =====================================
>>> 
         5 function calls in 11.504 seconds #局部变量測试结果: 11s    Ordered by: standard name    ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    3.135    3.135    3.135    3.135 :0(range)
        1    0.006    0.006    0.006    0.006 :0(setprofile)
        1    0.000    0.000   11.497   11.497 <string>:1(<module>)
        1    8.362    8.362   11.497   11.497 Learn.py:5(param_test)
        1    0.000    0.000   11.504   11.504 profile:0(param_test())
        0    0.000             0.000          profile:0(profiler)

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

while 1 > while True

while 1运行结果:

import cProfile

def while_1():
tag = 0
while 1:
tag += 1
if tag > 100000000:
break if __name__=='__main__':
cProfile.run('while_1()') >>> ===================================== RESTART =====================================
>>>
4 function calls in 5.366 seconds Ordered by: standard name ncalls tottime percall cumtime percall filename:lineno(function)
1 0.006 0.006 0.006 0.006 :0(setprofile)
1 0.000 0.000 5.360 5.360 <string>:1(<module>)
1 5.360 5.360 5.360 5.360 Learn.py:3(while_1)
0 0.000 0.000 profile:0(profiler)
1 0.000 0.000 5.366 5.366 profile:0(while_1())

while True运行结果:

import cProfile

def while_true():
tag = 0
while True:
tag += 1
if tag > 100000000:
break if __name__=='__main__':
cProfile.run('while_true()') >>> ===================================== RESTART =====================================
>>>
4 function calls in 8.236 seconds Ordered by: standard name ncalls tottime percall cumtime percall filename:lineno(function)
1 0.012 0.012 0.012 0.012 :0(setprofile)
1 0.000 0.000 8.224 8.224 <string>:1(<module>)
1 8.224 8.224 8.224 8.224 Learn.py:10(while_true)
0 0.000 0.000 profile:0(profiler)
1 0.000 0.000 8.236 8.236 profile:0(while_true())

NOTE: while 1比while True,运行快些,是由于Python 2.x中True相当于全局变量,非keyword.

尽管前者比后者快些,但后者可读性无疑更好些.

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

求交集集合比列表快

列表測试结果:

from time import time

t1 = time()
list_1 = [32,78,65,99,19,43,18,22,7,1,9,2,4,8,56]
list_2 = [3,4,8,56,99,100]
temp = []
for x in range(1000000):
for i in list_2:
for j in list_1:
if i == j:
temp.append(i)
t2 = time()
print "Total time:", t2 - t1 #測试结果:
>>>
Total time: 13.6879999638

集合測试结果:

from time import time

t1 = time()
set_1 = set([32,78,65,99,19,43,18,22,7,1,9,2,4,8,56])
set_2 = set([3,4,8,56,99,100])
for x in range(1000000):
set_same = set_1 & set_2 t2 = time()
print "Total time:", t2 - t1 #測试结果:
>>>
Total time: 0.611000061035

NOTE:
用集合的方式取交集速度快的多。

以下是经常使用的集合操作。

>>> set1 = set([2,3,4,8,9])  #集合1
>>> set2 = set([1,3,4,5,6]) #集合2
>>> set1 & set2 #求交集
set([3, 4])
>>> set1 | set2 #求合集
set([1, 2, 3, 4, 5, 6, 8, 9])
>>> set1 - set2 #求差集
set([8, 9, 2])
>>> set1 ^ set2 #求异或:即排除共同部分
set([1, 2, 5, 6, 8, 9])

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

直接交换两变量 > 借助中间变量

要交换X,Y的值。有两种方法:

1. 直接交换: X, Y = Y, X

>>> X,Y = 1,2
>>> X,Y
(1, 2)
>>> X, Y = Y, X
>>> X,Y
(2, 1)

2.借助中间变量: T = X, X = Y, Y = X

>>> X,Y = 1,2
>>> X,Y
(1, 2)
>>> T = X; X = Y; Y = T
>>> X,Y
(2, 1)

測试结果:



++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

is not速度快于!=

在if条件推断中。能够用 if a is not None:或者 if a != None 前者执行速度快于后者.

測试结果:

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

''.join(list)速度快于+或+=

+測试结果:



''.join(list)測试结果:

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

在循环体外运行函数比在循环中快

所以要降低函数的调用次数

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

**比pow()速度快

測试结果:



**其速度是pow()函数的几十倍,并且数据越大,越明显。

**相当于Python的移位操作: 右移(>>) 和 左移(<<)比方说 2**2 = 4相当于 2 << 1 其速度相当!

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

生成器速度比列表快

前者是列表解析对象。后者是generate对象。所需的内存空间与列表大小无关。所以速度快些.

