1.优化算法时间复杂度

2. 减少冗余数据

3. 合理使用copy与deepcopy

import copy
a=range(100000)
%timeit-n10copy.copy(a)# 运行10次 copy.copy(a)
%timeit-n10copy.deepcopy(a)
10loops,best of3:1.55ms per loop
10loops,best of3:151ms per loop

4. 使用dict或set查找元素

a=range(1000)
s=set(a)
d=dict((i,1)foriina)
%timeit-n10000100ind
%timeit-n10000100ins
10000loops,best of3:43.5ns per loop
10000loops,best of3:49.6ns per loop

5. 合理使用生成器（generator）和yield

%timeit-n100a=(iforiinrange(100000))
%timeit-n100b=[iforiinrange(100000)]
100loops,best of3:1.54ms per loop
100loops,best of3:4.56ms per loop

%timeit-n10forxin(iforiinrange(100000)):pass
%timeit-n10forxin[iforiinrange(100000)]:pass
10loops,best of3:6.51ms per loop
10loops,best of3:5.54ms per loop

def yield_func(ls):
    foriinls:
        yieldi+1
def not_yield_func(ls):
    return[i+1foriinls]
ls=range(1000000)
%timeit-n10foriinyield_func(ls):pass
%timeit-n10foriinnot_yield_func(ls):pass
10loops,best of3:63.8ms per loop
10loops,best of3:62.9ms per loop

6. 优化循环

a=range(10000)
size_a=len(a)
%timeit-n1000foriina:k=len(a)
%timeit-n1000foriina:k=size_a
1000loops,best of3:569µsper loop
1000loops,best of3:256µsper loop

7. 优化包含多个判断表达式的顺序

a=range(2000)  
%timeit-n100[iforiinaif10<i<20or1000<i<2000]
%timeit-n100[iforiinaif1000<i<2000or100<i<20]     
%timeit-n100[iforiinaifi%2==0andi>1900]
%timeit-n100[iforiinaifi>1900andi%2==0]
100loops,best of3:287µsper loop
100loops,best of3:214µsper loop
100loops,best of3:128µsper loop
100loops,best of3:56.1µsper loop

8. 使用join合并迭代器中的字符串

In[1]:%%timeit
   ...:s=''
   ...:foriina:
   ...:         s+=i
   ...:
10000loops,best of3:59.8µsper loop
In[2]:%%timeit
s=''.join(a)
   ...:
100000loops,best of3:11.8µsper loop

9. 选择合适的格式化字符方式

s1,s2='ax','bx'
%timeit-n100000'abc%s%s'%(s1,s2)
%timeit-n100000'abc{0}{1}'.format(s1,s2)
%timeit-n100000'abc'+s1+s2
100000loops,best of3:183ns per loop
100000loops,best of3:169ns per loop
100000loops,best of3:103ns per loop

10. 不借助中间变量交换两个变量的值

In[3]:%%timeit-n10000
    a,b=1,2
   ....:c=a;a=b;b=c;
   ....:
10000loops,best of3:172ns per loop
In[4]:%%timeit-n10000
a,b=1,2
a,b=b,a
   ....:
10000loops,best of3:86ns per loop

11. 使用if is

a=range(10000)
%timeit-n100[iforiinaifi==True]
%timeit-n100[iforiinaifiisTrue]
100loops,best of3:531µsper loop
100loops,best of3:362µsper loop

12. 使用级联比较x < y < z

x,y,z=1,2,3
%timeit-n1000000ifx<y<z:pass
%timeit-n1000000ifx<yandy<z:pass
1000000loops,best of3:101ns per loop
1000000loops,best of3:121ns per loop

13. while 1 比 while True 更快

def while_1():
    n=100000
    while1:
        n-=1
        ifn<=0:break
def while_true():
    n=100000
    whileTrue:
        n-=1
        ifn<=0:break    
m,n=1000000,1000000
%timeit-n100while_1()
%timeit-n100while_true()
100loops,best of3:3.69ms per loop
100loops,best of3:5.61ms per loop

14. 使用**而不是pow

%timeit-n10000c=pow(2,20)
%timeit-n10000c=2**20
10000loops,best of3:284ns per loop
10000loops,best of3:16.9ns per loop

15. 使用 cProfile, cStringIO 和 cPickle等用c实现相同功能（分别对应profile, StringIO, pickle）的包

import cPickle
import pickle
a=range(10000)
%timeit-n100x=cPickle.dumps(a)
%timeit-n100x=pickle.dumps(a)
100loops,best of3:1.58ms per loop
100loops,best of3:17ms per loop

16. 使用最佳的反序列化方式

import json
import cPickle
a=range(10000)
s1=str(a)
s2=cPickle.dumps(a)
s3=json.dumps(a)
%timeit-n100x=eval(s1)
%timeit-n100x=cPickle.loads(s2)
%timeit-n100x=json.loads(s3)
100loops,best of3:16.8ms per loop
100loops,best of3:2.02ms per loop
100loops,best of3:798µsper loop

17. 使用C扩展(Extension)

18. 并行编程

19. 终级大杀器：PyPy

20. 使用性能分析工具