python进阶--性能优化1

Python写法纠正和优化：

 

bad	good
if x == None: ...	if x is None: ...
mapping = {5 : "5", 6 : "6"} for key, val in mapping.items(): ... for key in mapping.keys(): ...	# Use iter* methods when possible mapping = {5: "5", 6: "6"} for key, val in mapping.iteritems(): ... for key in mapping: ...
if x < 10 and x > 2: ...	if 2 < x < 10: ...
al = [1, 2, 3] for i in xrange(len(al)-1, -1, -1): print al[i]	al = [1, 2, 3] for el in reversed(al): print el
if x == 1: y = fun1(x) else if x == 2: y = fun2(x) else if x == 3: y = fun3(x) else: y = None	if x == 1: y = fun1(x) elif x == 2: y = fun2(x) elif x == 3: y = fun3(x) else: y = None # But sometimes a dict is better: funs = {1: fun1, 2: fun2, 3: fun3} y = funs.get(x, lambda x:None)(x)
from operator import add sl = ["ab", "cd", "ef"] all = "" for s in sl: all += s # Or: sl = ["ab", "cd", "ef"] all = reduce(lambda x,y: x+y, sl, "")	sl = ["ab", "cd", "ef"] all = "".join(sl)
a = "this isn't a word, right?" a = a.replace("'", " ") a = a.replace(".", " ") a = a.replace("?", " ") a = a.replace(",", "")	# .replace can be fine. This is faster: from string import maketrans tab = maketrans("'.?", " ") a = "this isn't a word, right." afilt = a.translate(tab, ",")
class Foo(object): def __init__(self, x, y, z): self.x_public = x self.y_private = y self.z_veryprivate = z def getx(self): return self.x_public print Foo(1, 2, 3).getx()	# Generally getters and setters are not used. # Instance names starting with _ are meant as # 'to not mess with' by convention. # Instance names starting with __ are private # and receive name mangling. class Foo(object): def __init__(self, x, y, z): self.x_public = x self._y_private = y self.__z_veryprivate = z print Foo(1, 2, 3).x_public
a = "this isn't a word, right?" a = a.replace("'", " ") a = a.replace(".", " ") a = a.replace("?", " ") a = a.replace(",", "")	# .replace can be fine. This is faster: from string import maketrans tab = maketrans("'.?", " ") a = "this isn't a word, right." afilt = a.translate(tab, ",")
def mul(x, y): return x*y l = [2, 3] print apply(mul, l)	def mul(x, y): return x * y l = [2, 3] print mul(*l)
l = [0] * 4 m = [l] * 4 m[1][1] = 5 print m	# One correct way to create a matrix: m = [[0] * 4 for _ in xrange(4)] m[1][1] = 5 print m  创建4*4矩阵
# sorting on the second item of the tuple # try to remove the i index from the temporary tuples lp = [(5J,"b"),(2J,"c"),(3+1J,"a"),(1+2J,"a")] lp2 = [(c, i, n) for i,(n, c) in enumerate(lp)] lp2.sort() print [(n, c) for (c, i, n) in lp2]	from operator import itemgetter lp = [(5J, "b"), (2J, "c"), (3+1J, "a"), (1+2J, "a")] print sorted(lp, key=itemgetter(1)) 根据元组第二项排序
vals = [5, 7 ,8] tot = -2.0 for v in vals: tot += v	vals = [5, 7 ,8] tot = sum(vals, -2.0)
ll = [[1, 2, 3], [4], [5, 6]] print sum(ll, [])	data = [[1, 2, 3], [4], [5, 6]] result = [] for sublist in data: result.extend(sublist) # Or even, for max speed from itertools import imap data = [[1, 2, 3], [4], [5, 6]] result = [None] * sum(imap(len, data)) pos = 0 for sublist in data: lensl = len(sublist) result[pos : pos+lensl] = sublist pos += lensl
# To negate (inplace) each second # element of alist: result = [] for (i, v) in enumerate(alist): # faster than i % 2 if i & 1 == 0: result.append(v) else: result.append(-v) alist[:] = result	from operator import neg alist[1::2] = map(neg, alist[1::2]) # Or a bit slower but easier to read: alist[1::2] = [-el for el in alist[1::2]] 偶数项变负数
freqs = {} for c in "abracadabra": try: freqs[c] += 1 except: freqs[c] = 1	# Short way: freqs = {} for c in "abracadabra": freqs[c] = freqs.get(c, 0) + 1 # Often the fastest way: freqs = {} for c in "abracadabra": if c in freqs: freqs[c] += 1 else: freqs[c] = 1 # Or better with Python 2.5+: from collections import defaultdict freqs = defaultdict(int) for c in "abracadabra": freqs[c] += 1
someitems = set([1, 2, 3]) somemap = {1:2, 3:4, 5:6} print list(someitems)[0] print list(somemap)[0]	someitems = set([1, 2, 3]) somemap = {1: 2, 3: 4, 5: 6} print iter(someitems).next() print iter(somemap).next()
from time import clock	# This works well on Windows and Linux: from timeit import default_timer as clock # Or often use the timeit module

 

参照：
Python best practices：
http://www.fantascienza.net/leonardo/ar/python_best_practices.html

 

python性能提升：
1. python快速的内建函数
input, int, isinstance, issubclass, iter, open, ord, pow, print, property
 
2. jion连接字符串更快，因为+会创建一个新的字符串并复制旧的内容
同理，list用extend更快
 
3. 快速交换变量，不用temp
x, y = y, x
 
4. 尽量使用局部变量
 
5. 尽量使用in
 
6. 使用延迟加载加速：将import移入函数中，需要时再导入，可以加速软件启动
可以多用from……import……
 
7. while 1：比while True：高效
 
8. list comprehension（列表解析）比for，while快
eg：evens = [i for i in range(10) if i % 2 == 0]
 
9. xrange是range的兄弟，用于需要节省内存和超大数据集合
xrange不生成整个列表
 
10. itertools模块，迭代组合
 
11. python中列表是数组，append是尾部添加，比insert首部添加要高效
deque是双链表
dict，set是哈希表
 
12. Schwartzian Transform准则
装饰——排序——去装饰
例如：使用list.sort比list.sort(cmp)高效（cmp是自己编写的函数），因此对其他结构排序，可以将其他结构装换为list
 
13. GIL全局解释器锁会序列化线程，可以考虑用multiprocessing
 
参照：
《python性能鸡汤》：
中文翻译：http://blog.renren.com/share/101978396/12376358005
英文原文：http://blog.monitis.com/index.php/2012/02/13/python-performance-tips-part-1/
英文原文：http://blog.monitis.com/index.php/2012/03/21/python-performance-tips-part-2/

 

实用工具：
1. python -m profile 程序
性能监控
eg:
python –m profile –o stats x.py
>>> import stats
>>> p = pstats.Stats(‘stats’)   # stats为记录文件
>>> p.sort_stats(‘time’).print_stats(15)
 
详情参照：
python用profile：
http://blog.youkuaiyun.com/lanphaday/article/details/1483728

 

2. 一些有用的第三方库和工具：
NumPy：和Matlab差不多
Scipy：数值处理
GPULib
PyPy：JIT编译器，优化python代码
CPython，Boost等：python -> C
ShedSkin: python -> C++