在实际应用中,比方说创建一个集合,用生成器对象明显比列表对象要快些。

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

浅拷贝的速度比深拷贝速度

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

cPickle比pickle速度快

pickle模块和cPickle模块都能够将Python对象永久存储在系统文件里。

但cPickle是Python的C语言实现。因而速度更快些。

请看以下对照:

存储一个百万级别大小的列表,用cPickle模块差点儿10倍于pickle模块。

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

int()比int(math.floor())快

比方对浮点数取整,32.9,假设要取年龄的话。仅仅能是32。

能够有两种方式

一种: int(math.floor(32.9))   # floor先取32.0再转为整数

一种: int(32.9)                  # 直接向下取整,math.floor()下多余的

能够看出,另外一种方式要比第一种快的多。

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

读取文件操作for循环比while高效

測试项目: test.txt 3M的文件

測试结果: 245.83s

測试结果: 182.62s

for line in open('filename'):

process(line)

上面这样的方式应该读取文件的最佳方式:原因有三

1.写法最简单

2. 执行最高速

3. 从内存使用角度来看,也是最好的

NOTE: 

1.
readlines()是一次性载入全部的行,而xreadlines()是按须要载入。可避免大文件导致内存泄露

2.
readline()是迭代逐行读取。从内存角度来说。在大文件处理中。效率要比readlines()高

3. 超大文件的话,用readlines()方式能够会导致内存崩溃。

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

字典的迭代取值比直接取值高效

>>> Timer('for v in d.values(): pass',"d={'a':1,'b':2,'c':3,'d':4,'e':5}").timeit(100000000)
43.297132594847085
>>> Timer('for v in d.itervalues(): pass',"d={'a':1,'b':2,'c':3,'d':4,'e':5}").timeit(100000000)
36.16957129734047

也就是说:

# d.itervalues() 比d.values()要快些
# d.iteritems() 比d.items()要快些
# d.iterkeys() 比d.keys()要快些

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

创建字典常规方法比工厂方法要快

常规方法:

d={'x':1, 'y':2, 'z':3}

工厂方法:例如以下面几种

1. dict((['x',1], ['y',2], ['z',3]))

2. dict(zip(('x', 'y', 'z'), (1, 2, 3)))

3. dict(x=1, y=2, z=3)

>>> Timer("d={'x':1, 'y':2, 'z':3}").timeit()
0.19084989859678103
>>> Timer("dict((['x',1], ['y',2], ['z',3]))")
<timeit.Timer instance at 0x0000000002E06288>
>>> Timer("dict((['x',1], ['y',2], ['z',3]))").timeit()
1.5503493690009975

相同工厂方法创建新字典速度比copy()函数慢

>>> Timer("D2=copy.copy({'x':1, 'y':2, 'z':3})","import copy").timeit()
1.074277245445046
>>> Timer("D3=dict((['x',1], ['y',2], ['z',3]))").timeit()
1.5830155758635556

最新文章

  1. Java里List取并集方法retainAll不能用来判断是否有重复数据!
  2. 爬虫之scrapy框架
  3. BZOJ3276 : 磁力
  4. C# testJsonAsXMLNodeAttribute - XML&amp; json &amp; Collections - XmlNode, XmlElement, XmlAttribute,Dictionary,List
  5. SortedList的用法
  6. SVNclient安装与使用
  7. SRM 624 Building Heights DivI 解读
  8. linux 安装nginx 详解
  9. Spring Security 入门(1-3-1)Spring Security - http元素 - 默认登录和登录定制
  10. centos7网络配置方法
  11. IP网际协议
  12. cmake教程
  13. 安卓BLE测试apk
  14. 解决关于ios访问相机闪退问题
  15. 使用 Topshelf 结合 Quartz.NET 创建 Windows 服务
  16. MyChatRoom——C#自制聊天室
  17. boost::asio 学习草稿
  18. nodejs 发起http请求
  19. k8s常用指令集(kubectl kubeadm)
  20. LINUX 命令&mdash;netstat [简单实用]

热门文章

  1. Mac 因误使用chmod -R 777 命令更改 /usr/bin 造成终端不能实用,提醒进程已结束的完美解决方案!
  2. mysql环境变量配置(复制)
  3. poj3083 Children of the Candy Corn BFS&amp;&amp;DFS
  4. LAMP第二部分apache的配置
  5. 【bzoj2427】[HAOI2010]软件安装 Tarjan+树形背包dp
  6. 个人环境搭建——ANT安装
  7. jquery 获取被点击元素的id属性值
  8. 【HDOJ5556】Land of Farms(最大团)
  9. 自动从网站上面下载文件 .NET把网站图片保存到本地
  10. 禁止复制(copy),禁用鼠标右键!